African Successes, Volume III: Modernization and Development 9780226315867

Citation preview

African Successes, Volume III

National Bureau of Economic Research Conference Report

African Successes, Volume III Modernization and Development

Edited by

Sebastian Edwards, Simon Johnson, and David N. Weil

The University of Chicago Press Chicago and London

Sebastian Edwards is the Henry Ford II Professor of International Economics in the Anderson Graduate School of Management at the University of California, Los Angeles. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and professor of global economics and management at the MIT Sloan School of Management. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University. All three editors are research associates of the NBER.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2016 by the National Bureau of Economic Research Printed in the United States of America 25 24 23 22 21 20 19 18 17 16 1 2 3 4 5 ISBN-13: 978-0-226-31572-0 (cloth) ISBN-13: 978-0-226-31586-7 (e-book) DOI: 10.7208/chicago/9780226315867.001.0001 Library of Congress Cataloging-in-Publication Data Names: Edwards, Sebastian, 1953– editor. | Johnson, Simon, 1963– editor. | Weil, David N., editor. Title: African successes : government and institutions / edited by Sebastian Edwards, Simon Johnson, and David N. Weil. Other titles: National Bureau of Economic Research conference report. Description: Chicago ; London : Chicago and London, 2016– | Series: National Bureau of Economic Research conference report Identifiers: LCCN 2015050080 | ISBN 9780226316222 (cloth : alk. paper) | ISBN 9780226316369 (e-book) | ISBN 9780226316055 (cloth : alk. paper) | ISBN 9780226316192 (e-book) | ISBN 9780226315720 (cloth : alk. paper) | ISBN 9780226315867 (e-book) | ISBN 9780226315553 (cloth : alk. paper) | ISBN 9780226315690 (e-book) Subjects: LCSH: Economic development—Africa. | Africa—Economic conditions—21st century. Classification: LCC HC800 .A56873 2016 | DDC 330.96–dc23 LC record available at http://lccn.loc.gov/2015050080 ♾ This paper meets the requirements of ANSI/NISO Z39.48–1992 (Permanence of Paper).

National Bureau of Economic Research Officers Martin B. Zimmerman, chairman Karen N. Horn, vice chairman James M. Poterba, president and chief executive officer Robert Mednick, treasurer

Kelly Horak, controller and assistant corporate secretary Alterra Milone, corporate secretary Denis Healy, assistant corporate secretary

Directors at Large Peter C. Aldrich Elizabeth E. Bailey John H. Biggs John S. Clarkeson Don R. Conlan Kathleen B. Cooper Charles H. Dallara George C. Eads Jessica P. Einhorn

Mohamed El-Erian Linda Ewing Jacob A. Frenkel Judith M. Gueron Robert S. Hamada Peter Blair Henry Karen N. Horn John Lipsky Laurence H. Meyer

Michael H. Moskow Alicia H. Munnell Robert T. Parry James M. Poterba John S. Reed Marina v. N. Whitman Martin B. Zimmerman

Directors by University Appointment Timothy Bresnahan, Stanford Pierre-André Chiappori, Columbia Alan V. Deardorff, Michigan Ray C. Fair, Yale Edward Foster, Minnesota John P. Gould, Chicago Mark Grinblatt, California, Los Angeles Bruce Hansen, Wisconsin–Madison

Benjamin Hermalin, California, Berkeley Marjorie B. McElroy, Duke Joel Mokyr, Northwestern Andrew Postlewaite, Pennsylvania Cecilia Rouse, Princeton Richard L. Schmalensee, Massachusetts Institute of Technology David B. Yoffie, Harvard

Directors by Appointment of Other Organizations Jean-Paul Chavas, Agricultural and Applied Economics Association Martin J. Gruber, American Finance Association Arthur Kennickell, American Statistical Association Jack Kleinhenz, National Association for Business Economics William W. Lewis, Committee for Economic Development Robert Mednick, American Institute of Certified Public Accountants

Alan L. Olmstead, Economic History Association Peter L. Rousseau, American Economic Association Gregor W. Smith, Canadian Economics Association William Spriggs, American Federation of Labor and Congress of Industrial Organizations Bart van Ark, The Conference Board

Directors Emeriti George Akerlof Jagdish Bhagwati Carl F. Christ Franklin Fisher

George Hatsopoulos Saul H. Hymans Rudolph A. Oswald Peter G. Peterson

John J. Siegfried Craig Swan

Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the NBER is to ascertain and present to the economics profession, and to the public more generally, important economic facts and their interpretation in a scientific manner without policy recommendations. The Board of Directors is charged with the responsibility of ensuring that the work of the NBER is carried on in strict conformity with this object. 2. The President shall establish an internal review process to ensure that book manuscripts proposed for publication DO NOT contain policy recommendations. This shall apply both to the proceedings of conferences and to manuscripts by a single author or by one or more co-authors but shall not apply to authors of comments at NBER conferences who are not NBER affiliates. 3. No book manuscript reporting research shall be published by the NBER until the President has sent to each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the above principles of the NBER. Such notification will include a table of contents and an abstract or summary of the manuscript’s content, a list of contributors if applicable, and a response form for use by Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied and the main conclusions reached. 4. No volume shall be published until forty-five days have elapsed from the above notification of intention to publish it. During this period a copy shall be sent to any Director requesting it, and if any Director objects to publication on the grounds that the manuscript contains policy recommendations, the objection will be presented to the author(s) or editor(s). In case of dispute, all members of the Board shall be notified, and the President shall appoint an ad hoc committee of the Board to decide the matter; thirty days additional shall be granted for this purpose. 5. The President shall present annually to the Board a report describing the internal manuscript review process, any objections made by Directors before publication or by anyone after publication, any disputes about such matters, and how they were handled. 6. Publications of the NBER issued for informational purposes concerning the work of the Bureau, or issued to inform the public of the activities at the Bureau, including but not limited to the NBER Digest and Reporter, shall be consistent with the object stated in paragraph 1. They shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The Executive Committee of the Board is charged with the review of all such publications from time to time. 7. NBER working papers and manuscripts distributed on the Bureau’s web site are not deemed to be publications for the purpose of this resolution, but they shall be consistent with the object stated in paragraph 1. Working papers shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The NBER’s web site shall contain a similar disclaimer. The President shall establish an internal review process to ensure that the working papers and the web site do not contain policy recommendations, and shall report annually to the Board on this process and any concerns raised in connection with it. 8. Unless otherwise determined by the Board or exempted by the terms of paragraphs 6 and 7, a copy of this resolution shall be printed in each NBER publication as described in paragraph 2 above.

Contents

Series Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

1

Volume Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

5

I. Finance and Capital Allocation 1. Resolving the African Financial Development Gap: Cross-Country Comparisons and a Within-Country Study of Kenya Franklin Allen, Elena Carletti, Robert Cull, Jun Qian, Lemma Senbet, and Patricio Valenzuela 2. Challenges in Banking the Rural Poor: Evidence from Kenya’s Western Province Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson 3. The Financial Sector in Burundi: An Investigation of Its Efficiency in Resource Mobilization and Allocation Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

13

63

103

vii

viii

Contents

4. Were the Nigerian Banking Reforms of 2005 a Success . . . and for the Poor? Lisa D. Cook 5. Misallocation, Property Rights, and Access to Finance: Evidence from within and across Africa Sebnem Kalemli-Ozcan and Bent E. Sørensen

157

183

II. Mobile Phones 6. New Cellular Networks in Malawi: Correlates of Service Rollout and Network Performance Dimitris Batzilis, Taryn Dinkelman, Emily Oster, Rebecca Thornton, and Deric Zanera 7. Mobile Banking: The Impact of M-Pesa in Kenya Isaac Mbiti and David N. Weil

215

247

III. International Trade 8. African Export Successes: Surprises, Stylized Facts, and Explanations William Easterly and Ariell Reshef

297

9. AGOA Rules: The Intended and Unintended Consequences of Special Fabric Provisions Lawrence Edwards and Robert Z. Lawrence

343

Contributors Author Index Subject Index

395 399 403

Series Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

In the 1950s and early 1960s, there was a great deal of optimism about the prospects for economic development in sub-Saharan Africa. By early in the twenty-first century, the prevailing consensus had become much more negative—and for good reason. Decades of civil war, repeated state failure, corruption, and disappointing private sector performance characterized much of Africa’s postindependence experience. A wave of prominent papers in the economic literature tried to dig more deeply into the causes of these problems, with some scholars putting them in a broader comparative perspective and others focusing on the specifics of the African experience. There was no shortage of deep causes suggested as explanations for repeated African disappointments, including weak rule of law, a lack of democracy, colonial inheritance, the impact of the slave trade, the burden of tropical disease, some form of “resource curse,” and ethno-linguistic divisions among the population. The NBER Africa project, conceived in the middle of the first decade of the twenty-first century, took a different approach. Rather than trying to understand the causes of underperformance in Africa, we decided to focus on finding and understanding more positive aspects of what was happening Sebastian Edwards is the Henry Ford II Professor of International Economics at the University of California, Los Angeles, and a research associate of the National Bureau of Economic Research. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and Professor of Global Economics and Management at the MIT Sloan School of Management and a research associate of the National Bureau of Economic Research. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University and a research associate of the National Bureau of Economic Research. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13651.ack.

1

2

Sebastian Edwards, Simon Johnson, and David N. Weil

on that continent south of the Sahara, along several different dimensions studied by economists. Our timing proved good for three reasons. First, scholars were turning their attention increasingly to the details of what was working well or at least better with regard to African development. This new wave of research involved working with or, in many cases creating, data sets suited to careful analysis—including sophisticated use of econometrics. We are fortunate to have involved and supported many leading empirical researchers as they broke new ground in multiple directions. Most likely some of this work would have happened in any case, but at the very least we can claim to have provided an appropriate catalyst for many projects. Second, Africa is changing—and the prospects for nearly a billion people are looking up. A decade or more of relatively good performance does not a miracle make, but across a wide range of countries there has been better economic growth, considerable progress with improving public health and other social indicators, and a range of attempts to improve the performance of the state. Many of the chapters in this series of four volumes address some aspect of this profound and important transformation. Third, the study of Africa is becoming much more integrated with the rest of economics. Just as happened earlier with research on Japan and China, a line of work that previously might have been seen as of purely regional interest can now appear in top journals. The broader trend, of course, has been the resurrection of development economics as a vibrant field. But also encouraging—and a central part of the NBER Africa project—has been the increase in interest among nonspecialists regarding what is actually happening on this dynamic and fascinating continent. Volume I in this series focuses on the most basic building blocks of economic development, including the rule of law, civil conflict, and institutions more broadly. Volume II contains chapters on human capital development in Africa, including some important work on public health improvements, but also research into education on the continent. Volume III examines whether Africa can glean any advantage from being relatively late to economic development, including being able to bypass some large investments (for example, in fixed-line telephones or in traditional branch banking) or being able to learn from others (for example, in the development of an export sector). Volume IV looks at whether recent growth can be sustained, in terms of agricultural development and more broadly. The research presented in these volumes covers a wide territory, in geographic and intellectual terms. However, our project was never intended to be comprehensive. Rather we attempted to act as a catalyst for rigorous and innovative thinking applied to recent African development. The work published here far exceeds our earliest expectations—a reflection, primarily, of how many serious scholars are now deeply engaged with these issues.

Series Introduction

3

It has been a great honor for us, as program directors and now as editors of these volumes, to participate in and help to facilitate this surge in serious interest. In terms of making this possible, we must thank, first and foremost, the Bill & Melinda Gates Foundation for their willingness to encourage economists in this endeavor. We have worked with a range of officials at the Foundation over the past decade; their support has been unstinting and their perspective on development is always refreshing. We would specifically like to thank Sara Sievers for many of our early interactions and Oliver Babson for his subsequent engagement. We also greatly appreciate the ideas and energy of Geoffrey Lamb, Kim Hamilton, Negar Akhavi, Adam Gerstenmier, and Mumukshu Patel. We must also recognize the founding insight and energy of Martin Feldstein, who emphasized, among many other relevant and helpful points, that there is an important link between raising the quality of economic research on an issue and improving the odds that policy discussions will become more informed. Since becoming president of the NBER in 2008, James Poterba has also provided us with great leadership and encouragement—including sage advice at every stage of the process. Our four published volumes present results from thirty-nine research teams, with the findings organized along thematic lines. Of the 100 authors whose names appear on one or more of the chapters in our collection, nineteen are from Africa and thirty-two are affiliated with the NBER as either Faculty Research Fellows or research associates. They range from experienced specialists on African development to scholars who had never previously worked on the continent; in seniority they ran the gamut from newly appointed assistant professors to distinguished, established thought leaders. Every project included a significant on-the-ground component. Some research teams combined money we provided with other funds, for example, to support the collection of very large field data sets. In other cases, funding from the NBER made it possible for researchers to interview policymakers, business people (both large and small), farmers, health workers, or others engaged in the process of economic development. Results from the research project were presented at a series of conferences in the United States and in Africa. Given the large size of our project and our explicit goal of including scholars who had not previously been working on Africa, we began with a background conference in February 2008, at the NBER in Cambridge, MA, bringing together researchers and policymakers from the United States, Africa, and Europe. The conference featured presentations on current research and an overview of available data, aimed especially at scholars who were new to the area. Our next meeting, also in Cambridge, MA, in February 2009 was a preconference at which preliminary findings from ten papers were presented. This was followed by a conference in December of the same year, again in

4

Sebastian Edwards, Simon Johnson, and David N. Weil

Cambridge, MA, that took a hybrid form: final versions of five projects were presented, as were preconference versions of another twelve projects. This was followed in July of 2010 by a conference in Accra, Ghana, held in cooperation with the Institute of Statistical, Social, and Economic Research (ISSER). Our final research conference was held in Zanzibar, Tanzania, in August 2011, in cooperation with the Bank of Tanzania. One goal of our project from inception, with full support from the Gates Foundation, was to help connect economic research with the African policy community. We arranged some interactions along these lines throughout our project, but we were fortunate to be able to make a special effort at the end, with a meeting devoted specifically to the Next Macroeconomic Challenges in Africa, held again in Zanzibar, in December 2012, also in cooperation with the Bank of Tanzania. There are many people to thank for the successful completion of this project. Benno Ndulu, governor of the Bank of Tanzania, provided wise guidance throughout the process and particularly helped us to focus on narrowing the gap between research and policy, including our two fruitful meetings in Zanzibar. His team at the Bank of Tanzania was most helpful in many ways, and we have special thanks for Patricia Mlozi, Mechtilda Mugo, Msafiri Nampesya, and Pamella Lowassa-Solomon. Our conference in Accra benefited greatly from the engagement and support provided by Ernest Aryeetey, Kwesi Botchwey, and Robert Osei. And we had great cooperation in many ways across the entire project with African Economic Research Consortium (AERC); thanks to Olu Ajakaiye, William Lyakurwa, and Lemma Senbet for making this possible. We would like to thank everyone who attended our various conferences and who worked hard on all dimensions of these research projects. We are also most grateful for all the inputs received from members of the Advisory Committee, formed especially for this project: Robert Bates, Paul Collier, Martin Feldstein, Benno Ndulu, Franklyn Prendergast, Antoinette Sayeh, Nicholas Stern, and John Taylor. The entire project ran smoothly due to the hard work, precision, and attention to detail of Elisa Pepe at the NBER. Elisa was with us from beginning to end, and words cannot sufficiently express our gratitude for all she has done. Others at the NBER provided outstanding help on a wide variety of tasks. Our meetings in the United States and in Africa were organized with exemplary competence by Carl Beck of the NBER’s conference department. For all their help in the production, management, and dissemination of research, we would also like to thank Alex Aminoff, Laura Bethard, Daniel Feenberg, Helena Fitz-Patrick, Wayne Gray, Steve Harriman, and Alterra Milone. And for their assistance with the data portal, we are grateful to Binh Thanh Le, Dimitry Legagneur, and C. Michelle Tejada.

Volume Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

Volumes I and II in this series deal with some difficult fundamental issues, including whether African countries can put their civil wars behind them and now look forward to at least a reasonable life expectancy. Beset by both traditional tropical disease and the awful impact of HIV/AIDS, it is an uphill struggle. However, progress is being made—and this is a major theme of our research project—but a great deal of hard work remains to be done. Volume III turns to quite different dimensions of African development, including some aspects where it is possible to imagine there may even be a “late mover” advantage: finance, mobile phones, and exports. Finance and Capital Allocation For the financial sector there is much that can be gleaned from prior experience. At the very least, Africa should potentially be able to avoid some of the more serious problems seen elsewhere, and it is not unreasonable to imagine that some African countries could actually find institutional or technological innovations that will allow more effective, inclusive, and safer use of finance, either for firms or individuals or both.

Sebastian Edwards is the Henry Ford II Professor of International Economics at the University of California, Los Angeles, and a research associate of the National Bureau of Economic Research. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and Professor of Global Economics and Management at the MIT Sloan School of Management and a research associate of the National Bureau of Economic Research. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University and a research associate of the National Bureau of Economic Research. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13357.ack

5

6

Sebastian Edwards, Simon Johnson, and David N. Weil

On finance, a major issue remains how best to bring people into the organized financial sector—with bank accounts, payments- related services, and access to credit. In chapter 1, “Resolving the African Financial Development Gap: Cross-Country Comparisons and a Within-Country Study of Kenya,” Franklin Allen, Elena Carletti, Robert Cull, Jun Qian, Lemma Senbet, and Patricio Valenzuela emphasize the particular difficulties that arise from relatively low population density in the African context. They use comparative data to put this issue in appropriate context, but they also look for particular business models that may prevail despite this feature of the market environment. In contrast to most of the other work represented in these volumes, the authors focus on a single company that appears to offer a potential role model for others. It remains to be seen how well this model will travel across countries, but there are potentially important lessons for private financialsector firms across the continent and perhaps more broadly. In chapter 2, Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson marshal a great deal of evidence that suggests how hard it will be to ensure relatively poor rural residents receive appropriate services from the financial sector. “Challenges in Banking the Rural Poor: Evidence from Kenya’s Western Province” reviews relevant survey evidence and also reports on some insightful experiments run by this team of researchers. The problem is not so much supply- side constraints—that is, this is not about the banks being reluctant to lend. Rather the issue is much more about potential customers not actually wanting either the financial services already offered or what could reasonably become available in the near term. The authors’ investigations have uncovered a great deal of convincing detail regarding exactly what makes people reluctant to use banking services today—as well as some indications of what it would take to significantly change this situation. Simply reducing bank fees is unlikely to be enough. The authors also do a great service by outlining what should be the research agenda for this topic. We gain some rare holistic insight into financial- sector functioning with chapter 3, “The Financial Sector in Burundi: An Investigation of Its Efficiency in Resource Mobilization and Allocation” by Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya. The authors have analyzed a considerable amount of data regarding who has access to credit, including how this is related to political connections. Is it profitable to lend to the rising middle class? What is the relative role of traditional banking compared with microfinance? Does the arrival of foreign banks help or hurt access to credit? And how should we evaluate regulation and supervision in this context? The authors address all of these questions. Chapter 4 assesses a prominent episode from Nigeria that confirms that expanding access to finance is much harder than making the financial system more stable. In “Were the Nigerian Banking Reforms of 2005 a Success . . .

Volume Introduction

7

and for the Poor?” Lisa D. Cook argues that financial- sector safety, soundness, and accessibility have improved considerably over the past decade. Banks becoming better capitalized and more efficient is no small achievement. However, she also finds no evidence that financial services became more accessible to the poor. In addition, part of the improved performance may simply be due to a better macroeconomic environment. The reformed Nigerian banking system has not yet been fully stress tested by events. We probably need to see a full boom- bust cycle before we can really assess the resilience of the banks and their balance sheets. Chapter 5 links financial development to nonfinancial- sector outcomes, while also suggesting some deeper causes of weak capital allocation mechanisms. In “Misallocation, Property Rights, and Access to Finance: Evidence from within and across Africa,” Sebnem Kalemli-Ozcan and Bent E. Sørensen look at the marginal product of capital in ten African countries. They find evidence that the allocation of capital across opportunities is far from optimal, although there are important differences in this regard across types of countries—and also across specific kinds of firms. The authors also shed light on whether the issues here are more about country- level institutions, such as the strength of property rights and the quality of the legal system, or more specific financial- sector rules and regulations. This analysis should be informative to anyone seeking policy levers that can help capital flow to activities with high private or social returns. This topic deserves to receive a significant amount of attention from the official sector, but also from anyone who would like the African private sector to become more productive. Mobile Phones Mobile phones are obviously an important new sector, for Africa as for the rest of the world. It is also possible that because the legacy fixed- line systems are weaker, mobile phone networks will provide opportunities in Africa—and perhaps even have an impact on development—not seen elsewhere. Two projects in our portfolio explored this issue in more detail. Chapter 6, “New Cellular Networks in Malawi: Correlates of Service Rollout and Network Performance,” by Dimitris Batzilis, Taryn Dinkelman, Emily Oster, Rebecca Thornton, and Deric Zanera, considers who gets high- quality access to the mobile phone network. Infrastructure for mobile networks has expanded dramatically over the past two decades, providing access even in remote areas within developing countries that never previously had phone service. The authors examine the rollout of coverage in Malawi. Overall, despite the low average income level in that country, within ten years over half of the country obtained access to at least one network. Still, some initial advantages tend to help inhabitants get better access to

8

Sebastian Edwards, Simon Johnson, and David N. Weil

the mobile network. Higher population density and more education play a role. Areas that are far from a road are significantly less likely to get access to the network throughout the period. Further research needs to examine what kind of advantage is conferred by mobile phone access—and where the creation of these new networks actually reinforces existing inequality and barriers to development. Africa is not a global player in terms of developing the hardware and basic software for mobile telephony. But some African companies have led the way in terms of thinking both about the traditional voice communications side of this industry (for example, selling air time in very small increments), as well as about how to bring mobile phones and banking closer together. In chapter 7, “Mobile Banking: The Impact of M-Pesa in Kenya,” by Isaac Mbiti and David N. Weil, the authors examine a mobile phone- based money transfer system in Kenya. M-Pesa grew rapidly following its inception in 2007 and has had a great deal of impact—including on how financial services are used more broadly and on the cost of competing money transfer services. It does not appear to be the case that people use their M-Pesa accounts as a place to store wealth. However, there are strong indications that M-Pesa improves individual outcomes by promoting the use of banking services and by making it cheaper and easier to transfer money— particularly from urban to rural areas. Hopefully, researchers will be able to build on these results—both with regard to improved availability of financial services and more broadly. There are also important lessons for both private mobile phone companies and for their regulators. Relatively efficient regulation in Kenya has allowed a productive range of innovations in this product space. Other countries may consider changing the rules to allow or even encourage responsible mobile banking along Kenyan lines. International Trade Does Africa have some particular problem developing goods and services suitable for export? The authors of chapter 8 have a clear answer: no. In “African Export Successes: Surprises, Stylized Facts, and Explanations,” William Easterly and Ariell Reshef argue that the pattern of African exports is broadly in line with what we see elsewhere in the world. Export success is dominated by a small number of what they call “Big Hits.” Interestingly, such Big Hits are no more or less common in Africa as anywhere else. Some determinants of export success include: moving up the quality ladder, utilizing strong comparative advantage, trade liberalization, investment in technological upgrades, foreign ownership, ethnic networks, and personal foreign experience of the entrepreneur. Other successes are triggered by more idiosyncratic factors like entrepreneurial persistence, luck, and cost

Volume Introduction

9

shocks. At least in this reading of the evidence, the African export sector should have plenty of scope for growth. Textiles are a particularly important topic for any low- income country seeking to increase exports. The MFA (Multi-Fiber Arrangement) quotas on US imports of textiles, by inducing constrained countries to move into higher- quality products, created a favorable environment for low valueadded, fabric- intensive clothing production in countries with unused quotas. By allowing the least developed African countries to use third- country fabrics in their clothing exports to the United States, the African Growth and Opportunities Act (AGOA) of 2000 provided additional implicit effective subsidies to clothing that were multiples of the US tariffs on clothing imports. However, there is a darker side to this attempt to help Africa, as documented in chapter 9, “AGOA Rules: The Intended and Unintended Consequences of Special Fabric Provisions,” by Lawrence Edwards and Robert Z. Lawrence. Lesotho and other least developed African countries responded impressively to the preferences they were granted under AGOA with a rapid increase in their clothing exports to the United States. But this performance has not been accompanied by some of the more dynamic growth benefits that might have been hoped for—in part because these preferences discouraged additional value addition in assembly and stimulated the use of expensive fabrics that were unlikely to be produced locally. When the broader system of quotas was removed, constrained countries such as China moved strongly into precisely the markets in which AGOA countries had specialized. Although AGOA helped the least developed countries withstand this shock, they were nonetheless adversely affected. Preference erosion due to reductions in US clothing tariffs could similarly have particularly severe adverse effects on these countries. Conclusion If Africa is to develop economically, it will have to do so via the growth of firms in the private sector. The idea that it could be otherwise—for example, that the continent could achieve persistent income growth primarily based on state- owned enterprises—has now been thoroughly discredited. The two previous volumes in this series focus on what one might view as the preconditions for private- sector development: peace, a stable political system, effectively functioning institutions, and a population that is healthy and educated enough to engage in production. The nine chapters in this volume look more closely at economics of private production, with much of the focus being on the behavior of firms. Undoubtedly, these firms operate in a difficult institutional and physical environment. Outside of natural resource exports, industries that have to

10

Sebastian Edwards, Simon Johnson, and David N. Weil

compete in a world market (such as textiles) often struggle. At the same time, there exist pockets of dynamism, the most visible being mobile phones. In banking, the continent’s position trails significantly behind the rest of the world in terms of both allocative efficiency and access by the poor to financial services. In some places, however, there are rapid improvements, often aided by new technology. Volume IV completes our four- set volume with a series of chapters that are relevant for the question of whether or not recent African growth is likely to be sustained.

1

Resolving the African Financial Development Gap Cross-Country Comparisons and a Within-Country Study of Kenya Franklin Allen, Elena Carletti, Robert Cull, Jun Qian, Lemma Senbet, and Patricio Valenzuela

1.1

Introduction

Africa’s growth performance has long been disappointing, and it has been described as a tragedy by some commentators (e.g., Easterly and Levine 1997). Although less well documented and perhaps not surprising, the financial sectors of sub-Saharan African countries remain woefully underdeveloped, even relative to the standards of developing countries. This is despite the fact that most of these countries have undergone extensive financialFranklin Allen is the Nippon Life Professor of Finance and Economics at the Wharton School of the University of Pennsylvania and a research associate of the National Bureau of Economic Research. Elena Carletti is professor of finance at Bocconi University. Robert Cull is a lead economist in the Finance and Private Sector Development Team of the Development Research Group of the World Bank. Jun Qian is professor of finance at Shanghai Advanced Institute of Finance. Lemma Senbet is the William E. Mayer Chair Professor of Finance at the Smith School of the University of Maryland. Patricio Valenzuela is assistant professor of finance and economics at the Department of Industrial Engineering at the University of Chile. We thank the Financial Sector Deepening Trust Kenya (FSD Kenya), and in particular David Ferrand, for providing the household surveys, and Samuel Makome and John Staley for providing information on Equity Bank. We appreciate helpful comments from Patrick Alila, Meghana Ayyagari, Thorsten Beck, Shawn Cole, Sankar De, Marcella Lucchetta, Randall Morck, Chukwuemeka Okoye, Daniel Sarpong, Moses Sichei, Philip Strahan, and seminar/ session participants at Harvard Business School, Wesleyan University, the ERD Workshop in Accra, the Financial Intermediation Research Society meetings in Prague, the NBER African Successes 2011 conference in Zanzibar, and the Summer Research Conference at the Indian School of Business. We are grateful to Giulia La Mattina, Sailu Li, and Mircea Trandafir for excellent research assistance, and the Gates Foundation, NBER, and the authors’ respective institutions for financial support. We are responsible for all the remaining errors. For acknowledgments, sources of research support, and disclosure of the author’s or authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13362.ack.

13

14

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

sector reforms in the last two decades of the same proportions as other developing and emerging countries.1 The financial development gap in sub-Saharan Africa is stark, and the gap is visible in the data. For example, the population- weighted trends for developing regions in terms of liquid liabilities and private credit (both divided by gross domestic product [GDP]) show that the rest of the developing world has seen substantial improvement relative to Africa in the last two decades, though the patterns differ across regions. Based on the precrisis data, the liquid liabilities of African financial sectors averaged below 30 percent of GDP in 2007 (see table 1.2). In no other region of the developing world did that figure stand below 40 percent. The credit side of the picture looked even worse: private credit averaged 17 percent of GDP in Africa in 2007, compared with ratios ranging from 33 to 44 percent for other developing regions. These precrisis data show that the financial development indicators of Africa improved in the period 1995– 2005, but at a much slower pace than in other developing and emerging countries in Asia, Eastern Europe, Latin America, and the Middle East. The poor state of African financial development raises a number of important questions as to what went wrong with the financial reforms in Africa and what could be improved. Is African financial development slow in itself, or is it a reflection of broader economic and policy failures? Are the levels of financial development achieved in the developing world outside of Africa achievable for most African countries? What factors have inhibited African financial development to this point and how can they be overcome? Understanding these issues is of crucial importance given the ample empirical evidence of a linkage between financial development and economic development.2 Moreover, in the context of Africa, where poverty is so widespread, the positive finance- growth nexus is suggestive of a positive linkage between finance and poverty alleviation.3 There is very little rigorous academic research that addresses these questions. This chapter represents a first step in addressing key issues at the heart of African financial development. We have three goals. First, we assess whether African financial development is slower than it ought to be, using other developing countries as a benchmark. Second, we identify factors that have more pronounced impact on financial development in Africa than in other developing countries, using both country- and firm- level data sets. The 1. The reform packages included price and interest rate liberalization, removal of credit ceilings, introduction of a variety of measures for banking and capital market development, including financial regulatory schemes, and large- scale privatizations of state- owned enterprises (see, for instance, Nissanke and Aryeetey 1998; Senbet and Otchere 2006; Honohan and Beck 2007). 2. See, for example, Levine (2005), Levine, Loayza, and Beck (2000), and Rajan and Zingales (1998). 3. Recent cross- country studies also find links between financial development and (lower) incidence of poverty (Beck, Demirgüç-Kunt, and Levine 2007; Clarke, Xu, and Zou 2006).

Resolving the African Financial Development Gap

15

problem for banks that try to increase financial inclusion is whether they can generate sufficient profits to sustain their business model. In this regard, our third goal is to study Kenya, a country that has made significant strides in financial development in recent years, and Equity Bank, a leading private commercial bank devoted to providing banking services to lower- income people. We examine the impact of Equity Bank, along with other commercial banks, on financial access at the household level in Kenya. To calibrate the financial development “gap” between Africa and other developing countries, we first analyze the determinants of financial development in other developing countries (low- and middle- income countries) via regression models based on prior research.4 We use the regression coefficients to generate predicted levels of financial development, as measured by the ratios of liquid liabilities and private credit over GDP, for sub-Saharan African countries (excluding South Africa). Then we compare those predicted levels with the actual levels of financial development in the African countries. We find that most African countries tend to have lower levels of financial development than would be predicted based on their fundamentals. The average country falls 13 percentage points short of its predicted level for liquid liabilities over GDP and 12 percentage points for private credit over GDP. To identify common and unique factors associated with financial development (or lack of ) in Africa, we run regressions for other developing countries and African countries separately; we also examine the combined sample with an African dummy, and interact the African dummy with factors affecting financial- sector development. One of our main findings is that population density appears to be more important for banking- sector development as measured by liquid liabilities and private credit (relative to GDP) in Africa than in other developing countries. If frequent interactions among firms, households, and investors are a necessary condition for business transactions, and hence financial development, then our results are plausible in that many African countries are endowed with scattered population and roads. Furthermore, we find a nonlinear relationship between population density and banking- sector development, and the largest gap between Africa and the rest of the developing world is for those African countries “trapped” in low- density areas. We further explore the channels through which population density may affect banking- sector development by examining additional variables related to population density. We find that the percentage of population living in the largest cities, roads per square kilometer, and bank- branch penetration are all associated with a higher level of liquid liabilities relative to 4. Our main financial development variables cover the period 1995– 2007. The negative impact of the 2007– 2009 global financial crisis has been less severe in Africa than in other regions, in part because the financial sectors were more isolated from global markets. See, for instance, Kamil and Rai (2010) for more details.

16

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

GDP, but the results are weaker for the level of private (bank) credit. These results imply that the minimum viable banking- sector scale is best achieved in major cities, and that technological advances such as mobile telephone banking, could be one way to facilitate African financial development, especially on the savings side of the banking sector outside metropolitan areas.5 Concerning other factors explaining banking- sector development, we find that macroeconomic stability and broad measures of institutional quality are less important in Africa than in other developing countries, while the natural resource “curse” (as in, for example, Sachs and Warner 1995, 2001) is no worse in Africa than elsewhere. Finally, regarding indicators of stock market development, such as market capitalization and turnover, we do not find any of the factors associated with stock market development in the rest of the world to be significant in Africa. In large part this is because most African stock markets remain thin, illiquid, and dysfunctional; only very recently have sub-Saharan African countries made a policy commitment to their development. We also conduct firm- level analyses that enable us to control for both country- level variables and firm characteristics in the same regressions. Employing the widely used global World Bank Investment Climate Surveys (ICS), which are comprised primarily of firms that are not listed on stock markets, we examine factors determining firms’ access to external finance in the form of bank loans (from both domestic and foreign banks) and/or credit cards.6 The idea is that if a country’s banking sector is more developed, firms, and in particular privately owned firms, should have easier access to these types of formal finance. We run regression models where we include the same country- level factors as in our cross- country studies as well as firm characteristics that could explain substantial variations in the demand for and in the use of financial services. Once again, we include the African dummy and interact it with both country- level and firm- level variables. The results of the microlevel regressions confirm that population density is an important factor for access of firms to financial institutions. While bank branch penetration is an important factor for access to bank finance, the coverage of roads is not as important for African firms as their counterparts elsewhere. As with country- level studies, general infrastructural failures are not necessarily responsible for low levels of firm access to bank finance in Africa, but rather failures are specific to the banking sector and markets for loans. While manufacturers are more likely to access bank finance in other 5. Mbiti and Weil (2011) use the same household surveys in Kenya as in our chapter, and find that increased use of M-Pesa, one of the leading mobile phone- based money transfer systems, lowers the propensity of people to use informal savings mechanisms, but raises the probability of their use of banking services. 6. This is consistent with prior research showing that financial institutions provide the most important source of external finance in most developing countries (e.g., Beck, Demirgüç-Kunt, and Levine 2000).

Resolving the African Financial Development Gap

17

countries, they show no strong tendency to rely on external finance relative to other firms in Africa. Having established population density as a key factor for financial development in Africa based on both cross- country and firm- level data, we next conduct a within- country study on Kenya. There are several reasons for this choice. First, based on our cross- country analysis, Kenya’s level of financial development is not too far off from the predicted level, and it has witnessed a strong bank- branch expansion in recent years. This expansion has also coincided with the emergence of Equity Bank, a pioneering (forprofit) commercial bank that devised a banking- service strategy targeting low- income clients and traditionally underserved territories. The bank is listed on the Nairobi Stock Exchange, and has no government ownership share. Many scholars and policymakers agree that a key obstacle to financial development is access of the disadvantaged to finance, which would promote economic growth at the broadest scales. While the success of microfinance institutions (MFIs), such as Grameen Bank, has captured the attention of economists and policymakers, many MFIs currently are beset by nonrepayment problems while most established commercial banks view the sectors targeted by MFIs as “unbankable.” This makes Equity Bank and Kenya a particularly interesting “laboratory” for the study of financial access. Since institutional and macroeconomic conditions tend to be more homogeneous within a single country, the findings from our within- country study can reinforce the results from our cross- country study. The absence of microlevel data at both the household and bank- branch level has made the study of financial inclusion a challenge. Fortunately, we are able to overcome that challenge using a new data set on bank- branch penetration at the district level described in Allen et al. (2011), matched with household surveys of financial usage for Kenya that were conducted in 2006 and 2009. We find that Equity Bank presence is strongly associated with increases in households’ usage of bank accounts and bank credit between the two surveys. These effects are particularly strong for underprivileged households—those with low income and less education, who do not own a permanent house, and those that lack any member with a salaried job. We also find differences in branching strategies between Equity Bank and other commercial banks. While all bank types (including Equity Bank) open a greater number of branches in urban, highly populated, and Englishspeaking districts, Equity Bank was more likely to expand to underdeveloped districts than other types of banks between 2006 and 2009—less densely populated areas and areas where the dominant language was not English or Swahili. This different branching strategy is consistent with our earlier result that population density is a major obstacle for financial- sector development in Africa; more importantly, our results are encouraging in the sense that private institutions can arise in such an environment to help overcome the “no finance, no growth trap” in many African countries.

18

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

The rest of the chapter is organized as follows. Section 1.2 describes the data, methodology for benchmarking financial development across countries, and results on examining whether the actual financial- sector development in Africa is below what the fundamentals would predict using other developing countries as benchmarks. Then section 1.3 examines whether the variables associated with banking development in other developing countries from our base models are related in a similar way to African banking development. Section 1.4 presents results on the determinants of firm access to external finance in Africa and other developing countries. Section 1.5 presents results on financial access in Kenya and on the effect of bank- branch expansion, particularly Equity Bank branching strategy on the use of bank accounts and credit. Section 1.6 offers concluding remarks. 1.2

Benchmarking Financial Development

For cross- country comparison of financial development, the literature has turned increasingly to regression analysis to examine the level and variation of financial development across countries relying on some of the same variables that have been used to study the links between financial development and growth (Levine 2005).7 There is no general consensus in the literature about the factors that should be taken into account in explaining financial development, nor what indicators of that development are most reliable. In the choice of the latter, we ground our analyses on the recent attempt by Beck et al. (2008) to standardize the selection of financial- development indicators. Under this approach, the potential financial- development indicators are ranked on the basis of the following criteria: (a) the directness of their linkages to welfare, (b) the goodness of fit of regressions that explain variations in them, (c) their coverage in terms of countries and years, and (d) the degree to which an indicator is stable within a country from year to year, but varies substantially across countries.8 In most of our analysis we use the two standard indicators, namely the ratio of liquid liabilities in the banking system to GDP and the ratio of credit to private sector to GDP. Both measures score among the highest based on all four criteria listed above. They score especially high relative to others on the first criterion, because they are robustly associated with long- run economic growth (Levine 2005; Levine, Loayza, and Beck 2000). Moreover, our 7. As in other recent papers, we use these variables, including growth, to describe financial development (Cull and Effron 2008; Cull, Senbet, and Sorge 2005). By contrast, in the finance and growth literature, the financial indicators are among the explanatory variables used to explain growth. 8. Beck et al. (2008) call this the ratio of within- sample variance to between- sample variance. They worry that high within- country variation may reflect measurement errors or a high degree of comovement with the business cycle. They argue that indicators of financial development are (or at least should be) better suited to measuring longer- term differences across countries rather than fluctuations along the cycle for a given country.

Resolving the African Financial Development Gap

19

analysis is rooted in banking indicators because banks hold the vast majority of financial- sector assets in Africa and other developing countries.9 Despite this, we perform some robustness checks using measures of stock market development as indicators of financial development. In the choice of the explanatory variables for financial development, we rely again on previous studies, in particular those on the finance- growth nexus (e.g., Levine 2005) and from other studies that analyze the determinants of financial development (e.g., Beck et al. 2008; Cull and Effron 2008). These studies (in particular, Beck et al. 2008) regress the indicators of financial development on a set of variables that describe the environment in which such development takes place, but that are exogenous to that process such as population size and density, natural resources, and offshore centers. They also include per capita income as an exogenous regressor, claiming that its effect on financial development is contemporaneous while the effect of financial development on income is lagged. The residuals from those regressions, therefore, provide an indication of the extent to which the chosen government policies promote financial development. However, as our objective is to benchmark African financial development relative to a set of variables that have been robustly associated with financial development in countries outside of Africa, especially in low- and middle- income countries, we expand further the set of regressors by including macroeconomic variables such as inflation, real growth, and the current account balance, broad measures of institutional development, and variables describing banking- sector structure. We stress from the outset that we are not necessarily estimating causal relationships for the expanded set of regressors. For ease of exposition, however, we refer to all explanatory variables as determinants of financial development throughout the chapter. 1.2.1

Regression Model

We begin the introduction of our regression model with the dependent variable: Financial Development Indicator (FDi). This measures an indicator of financial development in country i. We average our indicators of financial development and our explanatory variables over multiple years (from 2001 to 2005), as is customary in the literature on financial development and growth, so as to reduce the influence of outliers. We therefore have only one observation per country. The results are qualitatively similar when we use all yearly observations from 1990 to 2006 in panel regressions, and when 9. The ratio of private credit to GDP can include lending by nonbank financial institutions. As table 1.3 shows, though, the correlations between private credit/GDP and stock market capitalization, and those between liquid liabilities/GDP and stock market capitalization/GDP are of similar magnitude (0.65 and 0.68), suggesting that the private credit measure is not driven by the activities of nonbank financial institutions involved in stock market transactions.

20 Table 1.1

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela Summary statistics on the country sample

(1) FDi = α + β1Populationi + β2Population Densityi + β3Natural Resourcesi + β4Offshore Centeri

Exogenous determinants, based on Beck et al. (2008)

+ β5Per Capita Incomei + β6Population ∗ GDP Per Capitai

Plausibly exogenous, financial development affects these variables at a lag, Beck et al. (2008)

+ β7Real GDP Growth Ratei + β8Inflation Ratei + β9Current Account Balance/GDPi

Macroeconomic variables

+ Β10KKM Indexi

Index of institutional development

+ Β11Manufacturing/GDPi + Β12Secondary/Primary enrollment + εi

Other variables

we lag the explanatory variables in either the panel regressions or the crosssectional regressions. Because our goal is to describe a general picture of the factors that are robustly linked to financial development, however, we present below only the simplest cross- country regressions in which the financial indicators and explanatory variables are contemporaneous.10 The regression model for the expanded set of explanatory variables is shown in table 1.1. We briefly explain our variables below: Population. We measure population both with size and density and we expect both variables to be positively associated with financial development. A larger population should spur financial development due to scale and networking effects that make provision of financial services more efficient in larger economies. Population density, as measured by the number of residents per square kilometer, should have a positive coefficient in part because it is easier for financial institutions to accumulate savings when a higher number of potential depositors have easy access to them. Natural Resources. An abundance of natural resources may have a negative effect on financial development via the so-called “resource curse.” Consistent with this, Sachs and Warner (1995, 2001) offer evidence that resourcerich developing countries have grown more slowly since 1960 than other developing countries. We measure the intensity of a country’s reliance on natural resources by using a comprehensive approach that measures resource abundance based on trade indicators rather than solely on oil exports: 10. For example, in the cross- sectional regression we average the explanatory variables from 1990 to 2000 and use them to explain financial- development indicators averaged from 2001 to 2005. In the panel regression, we lag the explanatory variables by one to three years.

Resolving the African Financial Development Gap

Natural resources = ∑ k

Exports k − Imports k , Population ages 15–64

⎧⎪ petroleum, forest, tropical, where k ∈ ⎨ animal, cereal, raw materials ⎩⎪

21

. ⎫⎪ ⎬ ⎭⎪

The key advantage of this approach is that this measure of net exports is available for most countries and, as shown by Lederman and Maloney (2008), it is more closely linked to actual natural resource reserves than other trade- based endowment measures. If there is a resource curse and it impacts financial development, we would expect the coefficient on this variable to be negative. Offshore Centers. The financial sectors of offshore centers are typically much larger than their economies would otherwise warrant. We measure this effect with a dummy variable for offshore financial centers and we expect it to be positive.11 Per Capita Income. Per capita income is expected to be positively linked to financial development because the volume and the sophistication of financial activities demanded is greater in richer countries and, on the supply side, richer economies can better exploit scale economies in the provision of financial services. Per capita income reflects an aggregation of past growth, and hence its coefficient summarizes the long- run positive relationship between growth and financial development. Moreover, per capita income may interact with population to produce even greater financial development than they do individually. We test this by including the interaction between those two variables in the regression. Indicators of Macroeconomic Stability. Financial development is more likely in a sound macroeconomic environment, and to measure this we include real growth, inflation, and the current account balance (relative to GDP) in the regressions. 1. Real Growth: The effect of real growth on financial development is ambiguous. On the one hand, countries with rapid growth may be associated with greater financial development. On the other hand, countries with higher levels of development, as reflected in GDP per capita, tend to have slower growth according to “conditional convergence” (Levine and Renelt 1992; Easterly and Levine 1997). Given that financial- development indicators tend to be highly correlated with per capita income, it is likely that the faster growing countries will also tend to have lower levels of financial development. We would therefore expect a negative coefficient for real growth in our 11. One could question whether this is an exogenous variable. We follow Beck et al. (2008) in taking it as exogenous.

22

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

financial- development regressions. The coefficient should also capture the negative short- run relationship between financial development and growth as emphasized in Loayza and Ranciere (2006). 2. Inflation: On the private credit side, inflation should slow financial development if it makes loan contracting over extended periods more difficult. Inflation could also have a dampening effect on liquid liabilities, making depositors more hesitant to place their savings in the formal financial system for fear of not being able to get them back quickly enough. Therefore, we expect the coefficient for inflation to be negative in our regressions. 3. Current Account Balance: The current account balance can be seen as a rough indicator of the health of the macroeconomic environment, and we thus expect it to be positively associated with our financial indicators. Institutional Development. We include in the regression KKM, which is the measure of broad institutional development created by Kaufmann, Kraay, and Mastruzzi (2007). Institutional development has been found to foster financial development in developing countries (Cull and Effron 2008), and thus we expect a positive coefficient for KKM in our regressions. Manufacturing. We include the share of GDP generated by the manufacturing sector. Industrial sectors that are relatively more in need of finance tend to grow faster in countries with well- developed financial sectors (Rajan and Zingales 1998). Manufacturing encompasses a broad variety of activities that tend to rely heavily on external finance so that we expect countries with a large manufacturing sector to have well- developed complementary financial institutions. We therefore expect a positive coefficient for manufacturing in our regressions. Secondary/Primary Enrollment. Finally, we want to measure the impact of risk management on financial development. The idea is that a lack of capacity in risk management may be a deterrent to banking- sector development and broader financial- sector development (e.g., stock markets). As measuring financial capabilities across countries directly is not possible, we proxy risk- management capability with the ratio of secondary to primary school enrollment and we expect its coefficient to be positive. Our argument is that risk- management capability is fundamentally a question of human capital development and thus of talented financial people. The summary statistics and the correlations between the key variables appear in tables 1.2 and 1.3. Table 1.2 shows that the mean values for the financial indicators are uniformly lower in Africa than the rest of the world, measured in terms of liquid liabilities, private credit provision, and stock market development. We also see marked differences in the explanatory variables between Africa and the rest of the world, particularly with respect to population density, per capita income, current account balance/GDP, and institutional development as proxied by the KKM index. The correlations in table 1.3 are instructive. All indicators of financial

Resolving the African Financial Development Gap Table 1.2

23

Summary statistics on the country sample World (minus Africa)

Sub-Saharan Africa

Variable

Mean

Standard deviation

Mean

Standard deviation

Liquid liabilities/GDP (%) Private credit/GDP (%) Stock market capitalization/GDP (%) Stock market value traded/GDP (%) Ln(population) Ln(population density) Natural resources Offshore center (%) Ln(per capita income) Population ∗ GDP per capita Real GDP growth rate (%) Inflation rate (%) Current account balance/GDP (%) KKM index Bank concentration Foreign-ownership share (%) State-ownership share (%) Manufacturing/GDP (%) Secondary/primary school enrollment Roads/area Railroads/area Urban population (%) Geographic branch penetration Demographic branch penetration

64.2 57.7 52.1 34.1 2.44 0.44 0.5 4.4 2.25 0.48 4.1 5.2 0.2 0.33 0.65 27.1 15.9 16.8 0.81 1.07 0.03 63.6 29.76 16.51

47.4 45.3 60.0 50.9 1.59 1.94 2.41 20.7 1.04 1.35 2.5 5.2 8.1 0.9 0.19 25.9 19.7 6.1 0.24 1.65 0.03 20.7 80.07 17.28

27.5 17.6 25.6 6.5 2.24 0.09 0.15 0.0 0.38 0.03 4.8 9.3 –3.8 –0.54 0.81 44.4 13.3 11.0 0.33 0.21 0 36.2 7.97 2.86

17.7 22.7 43.8 20.0 1.33 0.12 0.77 0.0 0.94 0.07 2.7 15.0 6.7 0.58 0.14 24.4 16.6 7.3 0.18 0.22 0 17.2 22.49 3.64

Sources: For liquid liabilities/GDP, private credit/GDP, stock market capitalization/GDP, stock market value traded/GDP, and bank concentration: World Bank Database on Financial Development and Structure. Natural resources is the Lederman and Maloney (2008) measure of net exports in resource-intensive industries as described in the text. Offshore center is a dummy variable defined by IMF (2007). The KKM index is the measure of broad institutional development created by Kaufmann, Kraay, and Mastruzzi (2007). Foreign- and stateownership share of banking-sector assets are from the World Bank Database of Bank Regulation and Supervision (2008). Geographic and demographic branch penetration are from CGAP (2009). All other variables are from World Development Indicators.

development—liquid liabilities/GDP, private credit/GDP, stock market capitalization/GDP, and stock market value traded/GDP—are highly correlated; population density is positively associated with two of them, liquid liabilities/GDP and stock market capitalization/GDP. The macroeconomic variables are strongly associated with the measures of financial development in an expected way. In particular, there is a negative relationship between the financial- development indicators and both inflation and real growth, and a positive relationship between the indicators and the current account balance. However, the relationship with real growth is weaker than

0.38* 0.18 –0.36* –0.20 –0.48* –0.35*

0.22* 0.79* 0.01

0.13 –0.17 –0.40*

0.37* 0.51* –0.11

0.25*

0.08 –0.14 –0.07

0.13 0.61* 0.06 –0.06 0.10 0.08 0.71* 0.51*

0.38* –0.22* 0.19 0.49*

0.51*

0.27*

–0.17 –0.02 –0.28* –0.28* 0.08 –0.15

0.45* 0.51* 0.12

0.70* 0.75* 0.03 –0.02

0.40* –0.08

1

0.65*

0.68*

0.34*

–0.25* –0.14 –0.02

0.27* 0.50* 0.04

0.47* –0.25* –0.32*

0.10 0.03 –0.08 0.51*

1 0.24*

–0.18

–0.43* 0.43* 0.35*

0.02 –0.19 –0.50*

0.47* –0.18 0.15

–0.20 –0.19 –0.25* –0.15

1

–0.01

0.21 –0.11 –0.18

0.58* 0.13 0.07

–0.05 0.31* –0.18

1 –0.21* 0.41* 0.18

Stock Stock mkt. mkt. val. capit./ traded/ Ln(pop GDP GDP Ln(pop) density)

1

Priv. credit/ GDP

0.64*

Liquid liab./ GDP

Correlations tests: World (minus Africa) sample

*Significant at the 5 percent level.

Private credit/GDP Stock market capitalization/GDP Stock market value traded/GDP Ln(population) Ln(population density) Natural resources Offshore center Ln(per capita income) Population ∗ GDP per capita Real GDP growth rate Inflation rate Current account balance/GDP KKM index Bank concentration Foreign-ownership share State-ownership share Manufacturing / GDP Secondary/primary school enrollment

N = 90

Table 1.3

–0.07 0.22* –0.16

1 0.04

Offsh. center

0.25* –0.12

–0.09 0.15 –0.09 –0.15 –0.26* –0.12

0.25* 0.18 0.16 0.01 0.21 –0.02

–0.12 0.00 –0.08

1 –0.07 0.22*

Nat. res.

0.76*

0.14 –0.41* 0.03

0.34* 0.88* 0.03

0.26* –0.26* –0.39*

1

Ln(per capita income)

1 –0.04

Real GDP gr. rate

0.14

–0.15

–0.23* 0.11 –0.03 0.08 0.05 –0.07

–0.02 0.09 0.21* –0.27* –0.35* –0.03

1 –0.19 –0.10

Pop ∗ GDP per capita

–0.32*

–0.13 0.26* 0.06

–0.17 –0.48* –0.16

1

Infl. rate

0.08

0.02 –0.05 –0.15

1 0.17 –0.06

CA balance/ GDP

0.76*

0.14 –0.39* 0.10

1 0.18

KKM index

0.19

0.08 –0.16 –0.22*

1

Bank conc.

0.16

1 –0.32* –0.02

Foreignown share

–0.33*

1 0.11

Stateown share

0.07

1

Manuf./ GDP

Resolving the African Financial Development Gap

25

the other macrovariables. The KKM index of institutional development is strongly and positively associated with all of the indicators of financial development. The same holds for the log of per capita income. This is indicative of a positive long- run relationship between finance and growth. Finally, the secondary/primary school enrollment is positively associated with the measures of financial development. 1.2.2

Empirical Results at Cross-Country Level: The African Financial Development Gap

To benchmark African financial development, we estimate regression equation (1) for countries outside Africa, which enables us predict what African financial development should be based on the experience of these other countries. Specifically, we first run the regressions excluding all African countries, and we derive out- of-sample predictions for African financial development. Then we compare these predictions with the actual levels of African financial development to measure the development gap.12 We present models for all countries and for low- and middle- income countries separately in table 1.4. The latter are more reflective of the African experience. Benchmarking Results for the Determinants of Financial Development The results for regression equation (1) are presented in table 1.4. The signs of almost all of the coefficients are as predicted and many of them are significant. When we use the full sample of countries and the limited set of regressors as a benchmark, (see columns [1] and [5]), only the per capita income variable is positively associated with both indicators of banking development, while our proxy for natural resource intensity is negatively linked to those indicators. Population density and offshore financial center are significantly positive in the liquid liabilities regression (column [1]) but not the private credit regression (column [5]). When we benchmark to a sample of only low- and middle- income countries, using the limited set of regressors (columns [3] and [7]), the offshore financial center is highly significant for both indicators, while the per capita income and population density variables are positive and significant, but only in the private credit regression.13 12. The same approach has been used, for example, to assess whether the levels of foreign direct investment received by China are abnormally high, or whether they can be explained by fundamentals (Fan et al., forthcoming). 13. These results are similar to those in Beck et al. (2008). Aside from our proxy for natural resource intensity, our regressions differ from theirs in two ways. First, we use the more conservative cross- sectional approach, whereas they use the full panel. We choose to do this because errors from observations from the same country are likely to be correlated with one another. This could artificially deflate standard errors, thus increasing the significance level of coefficients. For robustness we also ran our models on the full panel, and qualitative results are similar. Second, they include a poverty- gap variable in their regressions, which is the proportion of the population under the poverty line, times the average distance from the poverty line (Source: Povcal Net, World Bank). Since the poverty gap is so tightly linked to income levels,

26

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

The expanded regression results, where we include macroeconomic, institutional, and other explanatory variables, are presented in columns (2), (4), (6), and (8). The results show that real growth is negative and significant (or very nearly so) for both private credit and liquid liabilities for both samples of countries that we use. This indicates that the fastest- growing countries tend to have the least- developed financial sectors, providing support for conditional convergence. As discussed earlier, this is likely a reflection of a short- run relationship between financial development and growth (the positive significant coefficient for per capita income summarizes the long- run relationship between accumulated growth and financial development). Inflation is negative and highly significant across financial indicators and country samples. The coefficients are similar in size for the full set of countries and for the low- and middle- income countries, indicating that macroeconomic instability is strongly linked to financial underdevelopment even in developing countries. The current account balance/GDP is strongly positively linked to both indicators for both samples, though coefficients are larger for the low- and middle- income sample. In short, the coefficients support our hypotheses and indicate strong links between macroeconomic outcomes and financial development. Our proxy for the degree of institutional development, as represented by the KKM index, is positive and highly significant across financial indicators and samples, providing support for the notion that broad institutional development helps to foster financial development. The coefficients are larger for the full- country sample than for the sample of low- and middle- income countries, but they are still large even in the latter case: a one- point increase in the KKM index (five- point scale) is associated with a 22 percentage point increase in liquid liabilities/GDP and a 21 percentage point increase for the private credit ratio. Finally, in the expanded models, the offshore financial center variable is significant for both indicators for low- and middle- income countries. Predicted versus Actual African Financial Development We now use the regression coefficients in table 1.4 to derive a predicted level of financial development for each country in Africa. Because they are likely to be more reflective of the African experience, as our benchmark we rely on the models that include only low- and middle- income countries.14 Again, we are not claiming that the relationships we find in table 1.4 are they use the residuals from a regression of the poverty gap on income in their regressions. We did the same, but the variable provided little explanatory power and reduced our sample. Also, unlike the other variables in our analysis, the poverty counts are based on surveys that do not occur at regular intervals, and thus the panel is highly unbalanced. For these reasons, we present models without the poverty gap variable. 14. We also tried to estimate models based only on low- income countries outside of Africa, but there were too few observations to generate meaningful results.

KKM index

Current account balance/GDP

Inflation rate

Real GDP growth rate

Population ∗ GDP per capita

Ln(per capita income)

Offshore center

Natural resources

–0.017 (0.021) 0.074*** (0.023) –0.038*** (0.014) 0.247* (0.145) 0.222*** (0.036) 0.004 (0.032)

(1)

All

–0.003 (0.028) 0.031 (0.027) –0.069*** (0.019) 0.133 (0.170) 0.061 (0.083) –0.013 (0.033) –2.015 (1.514) –1.219 (0.751) 1.519*** (0.556) 0.239*** (0.093)

(2) –0.001 (0.019) 0.036 (0.022) –0.034 (0.033) 0.369*** (0.112) 0.056 (0.039) 0.006 (0.100)

(3) –0.035 (0.030) 0.032 (0.022) –0.062* (0.036) 0.341*** (0.121) –0.091 (0.065) 0.061 (0.101) –2.619** (1.159) –0.946* (0.505) 1.565** (0.658) 0.215*** (0.080)

(4)

Low + middle income

Liquid liabilities/GDP

0.003 (0.018) 0.018 (0.019) –0.022* (0.012) 0.191 (0.123) 0.275*** (0.031) 0.071** (0.027)

(5)

Regressions on banking-sector development for the sample of non-African countries

Ln(population Density)

Ln(population)

Table 1.4

All

0.015 (0.021) 0.019 (0.020) 0.0002 (0.014) 0.159 (0.127) –0.016 (0.062) 0.062** (0.024) –3.129*** (1.126) –1.092** (0.559) 0.392 (0.414) 0.362*** (0.069)

(6)

0.002 (0.017) 0.032* (0.019) –0.010 (0.028) 0.374*** (0.096) 0.089** (0.034) –0.031 (0.086)

(7)

–0.027 (0.026) –0.020 (0.019) –0.026 (0.031) 0.339*** (0.102) –0.063 (0.055) 0.015 (0.084) –1.769** (0.974) –0.894** (0.424) 1.236** (0.553) 0.210*** (0.067) (continued)

(8)

Low + middle income

Private credit/GDP

(continued)

0.41 111

0.383*** (0.123)

(1)

All

0.51 97

–1.313** (0.652) –0.335 (0.241) 1.188*** (0.223)

(2)

0.14 75

0.467*** (0.101)

(3)

0.36 67

0.436 (0.580) –0.040 (0.179) 0.990*** (0.167)

(4)

Low + middle income

Liquid liabilities/GDP

0.54 111

–0.041 (0.104)

(5)

All

0.70 97

–0.127 (0.485) –0.130 (0.179) 0.782*** (0.166)

(6)

0.22 75

0.248*** (0.087)

(7)

0.43 67

0.651 (0.487) –0.004 (0.150) 0.640*** (0.141)

(8)

Low + middle income

Private credit/GDP

Notes: This table presents ordinary least squares (OLS) regressions of banking-sector development, measured by liquid liabilities/GDP and credit to the private sector extended by deposit money banks/GDP ratios, on a set of country-level variables including endowment (population and resources), macroeconomics, institutions, banking structure, and other variables. We present models for all non-African countries and for (non-African) low- and middleincome countries separately. Standard errors are presented in the brackets below coefficients. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Adjusted R2 Observations/countries

Constant

Secondary/primary enrollment

Manufacturing/GDP

Table 1.4

Resolving the African Financial Development Gap

Fig. 1.1

29

Liquid liabilities/GDP in African countries, actual versus predicted values

Notes: Based on the specification in table 1.3, column (4); negative predicted values are replaced by zero.

causal. Rather, we are asking what the level of financial development would be if the same relationships held in Africa as in the rest of the developing world. To the extent that predicted and actual levels of financial development are similar, one can say that African financial development is about what it should be. 1. African Financial Development Gap: Liquid Liabilities. Figure 1.1 shows that only seven of thirty- one countries have levels of liquid liabilities/GDP that are at or above their predicted levels. Of those, three (Angola, Sierra Leone, and Zambia) are huddled in the lower- left- hand corner of the figure where actual and predicted values are very low. Because the predicted values are based on linear regressions, they tend to be very near zero for these countries.15 That their actual levels exceed zero by some small amount is little consolation. To a lesser extent the same is true of Mozambique. Ethiopia has predicted and actual levels of liquid liabilities that are quite similar. The result on Ethiopia seems surprising given the country is known 15. Indeed, for Angola the predicted value is negative (and large in absolute value). This is due to its high average GDP growth (10.6 percent), low KKM score (–.71), and high inflation (86.4 percent per year). Negative levels of financial development are not, however, possible. For countries that have negative predicted values, we reset them to zero in both the figures. The calculations are discussed in the text.

30

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

to have undertaken few financial- sector reforms and has a banking system dominated by the government. While state- owned banks collect deposits, it is unlikely that they intermediate (lend) well. In fact, the dominant bank in Ethiopia was known to have a high accumulation of liquidity and idle funds during the sample period, and the relatively high level of liquid liabilities (relative to GDP) may be reflective of excess liquidity. Overall, there are only two countries, namely Cape Verde and Mauritius—among the smallest in Africa—that exceed their predicted levels by a substantial amount, but neither of the two is particularly reflective of the African experience. As figure 1.1 shows, twenty countries have levels of liquid liabilities that are more than 10 percentage points below their predicted levels. The worst ten performers, for example, range between 24 and 47 percentage points below. The average country falls 13 percentage points short of its predicted level, which is quite sizable, given that the average ratio of liquid liabilities to GDP hovered between 26.5 and 30.9 percent from 2001 to 2005. 2. African Financial Development Gap: Private Credit. As shown in figure 1.2, the level of private credit/GDP for African countries lies far below predicted levels. Again, only seven countries exceed their predicted levels and only Mauritius does so by more than 10 percentage points. Of the twentyfour countries with predicted levels that are higher than their actual private

Fig. 1.2

Private credit/GDP in African countries, actual versus predicted values

Notes: Based on the specification in table 1.3, column (8); negative predicted values are replaced by zero.

Resolving the African Financial Development Gap

31

credit ratios, seventeen fall short by more than 10 percentage points. The ten poorest performers range from 22 to 34 percentage points below their predicted levels, and the average country falls short of its predicted level by 12 points. The magnitude of this measure of underperformance is quite large when one considers that the average ratio of private credit to GDP stood at 15.5 percent in 2005. The gap in the observed African financial development, relative to what is predicted by the benchmark determinants, is stark. The levels of liquid liabilities to GDP for African countries are about two- thirds the level predicted by statistical relationships that hold elsewhere in the developing world. Private credit ratios are even lower, slightly better than half their predicted levels. There is also general consistency in rankings across the two indicators. Of the bottom ten performers in terms of liquid liabilities, seven are also among the bottom performers in terms of private credit, and the correlation for the differences between predicted and actual levels for the two measures of financial development is 0.85. Our results point to systematic underperformance relative to the fundamentals that have been associated with banking- sector development in the rest of the world. To analyze this gap more in depth, in the next sections we look at whether the factors in our base models relate to African financial development differently than to financial development in the rest of the world. 1.3

Are the Determinants of African Financial Development Different?

So far we have defined underdevelopment in African financial sectors in terms of the determinants of financial- sector development in other parts of the developing world. However, the course of African financial- sector development might depend on a different set of factors than those that have been important elsewhere.16 While we are reluctant to accept that African financial sectors have a distinct model of development, it seems plausible that some factors may be somewhat more or less important in the African context. To see whether this is indeed the case, as a first step, we estimate the model in table 1.5 for the sample of African countries. Note that this method essentially accepts that the level of financial development in Africa is lower than that in the rest of the developing world, and then tries to explain variation around the African mean based on the explanatory variables in our base models. Still, the results are instructive.17 First, population density is much more strongly linked to both liquid liabilities/GDP and private credit/GDP than it was elsewhere in the world. 16. It may also be that the factors explaining financial development in Africa are the same as for the other countries, but these factors load up or contribute differently in the African context. We are open to either interpretation. 17. Note that the offshore financial center variable does not appear in table 1.5 because no African countries qualify.

Inflation rate

Real GDP growth rate

Population ∗ GDP per capita

Ln(per capita income)

Offshore center

Natural resources

0.038 (0.035) 0.441 (0.520)

–0.035 (0.026) 0.074*** (0.019) –0.033 (0.036) —

(1)

All

0.022 (0.078) –0.281 (0.508) –0.447 (1.031) 0.060

–0.004 (0.026) 0.080*** (0.022) –0.013 (0.047) —

(2)

0.043 (0.037) 0.057 (1.12)

–0.029 (0.031) 0.075*** (0.020) –0.032 (0.037) —

(3)

0.018 (0.080) 0.221 (1.197) –0.499 (1.057) 0.058

–0.011 (0.030) 0.077*** (0.024) –0.010 (0.049) —

(4)

Without South Africa

Liquid liabilities/GDP

Regressions on banking development for the sample of African countries

Ln(population density)

Ln(population)

Table 1.5

0.009 (0.032) 2.941*** (0.477)

–0.058** (0.024) 0.0314* (0.018) –0.030 (0.033) —

(5)

All

0.054 (0.068) 2.396*** (0.442) –0.607 (0.898) –0.076

–0.021 (0.023) 0.049** (0.020) –0.001 (0.041) —

(6)

0.048* (0.027) –0.141 (0.816)

–0.007 (0.022) 0.043*** (0.014) –0.024 (0.027) —

(7)

0.073 (0.056) –0.163 (0.834) –0.343 (0.736) –0.066

0.013 (0.021) 0.061*** (0.016) –0.015 (0.034) —

(8)

Without South Africa

Private credit/GDP

0.36 38

0.548*** (0.078) 0.53 33

(0.192) –0.082 (0.587) 0.116* (0.062) –0.376 (0.455) 0.372 (0.247) 0.528*** (0.2613) 0.34 37

0.545*** (0.079) 0.52 32

(0.196) –0.207 (0.656) 0.119* (0.063) –0.341 (0.470) 0.379 (0.252) 0.518*** (0.3044) 0.64 38

0.297*** (0.071)

0.78 33

(0.167) –0.879* (0.511) 0.073 (0.054) –0.085 (0.396) 0.179 (0.215) 0.260* (0.130) 0.24 37

0.275*** (0.058)

0.45 32

(0.136) –0.241 (0.457) 0.058 (0.044) –0.265 (0.327) 0.144 (0.175) 0.309*** (0.243)

Notes: This table presents OLS regressions of banking-sector development, measured by liquid liabilities/GDP and credit to the private sector from deposit money banks/GDP ratios, on a set of country-level variables including endowment (population and resources), macroeconomics, institutions, banking structure, and other variables. We present models for all the sub-Saharan African countries, with and without South Africa. Standard errors are presented in the brackets below coefficients. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Adjusted R2 Observations/countries

Constant

Secondary/primary enrollment

Manufacturing/GDP

KKM index

Current account balance/GDP

34

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

In fact, among the set of exogenous factors in the regression equation (1), population density is the only one that is robustly linked to our indicators of banking development for Africa. Moreover, the interaction between population and per capita income is positive and significant in two of the private credit regressions, although not when South Africa is dropped from the African sample. This is reasonable, given the relatively high level of economic development of this country. Second, among the factors that explain financial development in the rest of the developing world, only the KKM index of institutional development seems relevant for Africa, but only for the liquid liabilities measure. The other additional factors in the expanded regressions in table 1.5 are less important in Africa than elsewhere. Real growth is not significant. We argued earlier that the negative, significant coefficient for growth in the rest of the developing world was consistent with conditional convergence, because it indicated that the least financially developed countries had higher current growth rates. The nonresult for Africa suggests that such convergence did not occur there, a point that was also made for some African countries in Pritchett (1997). In addition, neither inflation nor the current account balance is significant in the African regressions, whereas both were highly significant in the rest of the world as shown in table 1.4. The poor results for inflation, the current account, and growth indicate that the macroeconomic fundamentals that are associated with financial development in other developing countries have not been important in Africa.18 1.3.1

A Closer Look at Differences in Financial Development Factors

The models in table 1.5 provide a strong initial indication that the factors associated with banking development in Africa differ from those in the rest of the world. However, those African models are based on only a small number of observations. To address the small sample issue we rerun our models for all low- and middle- income countries (including Africa) and include an interaction between each of our explanatory variables and an Africa dummy variable. We also include an Africa dummy variable on its own to test whether the constant for Africa in our regressions is the same as in the rest of the developing world. This is akin to a seemingly unrelated regression (SUR), but the coefficients on the interactions terms enable us to make more precise statements about whether the effects for Africa are 18. The results on macroeconomic fundamentals suggest that African countries were able to avoid the financial roller coaster ride associated with macroeconomic instability that other countries experienced during this period. However, this apparent benefit might have come at a substantial opportunity cost associated with the same low levels of financial development during this period. To some extent, therefore, our benchmark regressions used to predict African financial- sector development give credit to some African countries for having relatively low initial levels of development, stable prices, and a current account surplus. These factors contribute to higher predicted financial development than has actually occurred.

Resolving the African Financial Development Gap

35

statistically different than for the rest of the developing world and from zero (i.e., there is no effect). The results are presented in table 1.6. In the simplest regressions, we include the Africa dummy and the limited set of regressors (columns [1] and [4]). The Africa dummy is negative for both indicators but it is significant only in the liquid liabilities regression. Per capita income is positive and significant for both indicators; it remains positive when the Africa interactions are introduced (columns [2] and [5]), although it then becomes insignificant in the liquid liabilities regression. Moreover, per capita income is not significant when we introduce the African interaction terms for the expanded set of regressors (columns [3] and [6]). The offshore financial center variable is positive and significant across all specifications, though again no African country qualifies as an offshore center and so there is no Africa interaction for the variable. The most relevant explanatory variable for Africa is again population density. When only the limited set of regressors is included in the regression, population density is positive and highly significant for both indicators of financial development (columns [1] and [4]). When the Africa dummy is interacted with the limited set of regressors (columns [2] and [5]), the relationship is still positive but weaker in terms of significance. When interactions are included for the expanded set of regressors (columns [3] and [6]), the variable remains positive but is no longer significant. Despite the insignificance of the density variable in models 3 and 6, we reject that the sum of the coefficients for density and its interaction with the Africa dummy is zero at the p = 0.1 level for those models (see F-statistics near the bottom of table 1.6). Therefore, the relationship between population density and our banking development indicators for African countries is significantly different from zero, but not for the rest of the developing world in our most expansive models. In contrast, the relationship between natural resource intensity and banking indicators is about the same in Africa as it is in the rest of the developing world. The coefficient for our proxy of natural resource intensity is negative but tends not to be significant in most of the models; the coefficient for its interaction with the Africa dummy is never significant. Thus, we reject the hypothesis that the relationship between natural resources and banking development is somehow different for Africa. In other words, the relatively low levels of banking development cannot be attributable to the natural resource curse being more severe in Africa than in other parts of the developing world. Aside from the negative significant coefficient in model 3 (for liquid liabilities), the results provide little support for a resource curse in terms of financial development outside Africa as well. Concerning the regressors that are only included in the expanded set, the coefficient for real growth is negative and highly significant for both

Table 1.6

Regressions for the combined sample of low- and middle-income and African countries Liquid liabilities/GDP

Africa Ln(population) Ln(population density) Natural resources Offshore center Ln(per capita income) Population ∗ GDP per capita Real GDP growth rate

(1)

(2)

(3)

(4)

(5)

(6)

–0.105* (0.054) –0.009 (0.015) 0.050*** (0.016) –0.031 (0.025) 0.363*** (0.098) 0.052** (0.026) 0.030 (0.084)

0.078 (0.147) –0.001 (0.017) 0.036* (0.019) –0.034 (0.030) 0.369*** (0.101) 0.056 (0.035) 0.006 (0.089)

–0.472* (0.280) –0.035 (0.028) 0.032 (0.020) –0.062* (0.033) 0.341*** (0.110) –0.091 (0.059) 0.061 (0.091) –2.619** (1.053) –0.946** (0.458) 1.565** (0.598) 0.215*** (0.072) 0.436 (0.526) –0.040 (0.163)

–0.066 (0.045) 0.0005 (0.013) 0.036*** (0.013) –0.012 (0.021) 0.378*** (0.082) 0.070*** (0.022) –0.023 (0.071)

0.027 (0.124) 0.002 (0.015) 0.032* (0.016) –0.010 (0.025) 0.374*** (0.085) 0.089*** (0.030) –0.031 (0.075)

–0.331 (0.231) –0.027 (0.023) 0.020 (0.017) –0.026 (0.027) 0.339*** (0.091) –0.063 (0.049) 0.015 (0.075) –1.769** (0.869) –0.894** (0.378) 1.236** (0.493) 0.210*** (0.060) 0.651 (0.434) –0.004 (0.134)

–0.027 (0.048) 0.039 (0.035) 0.001 (0.061) –0.013 (0.065) 0.051 (1.634)

0.024 (0.054) 0.046 (0.041) 0.052 (0.081) 0.109 (0.136) 0.160 (1.833) 2.120 (1.930) 1.004* (0.547) –1.773 (1.168) –0.096 (0.121)

Inflation rate Current account balance/ GDP KKM index Manufacturing/GDP Secondary/primary enrollment Interactions with Africa dummy Ln(population) Ln(population density) Natural resources Ln(per capita income) Population ∗ GDP per capita Real GDP growth rate Inflation rate Current account balance/GDP KKM index

Private credit/GDP

–0.010 (0.040) 0.010 (0.029) –0.014 (0.051) –0.042 (0.054) –0.110 (1.373)

0.041 (0.044) 0.041 (0.034) 0.010 (0.067) 0.136 (0.112) –0.178 (1.513) 1.425 (1.592) 0.830* (0.452) –1.477 (0.963) –0.152 (0.100) (continued)

37

Resolving the African Financial Development Gap Table 1.6

(continued) Liquid liabilities/GDP (1)

(2)

Manufacturing/GDP Secondary/primary enrollment Constant Adjusted R2 Observations F(population density = 0 for Africa) Corresponding p-value F(GDP growth = 0 for Africa) Corresponding p-value F(inflation rate = 0 for Africa) Corresponding p-value F(KKM index = 0 for Africa) Corresponding p-value F(CA balance / GDP = 0 for Africa) Corresponding p-value F(natural resources = 0 for Africa) Corresponding p-value

Private credit/GDP (3)

–0.776 (0.891) 0.418 (0.418) 0.990** (0.152)

0.525*** (0.073)

0.467*** (0.091)

0.33 112

0.31 112 6.74

0.45 99 4.62

0.011

0.035 0.10

(4)

(5)

(6)

0.294*** (0.061)

0.248*** (0.076)

–0.916 (0.735) 0.148 (0.345) 0.640*** (0.125)

0.37 112

0.34 112 3.07

0.51 99 4.27

0.083

0.042 0.07

0.758 0.04

0.798 0.07

0.848 1.50

0.797 0.52

0.224 0.04

0.473 0.09

0.837 0.02

0.771 0.06

0.893

0.802

Notes: This table presents OLS regressions of banking sector development, measured by liquid liabilities/ GDP and credit to the private sector from deposit money banks/GDP ratios, on a set of country-level variables including endowment (population and resources), macroeconomics, institutions, banking structure, and other variables. We include all the low- and middle-income countries (including Africa, excluding South Africa), an African dummy variable, and interactions between each of our explanatory variables and the Africa dummy. Standard errors are presented in the brackets below coefficients, and F-statistics are shown at the bottom of the table. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

indicators of financial development (columns [3] and [6]). The coefficient for the Africa * growth interaction is positive and insignificant, but we cannot reject the hypothesis that the sum of the two coefficients (growth and Africa * growth) is zero (again see F-statistics near the bottom of table 1.6). This indicates again that the conditional convergence result does not hold in Africa, as was suggested by our simple comparison of the coefficients from the base model (table 1.4) with those from the African models (table 1.5).

38

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Similar results hold for the inflation variable that remains negative and highly significant when the interactions are included in the regressions. The interaction between inflation and the Africa dummy is positive, marginally significant, and of a magnitude similar to that of the simple inflation variable. We cannot, therefore, reject the hypothesis that the sum of the inflation and Africa * inflation variables is zero, which indicates that, like real growth, inflation also has not been associated with less financial development in Africa. A similar pattern also holds for the current account balance and KKM index of institutional development variables. Both sets of coefficients are positive and significant in both specifications (columns [3] and [6]), but their signs are negative when interacted with the Africa dummy variable. We cannot reject the hypothesis that the sum of the two coefficients is equal to zero for both variables; that is, neither the current account balance nor KKM index is positively related to African banking development. The regressions that include interaction terms therefore indicate that the high predicted levels of financial development for Africa from our base models derive, in part, from macroeconomic factors that matter only outside Africa. Toward Resolving the African Finance Puzzle Are the determinants of financial development different in Africa? The more refined analysis in table 1.6 indicates that the determinants of banking development in Africa differ from other developing countries in significant ways. Of particular interest is the set of factors that matter outside Africa but not in Africa, and those that matter in Africa but not in the rest of the world. While macroeconomic variables and the KKM index of institutional development appear to be important determinants of banking development in the rest of the developing world, they lose power in the context of Africa. We also observe that the notion of a natural resource curse, which is largely popularized in the context of resource- rich countries such as those in Africa, is insignificant in explaining the African financial development gap. The one explanatory variable that stands out for Africa is population density. Our results show a steeper relationship between population density and banking indicators for Africa than the rest of the developing world. In general, the differences between the predicted and actual values of financial development for Africa become smaller when the interaction terms are included in the regressions. In fact, the average difference for the models with full interactions is very close to zero. No difference is greater than 0.06 or less than – 0.06 for either indicator of banking development. The majority of the predicted values lie within 2 percentage points of their actual levels. The simple Africa dummy, which is really a measure of our ignorance about what drives African financial development, tends not to be significant (and never achieves significance at better than the 10 percent level). Our findings suggest that, if mechanisms for holding down inflation, gen-

Resolving the African Financial Development Gap

39

erating a current account surplus, and conditional convergence were to work in Africa as elsewhere, the levels of financial development would be higher. At the same time, it remains an open question as to why those mechanisms do not seem to work in Africa. Perhaps a period of sustained economic growth will ignite those mechanisms, but the African puzzle continues in a different form.19 Additional Tests and Robustness Checks We now perform additional tests and robustness checks. In particular, we analyze in further depth the role of population density, which shows up more strongly in Africa than elsewhere. Moreover, we consider indicators of stock market development to test whether a gap similar to the one found for banking development indicators arises in the context of stock markets. Although stock markets still remain small and illiquid, African countries seem to have committed to the equity sector in promoting financial development, and several stock exchanges have emerged in Africa in the recent past. 1. Population Density. We try to better understand why sparsely populated African countries have low levels of banking development by, first, examining the “shape” of the relationship between density and financial development based on our regression models and, second, including additional variables related to population density in the regressions. Figure 1.3 shows the relationship between population density and liquid liabilities/GDP for the typical African and non-African developing country based on table 1.6. Specifically, the vertical axis plots predicted values from model 2 in table 1.6. For all variables other than population density, we use the mean value for the subsample (either Africa or other developing countries) multiplied by the appropriate coefficient for that subsample. For population density, we run through the range of possible values and multiply by the appropriate coefficient for each subsample to generate the curvature. Hence, nonpopulation density factors (partially) account for the distance between the curves for Africa and that for other developing countries. Three features of the graph for liquid liabilities in figure 1.3 are worth noting: (a) the largest differences between Africa and the rest of the developing world come at the lower end of the density scale; (b) population density affects financial development in all countries, but the relationship is much steeper for Africa; and (c) although African countries tend to have lower population densities, there is substantial overlap for the two samples, which 19. Another possibility is that variables that are important for African financial development are omitted from our models. For example, Easterly and Levine (1997) demonstrate that ethnic fractionalization explains a large share of factors that are linked to (slow) growth within Africa. We experimented with two types of variables: fractionalization and armed conflict. We use the fractionalization data as in Alesina et al. (2003), and we use the UCDP/PRIO data set to calculate the average number of armed conflicts per year for each country. Neither variable is significant in our banking development regressions, and their inclusion does not alter the main qualitative results of our models.

40

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Fig. 1.3

Relationship between population density and liquid liabilities/GDP

Notes: Plots are for the typical African and non-African developing country based on table 1.5. Specifically, the vertical axis plots predicted values from model 2 in table 1.5. For all variables other than population density, we use the mean value for the subsample (either Africa or other developing countries) multiplied by the appropriate coefficient for that subsample. For population density, we run through the range of possible values and multiply by the appropriate coefficient for each subsample to generate the curvature.

suggests that comparisons across the full range of the density spectrum are valid. The picture points to important differences between Africa and the rest of the world, which could be due to the fact that the minimum viable scale in banking is greater in Africa than elsewhere or to other institutional or infrastructural failures. Regardless of the explanation, figure 1.3 highlights Africa’s difficulty in overcoming problems related to scale for the least densely populated countries. Similarly to figure 1.3, figure 1.4 shows the relationship between population density and private credit/GDP based on the coefficients of model 5 in table 1.6. As in the figure for liquid liabilities/GDP, the gap between Africa and the rest of the developing world is largest at the lowest levels of population density. However, unlike for the liquid liabilities measure, that gap does not dramatically close at intermediate density levels. This suggests that while differences in minimum viable scale could explain differences between Africa and the rest of the developing world with respect to savings, they cannot account for Africa’s lower levels of private credit over the full range of population densities.

Resolving the African Financial Development Gap

Fig. 1.4

41

Relationship between population density and private credit/GDP

Notes: The vertical axis plots predicted values from model 5 in table 1.5. For all variables other than population density, we use the mean value for the subsample (either Africa or other developing countries) multiplied by the appropriate coefficient for that subsample. For population density, we run through the range of possible values and multiply by the appropriate coefficient for each subsample to generate the curvature.

To understand the aspects of population density that contribute to low levels of banking development in Africa, we include additional variables in our regressions that summarize (a) spatial population characteristics, such as large population agglomerations; (b) infrastructural and transportation development, and (c) physical banking presence (i.e., branch- penetration measures). The results of these new regressions can help understand what policies may be effective in fostering African development. If population agglomeration and infrastructural development can account for most of the strong relationship between population density and banking underdevelopment in Africa, provision of financial services might improve as a natural consequence of broader economic/infrastructural improvements or policies that specifically encourage economic participation by those that live outside urban agglomerates. On the other hand, if measures of bank- branch penetration are better able to account for the density/banking development relationship in Africa, then it is more likely that specific banking market failures, perhaps related to relatively high minimum viable scale, are at the heart of African underperformance.

42

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

For spatial population characteristics, we include the percentage of the population residing in the largest city, the percent of the population in urban agglomerates, and a measure of population concentration with zero indicating a uniform distribution of residents across 20 km by 20 km cells and one indicating that the entire population is concentrated in a single 20 km by 20 km cell. For infrastructural development we include roads per square km and railroad track per square km. For bank penetration, we include commercial bank branches per 100,000 residents (demographic penetration) and branches per square km (geographic penetration). Inclusion of many of these variables reduces sample size substantially and their coefficients are insignificant. The results for the variables that do produce significant results (roads per square km, percentage of population in the largest city, and the two branch penetration measures) are presented in table 1.7. The population density variable is no longer significant when any of these variables are introduced to the liquid liabilities regressions (in fact, it becomes negative and significant when the geographic branch penetration variable is included), suggesting that roads, population in major cities, and bank- branch penetration are all relevant factors in understanding the strong link between population density and financial development in Africa. That said, only the coefficients for the percentage of population in the largest city and the geographic bank- branch penetration measure are consistently significant across the models in table 1.7. This pattern is consistent with the notion that physical proximity to providers of financial services accounts for much of the variation in African banking development and thus the minimum viable scale is best achieved in major cities. It is also worth noting that the links to financial development for the geographic penetration and largest city population variables are significant for liquid liabilities to GDP but not for private credit to GDP. Again, overcoming difficulties associated with population density appears more effective on the savings side than on the credit side. 2. Stock Market Development. We replace our indicators of banking development with indicators of stock market development, namely stock market capitalization and the value of traded shares, each divided by GDP, and rerun the models in table 1.6. These are yearly observations that we average over the sample period for each country as described above. We are unable to run our base models from table 1.7 on the sample of African countries because so few of them have indicators of stock market development. The results (not reported) are notable only because none of the variables that are associated with stock market development in the rest of the developing world are important in Africa. That is, we cannot reject the hypothesis that the relationship between any of the variables in our models and our indicators of African stock market development is zero. This is very likely attributable to the early stage of development of stock markets in Africa.

Observations Adjusted R2

Demographic branch penetration

Geographic branch penetration

Percent population in largest city

Roads/area

91 0.46

0.096*** [12.083]

(1)

71 0.58

0.050 [1.547] 0.289 [1.387]

(2)

63 0.54

1.141** [6.326]

0.013 [0.2910]

(3)

73 0.40

0.256*** [180.65]

–0.040*** [62.300]

(4)

73 0.39

73 0.42

(9)

73 0.41 (continued)

[2.245]

0.215*** [29.084]

(8)

[4.264] 63 0.56

1.222** [6.455]

(7)

–0.009

71 0.54

0.512*** [12.633]

(6)

–0.013**

–0.035 [2.438]

(5)

Dependent variable: Liquid liabilities/GDP

Factors that could explain the effect of population density on financial development (combined sample of low- and middle-income and African countries, excluding South Africa)

Ln(population density)

Table 1.7

(continued)

99 0.45

0.078** [4.62]

(1)

75 0.49

0.051 [0.40] 0.171 [0.18]

(2)

63 0.55

1.044 [0.80]

0.015 [0.06]

(3)

76 0.42

0.248 [0.41]

–0.050 [0.23]

(4)

76 0.42

76 0.42

(9)

76 0.41

[014]

0.183 [0.26]

(8)

[0.17] 63 0.57

1.090 [0.94]

(7)

–0.011

75 0.45

0.399** [4.32]

(6)

–0.012

–0.032 [0.10]

(5)

Dependent variable: Private credit/GDP

Notes: All models estimated via OLS with White’s heteroskedasticity-consistent standard errors. We use the same model specifications as those in table 1.5, columns (3) and (6), without the African dummy variable and the interactions between each of our explanatory variables and the African dummy. The Fstatistics are presented in square brackets. Area is defined in terms of square miles. Geographic branch penetration is the number of commercial bank branches per square km as reported in CGAP (2009). Demographic branch penetration is commercial bank branches per 100,000 residents. In unreported specifications, we also included percent of population in urban agglomerates, railroad track distance per square km, percent arable land, and a measure of the geographic concentration of population with 0 indicating a uniform distribution and 1 indicating that the entire population is concentrated in a 20 km x 20 km cell (as used in Collier and Hoeffler 2004). Inclusion of those variables substantially reduced sample size, and none produced consistently significant results for both indicators of financial development, and when population density was excluded from the specification. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Observations Adjusted R2

Demographic branch penetration

Geographic branch penetration

Percent population in largest city

Roads/area

Ln(population density)

Table 1.7

Resolving the African Financial Development Gap

45

Once these stock markets are better established, it might be easier to analyze the determinants of their development via cross- country regressions.20 1.4

Firm-Level Tests on the Access to External Finance

As mentioned earlier, while cross- country studies focusing on countrylevel variables are highly useful for summarizing patterns in the data and identifying relationships between variables, they are subject to methodological issues. For example, the observed results may be driven by measurement errors or unobservable factors. In an attempt to overcome these potential problems, we next turn to firm- level data drawn from the Investment Climate Surveys (ICS) conducted by the World Bank over the past decade.21 Our specific reasons for augmenting the cross- country study with the firmlevel analysis are as follows: First, while country- level indicators of banking development might suffer from measurement errors, firm- level responses about their use of financial services should not. Second, our cross- country level regressions involve one observation per country and thus have fewer degrees of freedom than what are available in firm- level regressions. Third, the firm- level data sets allow us to control for a number of potentially relevant firm characteristics (e.g., size, sector) that could explain substantial variation in the demand for, and use of, financial services. Thus, overall the firm- level regressions can provide more precise estimates of the relationships between financial development and our explanatory variables than the country- level regressions. Furthermore, to the extent that we derive similar qualitative results for the firm- and country- level regression, it provides additional, and even stronger, support for the conclusions that we have drawn to this point based only on the country- level regressions. Table 1.8 presents summary statistics on our firm- level sample. Firm characteristics (all dummy variables) include size (small, medium, and large based on the number of employees), industry (manufacturing, services, and others), exporter status and foreign- ownership status. It appears that there are more small firms and nonexporting firms in the African (excluding South Africa) sample than in other developing countries, while the fraction of firms with foreign- ownership stakes is higher in the African sample. Firmfinancing sources, for both short- term (working capital) and long- term (new investments) purposes, include internal (e.g., retained earnings) and external channels. Following prior research, we focus on long- term external financ20. Moreover, Bekaert, Harvey, and Lundblad (2005) find that stock market liberalization, leading to free flows of capital in and out of a home country’s equity markets, has a causal and positive impact on economic growth. But for most African countries, the stock market development (and associated institutions) has not reached the stage of liberalization. 21. These surveys have been used frequently in recent cross- country, firm- level studies on law, institutions, finance and growth (e.g., Beck, Demirgüç-Kunt, and Levine 2005; Beck, DemirgüçKunt, and Maksimovic 2005).

46

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Table 1.8

Summary statistics: Firm sample based on ICS surveys N

Mean

Std. dev.

Fraction of 0

A. African countries (excluding South Africa) Firm size-small (0 or 1) 2,591 0.46 0.50 Firm size-medium (0 or 1) 2,591 0.30 0.46 Foreign (0 or 1) 2,591 0.25 0.43 Exporter (0 or 1) 2,591 0.18 0.38 Manufacturing (0 or 1) 2,591 0.70 0.46 Services (0 or 1) 2,591 0.17 0.37 Access to external finance (0 or 1) 2,591 0.16 0.37

— — — — — — 83.91%

B. Other low- and middle-income countries (excluding Africa) Firm size-small (0 or 1) 30,525 0.34 0.47 Firm size-medium (0 or 1) 30,525 0.33 0.47 Foreign (0 or 1) 30,525 0.13 0.33 Exporter (0 or 1) 30,525 0.24 0.43 Manufacturing (0 or 1) 30,525 0.66 0.47 Services (0 or 1) 30,525 0.27 0.44 Access to external finance (0 or 1) 30,525 0.26 0.44

— — — — — — 73.75%

Notes: This table presents summary statistics on our firm-level sample (used in regressions in table 1.11), based on World Bank’s Investment Climate Surveys over the past decade. The surveys were conducted over the period 2002–2006, with the majority of the firms surveyed once. Firm characteristics (all dummy variables) include size (small, medium, and large based on the number of employees), industry (manufacturing, services, and others), exporter status and foreign-ownership status. The external finance dummy equals one if a firm has access to loans from domestic or foreign banks or credit card debts from these institutions, and zero otherwise; we only include firm observations for which data is available on whether the firm has access to all three external financing channels.

ing from financial institutions, the most important external financing channel in most developing countries. Specifically, the external finance dummy equals one if a firm has access to loans from domestic or foreign banks or credit card debts from these institutions, and zero otherwise. Almost 84 percent (74 percent) of firms in Africa (other developing countries) do not have access to external finance under this definition.22 Our basic strategy is to pattern our firm- level regressions on the countrylevel regressions, but also to include variables that summarize firm characteristics. We note, however, that we have all of the necessary variables for firms from only twelve countries in Africa. Since the standard errors in our models allow for clustering at the country level, we are not able to include all of the country- level regressors that were in the cross- country regressions 22. We only include firm observations for which data is available on whether the firm has access to all three external financing channels (domestic and foreign banks and credit cards). Summary statistics reported in table 1.8 represent the sample we use in regressions reported in table 1.9.

Resolving the African Financial Development Gap

47

in table 1.6. However, the most important regressors are included.23 Since the majority of firms in Africa and elsewhere do not have access to external finance (table 1.8), our dependent variable is a dummy indicating whether the firm received any financing from a formal provider, rather than a continuous variable showing the importance of institutional finance among all financing sources.24 Table 1.9 reports the marginal effects from probit regressions. In the baseline model, population density is once again positive and significantly associated with firms’ access to external finance (column [1]). The coefficient on the interaction between the Africa dummy and population density indicates that there is no difference in the impact of this variable for African and non-African firms. Moreover, we can reject that the sum of the coefficients for density and its interaction with the Africa dummy is zero at the p = 0.05 level for this model (see chi- squared statistics near the bottom of table 1.9). In model 2 we add road coverage (per square km), and find it is actually negatively associated with African firms’ access to external finance while it has a positive impact for firms from other developing nations. On the other hand, adding this variable strengthens the (marginally positive) impact of population density for African firms as the interaction with the African dummy is significant at 10 percent and the sum of the density and Africa * density coefficients is positive and significantly different from zero at the 5 percent level (column [2]). Another main result from table 1.9 is that both the geographic and demographic bank- branch penetration variables have a positive association with access to external finance for African countries. However, the result is significant only for geographic penetration (at 1 percent in column [4]). These variables are insignificant and of smaller magnitudes in the rest of the developing world. Population density remains positive and significant for non-African countries when the variables for demographic and geographic bank- branch penetration enter the regression (columns [4] and [5]). We can no longer reject that the sum of the coefficients for population density and its interaction with the Africa dummy is zero in the model with geographic branch penetration (though we continue to do so when demographic 23. We performed a series of robustness checks in which we added each of the country- level variables from the country- level regressions in table 1.6 that do not appear in the firm- level regressions in table 1.9, one at a time. Qualitative results are very similar to those reported in table 1.9. 24. For the surveys completed through 2006, the dummy variable is equal to one if the firm reported a positive value for finance from local banks, foreign banks, or credit cards. For the surveys after 2006, those questions were broadened to ask whether firms received finance from private or state- owned commercial banks, or from nonbank financial institutions. Since banks are the main providers of financial services throughout the developing world, this slight change in question format does not present a major problem. Regression results are very similar when we include a post- 2006 dummy variable.

Table 1.9

Regressions for the sample of low- and middle-income and African countries: Firm-level results

  Africa dummy Small Medium Foreign Exporter Manufacturing Services KKM index Ln(per capita income) Natural resources Age of the firm Ln(population density)

(1)

(2)

(3)

(4)

(5)

0.011 [0.104] –0.151*** [0.023] –0.066*** [0.011] –0.046*** [0.013] 0.053*** [0.011] 0.082*** [0.030] –0.029 [0.023] 0.067* [0.035] –0.007 [0.032] 0.015 [0.023] –0.001** [0.000] 0.054*** [0.019]

0.449** [0.189] –0.144*** [0.015] –0.062*** [0.012] –0.042*** [0.014] 0.037*** [0.011] 0.048** [0.024] –0.033 [0.024] 0.032 [0.028] –0.029 [0.032] 0.019 [0.018] –0.001** [0.000] 0.021 [0.026] 0.060* [0.034]

0.230 [0.169] –0.140*** [0.018] –0.059*** [0.011] –0.048*** [0.013] 0.059*** [0.011] 0.068*** [0.025] –0.038 [0.023] 0.092* [0.048] 0.006 [0.033] 0.001 [0.029] –0.001** [0.000] 0.045** [0.018]

–0.061 [0.111] –0.142*** [0.026] –0.058*** [0.012] –0.046*** [0.013] 0.053*** [0.010] 0.091*** [0.030] –0.035 [0.023] 0.055 [0.037] –0.001 [0.032] 0.018 [0.022] –0.001*** [0.000] 0.060** [0.026]

0.093 [0.147] –0.142*** [0.024] –0.058*** [0.012] –0.046*** [0.013] 0.053*** [0.011] 0.091*** [0.029] –0.035 [0.023] 0.056 [0.037] –0.000 [0.035] 0.018 [0.022] –0.001*** [0.000] 0.059*** [0.020]

Roads/area Foreign-ownership share

–0.197* [0.119]

Number branches per 1,000 km2

–0.004 [0.034]

Number branches per 100 km adults

–0.000 [0.001] –0.047 –0.022 [0.037] [0.041] –0.028 0.000 [0.029] [0.033] 0.005 0.000 [0.041] [0.040] –0.037 –0.012 [0.033] [0.028] –0.097** –0.103** [0.045] [0.042] –0.157*** –0.180*** [0.033] [0.027] –0.003 0.014 [0.089] [0.090] (continued)

Small ∗ Africa Medium ∗ Africa Foreign ∗ Africa Exporter ∗ Africa Manufacturing ∗ Africa Services ∗ Africa KKM index ∗ Africa

0.020 [0.042] 0.069 [0.043] 0.013 [0.032] 0.019 [0.031] –0.110*** [0.034] –0.184*** [0.021] 0.035 [0.079]

–0.014 [0.037] 0.022 [0.040] 0.017 [0.035] 0.034 [0.029] –0.076* [0.041] –0.148*** [0.032] 0.108 [0.080]

0.008 [0.045] 0.065 [0.043] 0.049 [0.035] 0.002 [0.031] –0.118*** [0.033] –0.179*** [0.032] 0.072 [0.085]

Table 1.9

(continued)

  Ln(per capita income) ∗ Africa Natural resources ∗ Africa Age of the firm ∗ Africa Ln(population density) ∗ Africa Roads/area ∗ Africa

(1) –0.045 [0.059] –0.035 [0.484] 0.002*** [0.001] –0.000 [0.031]

Foreign-ownership share ∗ Africa

(2) –0.002 [0.073] –0.451 [0.453] 0.001 [0.001] 0.086* [0.051] –0.483** [0.222]

test ln_pd + ln_pd ∗ africa chi2 p-value ln_pd test roads_area + roads_area_africa = 0 chi2 p-value roads test fos + fos_africa = 0 chi2 p-value fos test geobrpen + geobrpen_africa = 0 chi2 p-value geobrpen test demobrpen + demobrpen_africa = 0 chi2 p-value demobrpen

–0.049 [0.060] 0.279 [0.614] 0.002** [0.001] –0.014 [0.027]

(4)

(5)

–0.043 [0.084] 1.439*** [0.476] 0.002** [0.001] –0.018 [0.045]

0.012 [0.074] 0.725 [0.909] 0.002*** [0.001] 0.030 [0.039]

–0.260 [0.170]

Num. branches per 1,000 km2 ∗ Africa Num. branches per 100 km adults ∗ Africa Observations Adjusted R-square

(3)

0.340*** [0.100] 0.004 [0.013] 35,912 0.0874

30,054 0.0785

34,622 0.0904

32,276 0.0921

4.574 0.0325

5.909 0.0151

2.396 0.122

1.285 0.257

32,276 0.0910 7.006 0.00812

3.741 0.0531 13.51 0.000237 13.54 0.000233 0.0866 0.769

Notes: This table presents marginal effects from probit regressions with the dependent variable equal to 1 if a firm has access to loans from domestic or foreign banks or credit card debts from these institutions, and zero otherwise. The surveys were conducted over the period 2002–2006, with the majority of the firms surveyed once. The explanatory variables include firm characteristics and a subset of country-level variables used in table 1.7. We include firms from all the low- and middle-income countries (including Africa, excluding South Africa), an African dummy variable, and interactions between each of our explanatory variables and the Africa dummy. White’s heteroskedasticity-consistent standard errors are clustered at the country level, and presented in the brackets below coefficients. The F-statistics are shown at the bottom of the table. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

50

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

penetration enters the regression in model 5). The results for geographic branch penetration thus confirm those from the country- level regressions. Thus, both cross- country and firm- level regressions point to the importance of physical proximity to providers of financial services in Africa relative to the rest of the developing world. General infrastructural failures are not necessarily responsible for low levels of access to external finance for African firms, but rather the failures seem endemic to the banking sectors and markets for debt instruments. In table 1.9 we also introduce foreign- bank ownership (at the country level) as an explanatory variable. We did not include this variable in the country- level regressions (tables 1.4– 1.6) because we view this as an “outcome” variable for the development status of a country’s banking sector. However, this variable is plausibly exogenous from the point of view of small, privately owned firms in a given region. This variable is negatively associated with access to external finance for firms from the developing world (column [3]), and for Africa (see F-statistic for foreign ownership plus its interaction with the Africa dummy variable at the bottom of column [3]). We cannot reject that the sum of the coefficients for population density and its interaction with the Africa dummy is zero in this model, suggesting some interplay between foreign- bank presence in Africa and population density.25 Finally, we note that firm characteristics explain a substantial amount of variation in firms’ access to external finance, supporting the notion that firm- level regressions offer more precise estimates than the country- level regressions. The similarities and differences between Africa and the rest of the developing world are also instructive. Firm size affects the use of external finance similarly in Africa and elsewhere, with smaller firms being at a substantial disadvantage relative to larger ones. Foreign- owned firms use less external finance than local firms in Africa and elsewhere, perhaps because they have access to nonlocal sources of finance such as their parent companies. Differences also arise with respect to industry affiliation. In the rest of the developing world, manufacturing firms rely more heavily on external finance than other firms. In Africa, those firms show no strong tendency to rely on external finance more than others. In addition, older firms are less likely to have access to external finance outside Africa; more mature African firms appear to have an advantage in acquiring such finance. We conclude our cross- country and firm- level analyses with some observations. The cross- country approach allows us to measure the financial development gap between Africa and other developing economies and identify factors that are of particular relevance for Africa. However, we recognize the limitations with cross- country studies, and hence attempt to address them in three ways. First, we are cautious in drawing causation; we 25. The density variable remains negative and significant for the rest of the developing world when foreign bank- ownership shares are included in the regression.

Resolving the African Financial Development Gap

51

do, however, view variables, such as population density and natural resources endowment, as exogenous variables for predicting financial- sector development. Second, as noted above we complement our country- level tests with firm- level analyses, and the microlevel evidence is consistent with the country- level results. In particular, both the cross- country and firm- level studies are consistent with the importance of overcoming low population density to achieve financial development in Africa. Finally, we employ new data on the outreach of financial sectors, in particular bank- branch penetration measures, to better understand the channels through which population density may affect different aspects of financial- sector development. Moreover, we explore whether transportation infrastructure, urban agglomeration, and the share of arable land are driving the strong relationship between population density and banking- sector development in Africa. 1.5

Within-Country Study: Equity Bank and Financial Access in Kenya

Our cross- country and firm- level analyses both indicate that sparsely populated regions pose a particularly severe problem for African financial development. What types of financial institutions are best suited to providing services in such areas? A satisfactory answer to this question is well beyond the scope of the current chapter, but in this section we examine the effects of the recent expansion of bank- branch networks in Kenya to get some sense of the factors that drove a large increase in the use of bank accounts and loans within a relatively short time period. As background, we begin by noting that while economic growth over the past few decades has been disappointing in Africa, some countries began growing fast at the start of the twenty- first century, at least until the negative shock of the global crisis in 2007– 2009. The GDP growth for the fiveyear period running from 2002 to 2007 averaged over 5 percent in Africa, outpacing population growth and hence implying an increase in GDP per capita (Allen, Otchere, and Senbet 2011). However, and as discussed earlier, despite the adoption of extensive economic and financial reforms, most sub-Saharan African countries still face a severe gap in the development of their financial sectors. Figure 1.5 shows the relationship between demographic bank penetration (measured by the number of bank branches per 100,000 people) and GDP per capita across low- income economies in 2003– 2004. The figure suggests three stylized facts. First, there is a strong positive relationship between bank penetration and economic development. Second, sub-Saharan African countries are characterized by both low income per capita and low bank- branch penetration. Third, the bank- branch network in sub-Saharan African countries is underdeveloped relative to not only the higher- income economies, but other peers with similar income levels. The underdevelopment of the bank- branch network in sub-Saharan

52

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Fig. 1.5

Demographic bank penetration and GDP per capita

Note: This figure shows the relationship between GDP per capita and the number of bank branches per 100,000 people across low-income economies in 2003/2004. The sample considers countries with a GDP per capita below USD 4,000. The gray dots correspond to subSaharan African countries. The black dots correspond to the rest of the countries. The data source is Beck, Demirgüç-Kunt, and Martinez Peria (2007).

Africa has resulted in low access to banking services, as illustrated by our firm- level study (on the bank credit side). While early in the twenty- first century roughly half of the world’s adult population lacked any form of bank account, in sub-Saharan Africa over 80 percent of the adult population lacked an account (Chaia et al. 2011; Honohan 2008). In line with these facts, Kendall, Mylenko, and Ponce (2010) report that the average number of bank accounts per adult is 3.7 in developed countries, 0.9 in developing countries, and less than 0.5 in sub-Saharan African countries. The financial access problem in many sub-Saharan countries and the emergence of new business models that may contribute to solve this problem deserve particular attention. In this respect, Kenya offers an interesting case study for financial access, as in recent years it has witnessed a strong bank- branch expansion. This expansion has coincided with the emergence of Equity Bank, a pioneering commercial bank that devised a bankingservice strategy targeting low- income clients and traditionally underserved territories. The bank is listed on the Nairobi Stock Exchange, it has no government- ownership share, and it has played a key role in integrating provision of financial services to the underprivileged with formal finance. Figure 1.6 shows the stock performance of Equity Bank during the period of August 2006 to August 2011. It significantly outperforms two of its main competitors on this measure, Barclays Bank of Kenya and the Kenya Commercial Bank.

Resolving the African Financial Development Gap

Fig. 1.6

53

Stock market performance

Note: This figure shows the stock market performance of Kenya Commercial Bank (KNCB), Equity Bank (EQBNK), and Barclays Bank of Kenya (BCBL) since August 18, 2006. The data source is an interactive chart from Bloomberg (http://www.bloomberg.com).

Table 1.10

Number of bank branches by type of district

District

2006

2009

Change

Change (%)

Urban Rural Arid and semiarid Total

290 238 42 570

471 398 89 958

181 160 47 388

62 67 112 68

Note: This table reports the number of bank branches by type of district: urban, rural and arid, and semiarid. This table was constructed based on alternative sources, including phone calls, official websites, banks’ annual reports, and government publications.

The banking system in Kenya expanded greatly in terms of the number of branches in recent years. Table 1.10 reports that between 2006 and 2009 the total number of bank branches in Kenya increased from almost 600 to almost 1,000 (a 68 percent increase). This bank- branch expansion occurred not only in urban districts, but also in rural and arid and semiarid districts. In percentage terms, this expansion was greater in the arid and semiarid districts (112 percent), followed by rural districts (67 percent), and urban districts (62 percent). This expansion involved all ownership catego-

54

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Table 1.11

Bank branches over time: Commercial banks in Kenya

Local private banks

2006

2009

Change

Change (%)

Equity Bank Co-operative Bank of Kenya Ltd. Diamond Trust Bank of Kenya Ltd. Commercial Bank of Africa Ltd. All local private banks Foreign banks Barclays Bank of Kenya Ltd. Standard Chartered Bank Kenya Ltd. K-Rep Bank Ltd. All foreign banks Government and government-influenced banks Kenya Commercial Bank Ltd. National Bank of Kenya Ltd. CFC Stanbic Bank Ltd. All government and government-influenced banks

44 52 8 19 247

112 83 32 20 450

68 31 24 1 203

155 60 300 5 82

62 31 22 150

119 35 30 258

57 4 8 108

92 13 36 72

117 33 16 179

169 43 35 262

52 10 19 83

44 30 119 46

Note: This table reports the number of Kenyan bank branches by bank name and ownership in 2006 and 2009. The table considers only banks with twenty or more bank branches by 2009. This table was constructed based on alternative sources, including phone calls, official websites, banks’ annual reports, and government publications.

ries of banks, namely domestic private banks, foreign banks, and government banks. However, as shown in table 1.11, the expansion was particularly driven by domestic private banks, among which Equity Bank played an important role. The number of Equity Bank’s branches increased from 44 to 112 in the period 2006– 2009, representing an expansion of 155 percent. No other bank experienced such growth in the same period. The number of bank branches of Barclays and Kenya Commercial Bank (i.e., the banks with the largest branch networks after Equity Bank) increased from 62 to 119 and from 117 to 169, representing growth of 92 percent and 44 percent, respectively. Although most Kenyan banks have experienced branch expansion in recent years, banks with different ownership structures have followed different expansion strategies. While foreign banks prefer urban areas (perhaps in order to cherry pick a set of elite customers), domestic private banks seem to exploit their superior knowledge of the culture and have also entered rural and arid and semiarid areas. For example, as part of its expansion strategy, Equity Bank focused on the use of local languages in its branches. This strategy is quite important considering that 30– 40 percent of the people in central Kenya cannot speak either English or Swahili, and thus speak only a minority language. As a first exploration of the bank- branching strategies in Kenya, figure 1.7 presents the number of bank branches in 2006 and 2009 for three bank

Bank penetration in vulnerable districts

Notes: This figure shows the number of bank branches according to different ownership structures: foreign banks, government banks, and domestic private banks. Rural, arid and semiarid districts exclude Nairobi and Mombasa. High minority-language districts are the ones where the share of the population speaking a minority language is larger than the median. Low-populated districts are the ones where the density population is smaller than the median. Low-educated districts are the ones where the share of the population with secondary or tertiary education is smaller than the median.

Fig. 1.7

56

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

groups: foreign banks, government- owned and government- influenced banks, and domestic private banks.26 The figure shows the number of bank branches in districts that have been traditionally less attractive for major commercial banks: rural and arid and semiarid districts, districts in which minority languages are more common, less populated districts, and districts with low educational- attainment levels.27 It suggests two stylized facts. First, domestic private banks and government and government- influenced banks had greater presence in underdeveloped/more vulnerable districts, while foreign banks had the least presence in those districts. Second, the bank- branch expansion of local private banks has been much stronger than the branch expansion of government banks and foreign banks. These patterns suggest that domestic private banks, whose expansion has been led by Equity Bank (see table 1.11), played a vital role in fostering banking services in underdeveloped districts. The performance of Equity Bank over the last half decade has been impressive. It has experienced an explosive growth of its assets and a significant expansion of its banking services. Figure 1.8 shows that during the period 2006– 2010 the assets of Equity Bank increased sevenfold, while its customers and customer deposits increased by a factor of six. In fact, Equity Bank ended the year 2010 with almost six million customers. Of these, 5.4 million are in Kenya, 474,000 are in Uganda, and 28,000 are in South Sudan. As a consequence of this expansion, Equity Bank became the fifth largest bank in Kenya in terms of the share of gross assets and deposits of the whole system and is by far the bank with the largest number of deposit and loan accounts. The number of deposit and loan accounts of Equity Bank represents around the 50 percent and 30 percent of the total number of deposit and loan accounts in Kenya, respectively (see table 1.12). These deposit and loan market shares suggest that the emergence of Equity Bank has played an important role in expanding the bank- branch network to underserved territories and thus expanding the use of banking services to population segments ignored by other large traditional commercial banks. Moreover, these stylized facts suggest that Equity Bank’s model has not only expanded the use of financial services to broader segments of the population, but it has also created profits in the process. As shown in figure 1.8, Equity Bank’s profits before taxes and shareholder funds have experienced a strong positive trend. 26. Government- owned banks are those in which the government owns a majority of shares. In (the two) government- influenced banks the government maintains a large, but noncontrolling ownership share. Throughout most of the chapter, we refer to both groups collectively as “government banks.” 27. Rural, arid, and semiarid districts exclude Nairobi and Mombasa. High minority language districts are those where the share of the population speaking a minority language is greater than the median. Less populated districts are those where population density is less than the median. Districts with low educational attainment levels are those where the share of the population with secondary or tertiary education is less than the median.

Resolving the African Financial Development Gap

Fig. 1.8

57

Equity Bank’s performance indicators

Note: This figure shows the evolution of the number of customers, deposits, gross loan portfolio, total assets, profits before taxes, and shareholders’ funds over the period 2006–2010. The data source is the Equity Bank’s Annual Report and Financial Statements (2010).

There is preliminary evidence that the bank- branch expansion and emergence of Equity Bank have paid off in terms of access to banking services in Kenya. In fact, as reported in table 1.13, the proportion of individuals having a bank account in Kenya increased considerably going from 14 percent in 2006 to 23 percent in 2009. The proportion of individuals having a loan from a bank showed a more modest increase from 2.9 percent to 4.3 percent. This increase in access to banking services has coincided with a stronger presence of Equity Bank. While Equity Bank had branches in 48 per-

58

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

Table 1.12

Market participation of major commercial banks in Kenya

Kenya Commercial Bank Ltd. Barclays Bank of Kenya Ltd. Co-operative Bank Standard Chartered Bank Ltd. Equity Bank Ltd. CfC Stanbic Bank Ltd.

Gross assets (%)

Total deposits (%)

Total capital (%)

Number of deposit accounts (%)

Number of loan accounts (%)

13.5 10.4 9.0 8.2 8.0 5.9

13.2 9.9 9.8 8.3 7.8 6.2

15.4 12.2 7.5 7.6 10.7 3.8

10.93 6.77 12.26 0.92 47.39 0.47

19.8 14.1 12.0 1.5 33.8 1.8

Note: This table reports market shares of gross assets, total deposits, total capital, number of deposit accounts, and number of loan accounts as of January 31, 2011. This table was constructed based on alternative sources, including official websites, banks’ annual reports, and government publications. Table 1.13

Use of bank services in Kenya Variable

2006 (%)

2009 (%)

Bank account Bank loan Equity Bank presence

14.20 2.90 48

22.70 4.40 87

Note: This table reports the percentage of people having a bank account and a loan from a bank in Kenya. It also reports the presence of Equity Bank across districts. This table was constructed based on FinAccess 2006 and 2009 surveys and alternative sources, including phone calls, official websites, banks’ annual reports, and government publications.

cent of the districts in Kenya in 2006, it had branches in 87 percent of them in 2009. In light of the preliminary evidence presented above, in an ongoing work Allen et al. (2011) explore the impact of Equity Bank, as well as other banks, on the use of banking services in Kenya by combining a new data set on bank presence with household- level survey data in 2006 and 2009. The data set on bank presence is based on branch- level information that is aggregated to construct a district- level panel on the number of branches by bank (see Allen et al. 2011 for details). The panel covers forty- five commercial banks that operated in sixty- five Kenyan districts in the period between 2006 and 2009.28 This new data set allowed the authors to explore the relationship between bank expansion and access to banking services over time and to 28. Although Kenya is currently divided into forty- six districts, the data set exploits a more disaggregated country division as of the 1999 census. In the 1999 census, Kenya comprised eight provinces (Central Coast, Eastern, Nairobi, North Eastern, Nyanza, Rift Valley, and Western) that were subdivided into sixty- nine districts. Of these sixty- nine districts, the survey data considered covers sixty- five.

Resolving the African Financial Development Gap

59

exploit within- district variation in bank presence. The household- level data is from the FinAccess surveys conducted by Financial Sector Deepening Trust Kenya (FSD Kenya) in 2006 and 2009. The 2006 survey consisted of 4,420 completed interviews, while the 2009 survey consisted of 6,598 completed interviews. Allen et al. (2011) find that the presence of domestic private banks, government banks, or foreign banks has a positive impact on access to bank accounts at the district level. Interestingly, they also find that the presence of Equity Bank in a specific district is strongly positively related to the residents’ probability of having a bank account and/or a bank loan, which goes beyond the effect of bank- branch expansion and presence of other commercial banks in Kenya. These results are consistent with the stylized facts that we discussed above, and they highlight the importance of institutions, such as Equity Bank, with a business model focused on the provision of financial services to population segments ignored by traditional commercial banks while generating sustainable profits in the process. Such institutions can be an important solution to the financial- access problem that has hindered the development of inclusive financial sectors in many developing countries. 1.6

Conclusions

The available evidence provides a convincing linkage between financial development and economic development. Yet the level of financial development remains very low in Africa based on the standard indicators of banking and capital market development. Benchmarking based on the correlates of financial development in other developing countries reveals a substantial gap between predicted and actual levels of African financial development. In addition, both country- level and firm- level tests indicate that the determinants of banking development in Africa differ from the rest of the world in telling ways. For example, inflation and the current account balance explain no variation in African financial development although they do so in other developing countries, and measures of institutional development explain less variation in Africa than elsewhere. Why this collection of macroeconomic fundamentals is not strongly linked to African financial development remains unclear. However, if these macroeconomic and institutional mechanisms could be made to function, our results suggest that the levels of financial development would rise. Most importantly, population density appears to be more important in Africa than elsewhere, and our firm- level study on the access to external financing indicates that population density is linked more to bank- branch penetration in Africa than in the rest of the developing economies. We also find a nonlinear relationship between population density and bankingsector development, with the largest gap between Africa and other developing countries observed for those African countries “trapped” in the low-

60

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

density area. Presumably, bank- branch penetration figures remain low in Africa because of difficulties in achieving minimum viable scale in sparsely populated, low- income areas. Therefore, technological advances, such as mobile banking, that enable users of financial services to be located far away from their financial institutions, provide a promising way to facilitate African financial development outside major cities, a topic that has been studied in the context of Kenya, where the mobile payments services of M-Pesa are now widely used (Mbiti and Weil 2011; Jack and Suri 2010). While use of mobile payments services has increased rapidly in Kenya, those services have not proven to be effective vehicles for savings accumulation nor provision of credit (Mbiti and Weil 2011). We present the stylized facts showing the effects of the expansion of bank- branch networks in Kenya from 2006 to 2009 and, in particular, that of Equity Bank. While banks of all ownership types (private, foreign, and government) expanded their networks during this period, Equity’s expansion was most pronounced and its association with take-up of bank accounts and loans greater than for other banks. Equity Bank was also more likely to enter previously underserved districts and the association between Equity’s presence and increased usage of bank accounts was larger for Kenyans with less wealth and education, that did not own a permanent home, and that lacked a salaried job. The more detailed analysis of the role of Equity Bank is a subject of an ongoing work by Allen et al. (2011), and the current findings support the observations that we make in this chapter. Overall, Equity Bank’s expansion experience suggests that it is possible to serve poorer segments of the African population while generating sustainable profits. At the same time, we recognize that the last chapter on this topic is yet to be written and that the current configuration of banking- services provision in Kenya presents its own challenges. For example, a single bank now maintains 55 percent of the deposit accounts in the banking system, and those deposits are covered by a deposit insurance fund, which could have implications for systemic stability. While we recognize that our contribution is closer to the first than the last word on the determinants of financial development and inclusion in Africa, our hope is that by combining country-, firm-, and household- level analyses, we have been able to shed some light on the contours of the current financial development gap and financial access issue in Africa.

References Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. “Fractionalization.” Journal of Economic Growth 8:155– 94. Allen, Franklin, Elena Carletti, Robert Cull, Jun Qian, Lemma Senbet, and Patricio

Resolving the African Financial Development Gap

61

Valenzuela. 2011. “Improving Access to Banking: Evidence from Kenya.” Working Paper, University of Pennsylvania. Allen, Franklin, Isaac Otchere, and Lemma Senbet. 2011. “African Financial Systems: A Review.” Review of Development Finance 1:79– 113. Beck, Thorsten, and Asli Demirgüç-Kunt. 2009. “Financial Institutions and Markets across Countries and over Time: Data and Analysis.” Policy Research Working Paper no. 4943, Washington, DC, World Bank. Beck, Thorsten, Asli Demirgüç-Kunt, and Ross Levine. 2000. “A New Database on the Structure and Development of the Financial Sector.” World Bank Economic Review 14 (3): 597– 605. ———. 2005. “SMEs, Growth and Poverty: Cross-Country Evidence.” Journal of Economic Growth 10:199– 229. ———. 2007. “Finance, Inequality, and the Poor.” Journal of Economic Growth 12 (1): 27– 49. Beck, Thorsten, Asli Demirgüç-Kunt, and Vojislav Maksimovic. 2005. “Financial and Legal Constraints to Firm Growth: Does Size Matter?” Journal of Finance 60:137– 77. Beck, Thorsten, Asli Demirgüç-Kunt, and Maria Soledad Martinez Peria. 2007. “Reaching Out: Access to and Use of Banking Services across Countries.” Journal of Financial Economics 85:234– 66. Beck, Thorsten, Erik Feyen, Alain Ize, and Florencia Moizeszowicz. 2008. “Benchmarking Financial Development.” Policy Research Working Paper no. 4638, Washington, DC, World Bank. Bekaert, Geert, Campbell Harvey, and Christian Lundblad. 2005. “Does Financial Liberalization Spur Growth?” Journal of Financial Economics 77:3– 55. Chaia, Alberto, Aparna Dalal, Tony Goland, Maria Jose Gonzalez, Jonathan Morduch, and Robert Schiff. 2011. “Half the World is Unbanked.” Working Paper, New York University. Clarke, George R. G., L. Colin Xu, and Heng-Fu Zou. 2006. “Finance and Income Inequality: What Do the Data Tell Us?” Southern Economic Journal 72 (3): 578– 96. Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers 56 (4): 563– 95. Consultative Group to Assist the Poor (CGAP). 2009. “Financial Access 2009: Measuring Access to Financial Services in 135 Countries.” Washington, DC, World Bank. Cull, Robert, and Laurie Effron. 2008. “World Bank Lending and Financial Sector Development.” World Bank Economic Review 22 (2): 315– 43. Cull, Robert, Lemma W. Senbet, and Marco Sorge. 2005. “Deposit Insurance and Financial Development.” Journal of Money, Credit, and Banking 37 (1): 43– 82. Easterly, William, and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.” Quarterly Journal of Economics 112 (4): 1203– 50. Equity Bank. 2010. Equity Bank’s Annual Report and Financial Statements 2010. http://www.equitybank.co.ke/investors.php?subcat=6. Fan, Joseph P. H., Randall Morck, Lixin Colin Xu, and Bernard Yeung. Forthcoming. “Does ‘Good Government’ Draw Foreign Capital? Explaining China’s Exceptional Foreign Direct Investment Flow.” World Development. Honohan, Patrick. 2008. “Cross-Country Variation in Household Access to Financial Services.” Journal of Banking and Finance 32:2493– 500. Honohan, Patrick, and Thorsten Beck, eds. 2007. Making Finance Work for Africa. Washington, DC: World Bank. International Monetary Fund (IMF). 2007. “Financial Soundness Indicators: Expe-

62

Allen, Carletti, Cull, Qian, Senbet, and Valenzuela

rience with the Coordinated Compilation Exercise and Next Steps.” Statistics Department, IMF. http://www.imf.org/external/np/pp/2007/eng/101807a.pdf. Jack, William, and Tavneet Suri. 2010. “The Economics of M-PESA.” Working Paper, MIT Sloan. Kamil, Herman, and Kulwant Rai. 2010. “The Global Credit Crunch and Foreign Banks’ Lending to Emerging Markets: Why Did Latin America Fare Better?” IMF Working Paper no. 10/102, International Monetary Fund. Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2007. “Governance Matters VI: Governance Indicators of 1996– 2006.” Policy Research Working Paper no. 4280, Washington, DC, World Bank. Kendall, Jake, Nataliya Mylenko, and Alejandro Ponce. 2010. “Measuring Financial Access around the World.” Policy Research Working Paper no. 5253, Washington, DC, World Bank. Lederman, Daniel, and William F. Maloney. 2008. “In Search of the Missing Resource Curse.” Economía 9 (1): 1– 58. Levine, Ross. 2005. “Finance and Growth: Theory and Evidence.” In Handbook of Economic Growth, edited by Philippe Aghion and Steven Durlauf. Amsterdam: Elsevier Science. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. “Financial Intermediation and Growth: Causality and Causes.” Journal of Monetary Economics 46 (1): 31– 77. Levine, Ross, and David Renelt. 1992. “A Sensitivity Analysis of Cross-Country Growth Regressions.” American Economic Review 82 (4): 942– 63. Loayza N., and R. Ranciere. 2006. “Financial Development, Financial Fragility, and Growth.” Journal of Money, Credit, and Banking 38:1051– 76. Mbiti, Isaac, and David Weil. 2011. “Mobile Banking: The Impact of M-Pesa in Kenya.” NBER Working Paper no. 17129, Cambridge, MA. Nissanke, Machiko, and Ernest Aryeetey. 1998. Financial Integration and Development in Sub-Saharan Africa. New York: Routledge. Pritchett, Lant. 1997. “Divergence, Big Time.” Journal of Economic Perspectives 11 (3): 3– 17. Rajan, Raghuram G., and Luigi Zingales. 1998. “Financial Dependence and Growth.” American Economic Review 88 (3): 559– 86. Sachs, Jeffrey D., and Andrew Warner. 1995. “Natural Resource Abundance and Economic Growth.” NBER Working Paper no. 5398, Cambridge, MA. ———. 2001. “The Curse of Natural Resources.” European Economic Review 45:827– 38. Senbet, Lemma, and Isaac Otchere. 2006. “Financial Sector Reforms in Africa: Perspectives on Issues and Policies.” In Annual Bank Conference on Development Economics, edited by Francois Bourgignon and Boris Pleskovic, 81– 119. Washington, DC: World Bank. World Bank. 2008. Database of Bank Regulation and Supervision (2008). Washington, DC, World Bank.

2

Challenges in Banking the Rural Poor Evidence from Kenya’s Western Province Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

2.1

Introduction

Access to basic banking services in sub-Saharan Africa remains limited, and lags far behind even other parts of the developing world. Chaia et al. (2009) combine a number of data sources to estimate that only about 20 percent of households in sub-Saharan Africa were banked early in the twentyfirst century.1 While there has been some progress in recent years, Kendall, Mylenko, and Ponce (2010) obtain similar results using more recent data. While developing countries have only 28 percent as many bank accounts per adult as do developed countries, the figure in sub-Saharan Africa is far lower (only 16 percent). Lack of access is particularly acute in rural areas: representative household survey data we collected between 2009 and 2011 suggest that only between 15 and 21 percent of households are banked in rural areas of Kenya, Malawi, and Uganda, respectively.2 Pascaline Dupas is associate professor of economics at Stanford University and a research associate of the National Bureau of Economic Research. Sarah Green is senior program officer and researcher for the High-Level Task Force for the International Conference on Population and Development. Anthony Keats is assistant professor of economics at Wesleyan University. Jonathan Robinson is associate professor of economics at the University of California, Santa Cruz, and a faculty research fellow of the National Bureau of Economic Research. We thank Kathy Nolan and Kim Siegal for excellent research assistance and IPA Kenya for managing the field work. We thank Cynthia Kinnan, William Lyakurwa, and conference participants at Strathmore University and the 5th NBER Africa conference for helpful comments and suggestions. This study was funded through grants from the International Growth Center, the NBER Africa project, and the International Initiative for Impact Evaluations (3ie). All errors are our own. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters /c13363.ack. 1. Much of their financial access data is from Honohan (2008). 2. At the country level, Chaia et al. (2009) find a weak relationship between urbanization and financial access.

63

64

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Such limited access could potentially have important repercussions on people’s lives. If lacking a formal bank account makes it more difficult for people to save, they will be unlikely to have enough saved up to cope with unexpected emergencies such as household illness. When such shocks occur, rather than withdraw money or take a loan from the bank, people might have to take much costlier actions.3 Lack of banking access might also make it difficult for people to save up large sums or obtain credit for lumpy purchases such as start-up costs for a business, agricultural inputs, or even preventative health products like antimalarial bed nets. Given this, expanding access to even very basic savings and credit services could have large effects. The existing evidence on this issue is somewhat mixed, however. Recent studies suggest that expanding access to microloans alone has only modest effects on most outcomes (e.g., Banerjee et al. 2010; Crépon et al. 2011; Karlan and Zinman 2009). In contrast, studies of programs that increased access to both credit and savings services have found important welfare impacts (see Burgess and Pande [2005] in India, and three studies in Mexico by Aportela [1999], Bruhn and Love [2009], and Ruiz [2010]). Expansion of saving services alone also appears to have the potential to be beneficial. In an earlier experimental study in Kenya, Dupas and Robinson (2009) provided small-scale entrepreneurs access to accounts in a local village bank, and found large effects on business investment and income among a subsample of the study population (market vendors, who are mostly female). In a similar experiment in Nepal, Prina (2011) also finds large impacts of expanding access to savings accounts for women. From a policy standpoint, in addition to understanding the impact of financial inclusion, a critical question is how to achieve it. This is an area that has seen a lot of innovation in the last five years. These recent innovations ultimately amount to either reducing barriers to access to existing financial institutions (e.g., reducing fees), or bringing banking options geographically closer to people.4 For example, a number of countries have adopted “correspondent” or “agent” banking in which people can deposit into and withdraw money from their bank account using a nonbank agent (for example, a retail store).5 A closely related option that has received a substantial amount of recent attention is “mobile money,” in which people can transfer, deposit, 3. Examples of such costly actions include taking children out of school to work on the farm (see Ferreira and Schady [2009] for a recent review article), selling off assets such as business inventory (Dupas and Robinson 2009) or productive animals (Rosenzweig and Wolpin 1993), or engaging in income-generating activities that entail health risk (Robinson and Yeh 2011). 4. Examples of the former type of innovations include the 2006 call made by the Reserve Bank of India to all commercial banks to introduce free “no-frills” accounts (Thyagarajan and Venkatesan 2008), or the 2010 pledge by the Bill and Melinda Gates Foundation to contribute $500 million over five years toward increasing access to savings accounts in poor countries (Bill and Melinda Gates Foundation 2010). 5. See Kumar et al. (2006) for evidence on agent banking in Brazil. McKinsey and Company (2010) provide some background on correspondent banking in several other Latin American countries.

Challenges in Banking the Rural Poor

65

and withdraw money using their cell phone (Jack and Suri 2011). A third approach is a “bank on wheels” in which a vehicle visits a town at regular intervals for people to make transactions.6 A less glamorous approach would be to simply build more ATMs or bank branches (as Equity Bank has done in Kenya with great success; see Allen et al. [2011]). While much attention has recently been paid to these various strategies to expand access, comparatively little attention has been paid to the quality of financial services in very rural areas. If people are not banked because they do not trust banks or banking agents, because they find services to be unreliable, or because account maintenance or withdrawal fees are prohibitive, then expanding such flawed services is unlikely to be appealing. On the demand side, little attention has been paid to understanding reasons other than access for why people may choose to stay out of the formal banking system. This chapter combines survey and experimental evidence from Western Kenya to show that addressing these supply and demand factors is crucial if financial services are to be expanded usefully to unbanked populations. Our study takes place in an area spanning multiple villages surrounding three rural market centers in Western Kenya, and in which banking options remain very limited. In this part of Kenya, large bank branches are located only in major towns, and the villages in our study are far enough away from a town that the cost of traveling there for banking is prohibitive. Locally, there are only two options: a “village bank,” owned by shareholding villagers and affiliated with a microfinance organization, and a partial-service branch (essentially a sales and information office with an ATM) for a major commercial bank. Both banks have substantial minimum balance requirements and withdrawal fees. The village bank also has an account opening fee. The village bank does not pay interest on deposits; effectively, neither does the commercial bank, at least for the poor (interest is only paid if the account balance exceeds 20,000 KSh, or about US$210). To examine financial access among this population, we conducted a census of 1,898 households in the study area between September and December 2009. Account ownership was quite low: only 20 percent of households had at least one member with a bank account. Knowledge of banking options was also limited, as only 60 percent of adults knew of the bank branches in the study area. Almost no one knew the fee schedule for account opening or maintenance. The 1,565 unbanked individuals formed the final experimental study sample. To test whether opening costs (information acquisition, account-opening fees, and administrative requirements) explained the low rates of account ownership, we randomly selected 55 percent of the 1,565 unbanked indi6. Though such banking products exist in many countries, there are few academic studies of their impact. See Stuart, Ferguson, and Cohen (2011) for evidence from Malawi and Nguyen Tien Hung (2004) for evidence from Vietnam.

66

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

viduals to receive a free account at either of the two local banks. We paid the account-opening fees and provided the minimum balance, and arranged for the banks to simplify the account-opening procedure for our study participants. We did not waive withdrawal fees. The majority of people opened accounts when offered this opportunity: take-up was over 60 percent. But actual account usage was much lower. Only 28 percent of those who opened an account (18 percent of those randomly selected for a free account) made at least two deposits on their account in the twelve months after account opening. Many did not use the account at all. Why didn’t the other 80 percent of those selected to receive a free account actively use it? To shed light on this question, we administered qualitative surveys in which respondents could discuss their concerns with the various savings mechanisms available to them. A significant proportion listed risk of embezzlement, unreliable services, and transaction fees as concerns with formal banking. Many of these concerns are valid: the fees are indeed quite high in both the village and commercial bank, and the services in one branch of the village bank were relatively poor during this time period. Furthermore, another branch of the village bank had a recent banking scandal in which withdrawals were frozen for some account holders for a long period. Not surprisingly, we find that trust concerns are more pronounced for the village with the branch with the recent scandal, and reliability concerns are worse for those near the branch with poor service. Interestingly, these concerns were reinforced by exposure to the bank: those who did use their account were more concerned with both the risk of fraud and the lack of reliability than those who did not use the account. We use a similar combination of survey and experimental evidence to examine the demand for formal loans. The banks offer a variety of loans that range in interest between 1.25 and 1.5 percent per month (16 percent– 19.5 percent annual percentage rate [APR]), well below that of many microfinance banks in other parts of the world,7 and well below recent estimated returns to capital, including estimates from previous work in this part of Kenya.8 Yet, very few people take out loans. Of those in our experimental sample, only 6 percent had ever applied for a formal loan at baseline. As with savings options, knowledge of loan options appears extremely limited—very few people know what the conditions are for loans with either bank. Further, when asked, very few people reported wanting loans for agricultural inputs such as fertilizer, despite the high estimated returns to usage in Kenya (Suri 2011; Duflo, Kremer, and Robinson 2011). 7. Kneiding and Rosenberg (2008) report a worldwide average APR of 35 percent. The average in Kenya is over 50 percent per year. See Armendáriz and Morduch (2007), Morduch (1999), and Demirgüç-Kunt, Cull, and Morduch (2009) for more background. 8. See, for example, de Mel, McKenzie, and Woodruff (2008), Fafchamps et al. (2011), and McKenzie and Woodruff (2008). For Western Kenya, see Kremer et al. (2011) and Dupas and Robinson (2009).

Challenges in Banking the Rural Poor

67

To better understand why people do not take up loans, we conducted a randomized credit intervention with two components: (a) an information intervention in which we told people about the requirements and procedures to apply for a loan; and (b) an intervention in which we gave people a voucher that lowered the eligibility requirements necessary to begin taking out loans with the village bank. Though the vast majority of people took the vouchers when offered them, and 40 percent redeemed them, only 3 percent of our experimental sample had even started the process of applying for a loan at the time of writing (six months after the credit information and voucher interventions). Evidence from qualitative surveys on barriers to borrowing suggests that the fear of losing one’s collateral if one cannot repay the loan is the primary deterrent. These results are in line with numerous recent studies in microfinance that show limited demand for microcredit at market rates (e.g., Johnston and Morduch 2008; Banerjee et al. 2010; Crépon et al. 2011). They are also roughly consistent with a recent informational experiment in Sri Lanka that found that only 10 percent of entrepreneurs who were given information about credit options took out loans (de Mel, McKenzie, and Woodruff 2011). Overall, our data reveal a number of challenges with the current supply of financial services. Simply expanding those existing services is not likely to massively increase formal banking use among the majority of the poor unless quality can be ensured, fees can be made affordable, and trust issues are addressed. Our results also suggest that marketing could be improved— a large percentage of people lack even basic information about banking options. Note that while our results are based on two particular banks in one part of Kenya, and concern “classical” banking services rather than agent- or mobile phone-based banking, the general takeaway is that service quality, fees, and trust are important and often overlooked factors. Even M-Pesa, Safaricom’s mobile money network in Kenya and arguably the most developed mobile money product in the world, is ultimately similar in structure to the banks we study here—people must still make deposits and withdrawals in person, in cash, and the fees are substantial. Moreover, M-Pesa, as it is currently constituted, cannot function well as a bank. To guarantee solvency, Safaricom requires agents to pay in advance for any mobile money they purchase. Safaricom then holds this money in bank accounts with several large commercial banks, and gives all interest to charity (Jack and Suri 2011). Clearly, M-Pesa cannot lower fees unless it can invest its deposits for profit—which, in turn, will likely require some form of regulation (for instance, deposit insurance) if people are to trust money with it.9 On top of 9. Of course, some countries may not require even banks to have deposit insurance, which will create a host of other problems. See Demirgüç-Kunt, Karacaovali, and Laeven (2005), which shows that deposit insurance in Africa lags behind other regions.

68

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

this, banks would lobby vociferously to prevent a new entrant into the banking sector—see Mas and Radcliffe (2010) for evidence on this in regard to M-Pesa in Kenya. Given this, it seems that the most likely future for mobile banking is as a platform through which people can transfer money into an account in a formal bank.10 Thus, the issues we raise here remain quite pertinent to mobile banking as well. Our finding that a nonnegligible proportion of people distrust banks generally is somewhat surprising, since the banking sector in Kenya has been relatively stable for some time: while Kenya has had a number of banking scandals, many of these were in the 1980s and 1990s (Central Bank of Kenya 2009), and many involved nonbank financial institutions such as Savings and Credit Co-operations (SACCOs). However, even though the number of bank scandals have been limited in recent years, it is likely that other non-bank-related financial scandals have made people wary, especially of the village bank for which deposits are not insured by the central government. For example, Kenya has had a number of very high-profile pyramid schemes in which an estimated 148,000 people had invested over $90 million (Ministry of Co-Operative Development and Marketing 2009). Both Kenya and neighboring Tanzania have also had high-level scandals that ultimately forced their respective central bank governors to step down. Such scandals might quite naturally cause general mistrust of financial institutions. Our results indicate that, once established, such mistrust sticks for a very long time, and limits the extent to which people seek out information about available financial services, even decades later. This suggests that any effort to expand financial access, if it is to successfully achieve financial inclusion, needs to include an important communication component in order to bring awareness of the various options available as well as the regulation around them (especially deposit insurance). 2.2 2.2.1

Background Information on Rural Banking Financial Institutions in our Study Area

Our data comes from farming villages located near three market centers in Western Province, Kenya. For confidentiality purposes we call these three market centers A, B, and C. Two separate financial institutions operate in this area, a village bank and a commercial bank. The village bank is a community-owned and operated entity that receives support from a local microfinance institution (MFI). Deposits in the bank are not insured by the central bank (though the bank does purchase a limited 10. Safaricom has recently entered into a partnership with a bank to link the M-Pesa account to a formal bank account through the M-Kesho service (Opiyo 2010). Since then, other banks are developing similar services allowing customers to manage their accounts using M-Pesa.

Challenges in Banking the Rural Poor

69

amount of private insurance), and the village bank is classified as a nonbank financial institution. The village bank has three branches in our study area. The main branch is located in market A and opened in 2000. There is a smaller branch in market B, which opened in 2008, and a part-time branch in market C, which opened in September 2009. Branch C only handles account opening, loan applications, and deposits (withdrawals can be made at either of the other two branches, each a bit over 11 km away). The branches in markets A and B are open Monday through Friday from 8:30 a.m. to 4:00 p.m., and on Saturdays from 8:00 a.m. to 1:00 p.m. The branch in market C is only open Monday, Wednesday, and Friday from 9:00 a.m. to 4:00 p.m. The second local financial institution, the commercial bank, is a small branch of a large corporate bank. The branch has an ATM through which people can make deposits and withdrawals at any time and a small staff that assists with these transactions (as well as with account opening and loan applications) during normal business hours (Monday through Friday from 8:30 a.m. to 4:00 p.m. and Saturday from 8:00 a.m. to 12:00 p.m.). The main, full-service branch, where tellers process loan applications and handle transactions, is located in a town about 30 km away. Deposits in the commercial bank are insured. Savings Products The village bank offers just one type of savings account, which does not pay interest. At the time this project started, opening an account at the village bank cost 300 Kenyan shillings (KSh). All accounts must also hold a 100 KSh minimum balance, making the total account opening fee 400 KSh, or about $4.25 US at the current exchange rate. Deposits are free and there are no monthly fees, but there is a fee to make withdrawals.11 No ATM services are available, so savings are illiquid beyond the opening hours mentioned above. The basic savings account at the commercial bank has no account opening fee, but a 200 KSh ($2.10 US) minimum balance requirement.12 The account comes with a free ATM card. The bank charges 30 KSh ($0.32) for withdrawals of any size from the ATM, and 100 KSh ($1.05 US) for withdrawals of any size made at an urban branch. The account pays no interest unless the customer maintains a balance of 20,000 KSh ($210 US) for at least a threemonth period, in which case interest is paid.13 A final way that people in the study area can potentially save is through mobile money, as there are a number of mobile money agents in the area. 11. The withdrawal fee is 10 KSh ($0.10) to withdraw amounts under 1,000 KSh ($10.50), 20 KSh ($0.21) to withdraw amounts between 1,000 and 4,999 KSh ($53), and 100 KSh ($1.05) for amounts of 5,000 KSh or higher. 12. The commercial bank also offers a youth savings account with a smaller minimum balance requirement. 13. The interest rate is variable, ranging from 2–4 percent within the study period.

70

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Mobile money is much more commonly used for transfers than for savings, however, for several reasons. First and most obviously, people need to have access to a cell phone (and only 47 percent of households in our census have a phone). Second, it is not always possible to withdraw money immediately. On the main mobile money network (M-Pesa), the currency of mobile money is “e-float.” The agent holds a balance of e-float on his own cell phone and must decide how much cash to hold to pay out withdrawals. If the agent has a large number of withdrawals on a given day, he may lack the liquidity to cover them all. On the other hand, if there are many deposits, the agent may have no e-float left to sell to allow deposits. These sorts of problems are cited as a drawback by many respondents in our sample. In addition, M-Pesa markets itself as a money transfer, rather than savings, product. Finally, withdrawal fees are substantial (though this is true of both of the banks in our study as well).14 Credit Products While both the village bank and the commercial bank offer credit products, the terms for borrowing vary quite a bit across the two institutions. The village bank, like many MFIs, requires the formation of a group of at least five people who approve the purpose and amount of each other’s loans, and who serve as mutual guarantors. To take out a loan, borrowers must purchase a share (valued at 300 KSh each, or $3.20 US) in the bank. Borrowers are then eligible to borrow up to four times the value of shares owned. In addition, the bank requires borrowers to attend several training sessions on loan management. The village bank offers several different types of loans, most at interest rates between 1.25 and 1.5 percent per month (16–19.5 percent APR). Loans are to be used for business purchases, with the exception of a loan for emergencies, which features a higher interest rate (2.25 percent per month).The commercial bank grants microloans for existing businesses to individuals who have had an account at the commercial bank or with another commercial bank for at least three months. Prospective borrowers must also be visited by a loan officer to assess the state of the business. Loans must be repaid within six months, with an interest rate of 1.5 percent per month. Two guarantors and full collateral are required for each loan.15 2.2.2

History of Financial Scandals

One of the key results of this study is that the level of interest and trust in financial institutions is quite low among rural households. This finding is not particularly surprising when it applies to nonregulated financial institu14. See Jack and Suri (2011) and Mbiti and Weil (2011) for more detail on these issues. 15. Besides these two banks, credit is available from a third institution, which until recently did not take deposits. However, that organization lends only to women with licensed businesses.

Challenges in Banking the Rural Poor

71

tions, such as the village bank that operates in our study area, or Savings and Credit Cooperatives (SACCOs), which have a somewhat charged history of financial scandals, up to the present day. In fact, the village bank in our study area suffered a financial scandal at its main branch (in market A) shortly after we started working in the area. The branch manager was fired for embezzlement, resulting in a months-long liquidity crisis during which existing customers were barred from withdrawing funds over $10.50 a day per customer. During the crisis, the satellite branch in market C was temporarily closed. Though nobody has officially lost deposits, liquidity remains a problem to this day. What is more surprising is that trust in fully regulated financial institutions also appears relatively low, despite the fact that Kenya has had relatively few scandals specific to the regulated banking sector in recent years. Why have rural populations not embraced banks that offer insured deposits? One hypothesis is that they do not make a clear distinction between regulated and unregulated institutions, and their somewhat well-founded mistrust of village banks and SACCOs expands to the banking sector more generally. Another hypothesis is that they remember the various banking crises that Kenya had in previous decades, particularly from 1983 until the late 1990s. These crises were dramatic and hugely costly. For instance, eleven banks were put under liquidation in 1993 alone (Central Bank of Kenya 2009). Outright fraud during crises between 1993 and 1995 was estimated to cost 3.8 percent of GDP (Economist Intelligence Unit 1995), and affected 30 percent of total bank financial assets (Daumont, Le Gall, and Leroux 2004). While many of these crises occurred a number of years ago, it’s likely that memory of them continues to have some effect on perceptions. Another reason that people might be wary is that Kenya has had a number of pyramid schemes and other scams, including a number in recent years. The problem was so severe that the government put together a Task Force on Pyramid Schemes in 2009. The final report of that task force reported that over 148,000 people had invested over $90 million in various pyramid schemes. The largest of these (DECI) had over 93,000 investors alone (Ministry of Co-Operative Development and Marketing 2009). Other scandals have involved corruption at very high levels. In the early 1990s, a number of government officials, including the Governor of the Central Bank, were implicated in the notorious “Goldenberg” scandal, which led to a minimum of $600 million in fraud (Warutere 2005). Note that these issues are not specific to Kenya. A number of other African countries had major banking scandals during the 1980s and 1990s (Caprio and Klingebiel 1997; Brownbridge 1998), several of which cost over 10 percent of gross domestic product (GDP) (Daumont, Le Gall, and Leroux 2004). Within East Africa, Uganda had a banking crisis in which four commercial banks, holding over 12 percent of the nation’s deposits, collapsed over just thirteen months in 1998 and 1999 (Habyarimana 2005;

72

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Brownbridge 2002). Tanzania had a major banking crisis in the late 1980s in which government-owned banks (accounting for 95 percent of total bank assets) became insolvent. Total estimated losses from this crisis were equivalent to about 10 percent of GDP (Daumont, Le Gall, and Leroux 2004). Also in Tanzania, a $120 million banking scandal in 2005 led to the firing of the Governor of the Central Bank (BBC 2008). 2.3 2.3.1

Study Sample, Design, and Data Sample

We first conducted a census of all households living within a four kilometer radius of the three market centers in our study area. The census survey collected information on demographic characteristics of the household, sources of income, as well as access to financial services, knowledge and perceptions of available financial services, and saving practices more generally. A total of 1,898 households were surveyed during the census exercise. Table 2.1 presents some summary statistics on those households. Panel A presents demographic information. The average household had just below six members. Only a few households (11 percent) did not have a female head living in the homestead, but around 31 percent of households did not have a male head living in the homestead.16 Polygamy is still somewhat prevalent—8 percent of households are polygamous. The average household in the study area owned just under two acres of land, and had just above 4,000 KSh ($42) in animal assets. Almost half (47 percent) of households owned a cell phone. Panel B of table 2.1 presents statistics on access to banking services. Only 20 percent of households had a member with a bank account, despite the fact that the average distance to the closest deposit-taking financial institution is only 1.6 kilometers, suggesting that physical access is unlikely to be the barrier. Table 2.2 presents statistics at the individual level, separately for women (panel A) and men (panel B). Average educational attainment is relatively low, with just about six years of education for women and eight years for men. Sixty-five percent of women and 90 percent of men were literate. Almost three-quarters of women reported farming as their primary activity, while only a little over one-third of men did. Own enterprise was the primary occupation of 19 percent of women and 36 percent of men. The remainder worked in physical day labor (mostly associated with agriculture), worked for a wage, or had no job. Table 2.2 also includes individual-level statistics on access to banking. While 21 percent of men had a bank account, only 10 percent of women did. 16. This is the result of two main factors: (a) it is much less common for a widow to remarry than it is for a widower, and (b) some men leave their family behind to work in urban areas.

Challenges in Banking the Rural Poor Table 2.1

73

Baseline household characteristics Full sample (1)

Restricted experimental sample (2)

5.83 (3.05) 0.31 0.11 0.08 3.38 (2.34) 683 (3,058) 0.43 0.48 0.17 0.47 10,482 (9,852) 4142 (9,278) 1.90 (2.86)

5.67 (2.95) 0.38 0.03 0.00 3.34 (2.28) 508 (1,502) 0.39 0.45 0.13 0.40 9,073 (8,448) 4277 (9,424) 1.74 (1.90)

0.20 1.60 (0.74) 2.78 (2.32)

0.00 1.60 (0.71) 3.01 (2.45)

A. Demographic information Total household size No male head No female head Polygamous household Number of children Household health expenditures last month (in KSh) Household treats drinking water with chlorine Iron roof at home Cement floor at home HH has cell phone Value of physical assets (in KSh) Value of animals (in KSh) Land holdings (acres) B. Access to banking At least one member of household has a bank account Distance to closest deposit-taking branch (in km) Distance to closest branch offering withdrawals (in km)

C. Eligibility for experimental treatments Eligible for randomized saving and credit experiments 0.52 Number of households 1,898

1.00 989

Note: Standard deviations in parentheses. The exchange rate at the time of the study was around 80 KSh to US $1 on average.

To construct a sample, the primary eligibility criterion was that nobody in the household had a bank account. However, we also decided to exclude all polygamous households and all households with no female head. The rationale for doing this is that those two categories of households are likely very different from others, yet there are too few of them to do subgroup analysis. In the case of polygamous households, another reason is that measuring expenditures and savings in such households is difficult and time intensive. Given this eligibility criteria, 989 of the 1,898 households in the census were selected to participate in the randomized experiment, comprising 1,565 individuals. As is to be expected, households in the experimental sample are

74

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Table 2.2

Baseline individual characteristics Full sample (1)

Age Years of education Can write in Swahilia Primary occupation: Farming Own enterprise Physical labor Employee None Has bank account Included in experimental sample Number of women Age Years of education Can write in Swahili Primary occupation: Farming Own enterprise Physical labor Employee None Has bank account Included in experimental sample Number of men

A. Women 39.27 (15.98) 6.09 (3.88) 0.65 0.72 0.19 0.02 0.03 0.05 0.10 0.56 1,686 B. Men 41.73 (15.28) 8.10 (3.58) 0.90 0.38 0.36 0.10 0.10 0.06 0.21 0.47 1,299

Restricted experimental sample (2) 40.39 (17.02) 5.34 (3.65) 0.58 0.78 0.15 0.02 0.00 0.05 0.00 1.00 949 40.02 (15.33) 7.35 (3.25) 0.90 0.43 0.36 0.11 0.05 0.04 0.02 1.00 606

Note: Standard deviations in parentheses. a We use writing in Swahili as a proxy for literacy because we have fewer observations with data on being able to read in Swahili. Results look very similar with alternate definitions, however.

poorer, less educated, and more likely to be farmers than other households (see column [2] in tables 2.1 and 2.2). 2.3.2

Experimental Design

Savings Experiment After constructing the sample, we randomly selected individuals for the savings intervention. Randomization was done at the individual (rather than

Challenges in Banking the Rural Poor

75

household) level, stratified by household composition (single female-headed or dual-headed), primary occupation, and market center. The savings intervention was rolled out between May and June 2010. Those individuals who were selected for this intervention received a nominal, nontransferable voucher for a free savings account. For those living within four kilometers of market A (where the commercial bank has an ATM), the voucher was redeemable at either the village bank or the commercial bank. For everybody else, the voucher was for the village bank only. The experiment made it financially costless to open an account: the vouchers covered all account-opening fees (where applicable), including the minimum balance requirement. The vouchers were delivered to people in their homes. During that visit, individuals received information on how the bank and the account work, and when and how to redeem the voucher.17 Among households with no male head, 50 percent were randomly selected to receive an account voucher, which was given to the female head. Among households with both a female and a male head, 20 percent received no voucher, 30 percent received two vouchers (one for each head), and 50 percent received one voucher (in 25 percent of households, the male received the voucher; in the other 25 percent, only the female received the voucher). In total, 55 percent of the sample was selected to receive vouchers. Credit Experiment In February 2011, a second randomization was conducted to lower informational and financial barriers to credit. The intervention differed slightly according to whether individuals had received the savings intervention nine months earlier. Among those who had not received the savings intervention, half were randomly selected to receive information about local credit opportunities. Trained staff visited those individuals at their homes and delivered a detailed script explaining the rules and procedures for obtaining a loan from either of the two local institutions. No financial assistance was given, however. Among those who had received the savings intervention, half were selected to receive the same financial information script as above. However, they were also given a voucher redeemable for one free share at the village bank (valued at 300 KSh, or $3.20). As discussed in the Credit Products section, one of the requirements for getting a loan from the village bank is that an individual must purchase a share (in addition to having a bank account). In particular, the maximum amount that anyone can borrow is four times the amount of share capital they own. While the share is not the only requirement to get a 17. The vouchers expired after two weeks. In practice, most of those who redeemed did so immediately. Commercial bank customers had to visit the branch twice, once to redeem the voucher and again two weeks later in order to pick up their ATM cards and receive training in their use.

76

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

loan (in particular, people must form a group with four other bank clients who approve their loan and act as guarantors), getting a free share does lower the barrier to getting a loan. 2.3.3

Data

We use three main sources of data in this project. First, we have background information (described above) from the census. Second, we have administrative data on deposits, withdrawals, and loan applications from the bank. All study participants that opened an account agreed to sign a waiver allowing their bank to release their bank statements to the research team. We use these bank statements to monitor the saving activity as well as the credit history of our restricted experimental sample. Third, a semistructured survey was administered to a randomly selected half of the restricted experimental sample after nine months. The survey asked respondents openended questions about their current savings practices, perceived barriers to saving, and perceptions of the various saving mechanisms available to them. For those who had received an account voucher but had not redeemed it, the survey also asked why they had not opened an account. The survey also included a number of questions about familiarity with and interest in local credit options. 2.4

Rural Households and their Money: A Snapshot

In tables 2.3 and 2.4 we present information from the census to show how rural households in the study area save. Table 2.3 presents means at the household level, separately by household type. Table 2.4 present means at the individual level, separately by gender. The first striking observation in table 2.3 is the fact that access to formal saving services is very limited. Among the 20 percent of households that have at least one family member with a savings account, only about 12 percent have accounts in a commercial bank (this includes all commercial banks in Kenya, not just the bank that participated in the experiment), 8 percent in the village bank that participated in the experiment, and 3 percent in the post office savings bank. Nobody saves in a microfinance institution. Note that some households have multiple accounts, so these categories are not exclusive. Interestingly, 25 percent of households have a mobile money account. However, most Kenyan households do not currently save in such accounts and instead use them only for transfers (Mbiti and Weil 2011).18 Nevertheless, 18. As discussed in the introduction, formal banks in Kenya have lobbied against the entry of M-Pesa. In part to avoid this controversy, M-Pesa markets itself as a service for transferring money and not for saving. This (along with the withdrawal fees) is likely a big reason that people do not much use M-Pesa as a savings vehicle.

Table 2.3

How do households save?

A. All households Informal savings Owns animals Value of animals (for those who own) Someone in household participates in ROSCA ROSCA contributions in past year (if any) Formal savings Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account B. Married households Informal savings Owns animals Value of animals (for those who own) Someone in household participates in ROSCA ROSCA contributions in past year (if any) Formal savings Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account

All

No. obs.

0.50 4,358 (9,469) 0.53 7,231 (13,121)

1,806 1,751

0.20 0.12 0.03 0.08 0.00 0.03 0.25

1,752 1,752 1,752 1,752 1,752 1,752 1,752

0.52 4762 (9,944) 0.56 8,361 (14,812)

1,195 1,153

0.25 0.16 0.03 0.09 0.00 0.04 0.32

C. Single-headed female households Informal savings Owns animals 0.47 Value of animals (for those who own) 3,637 (8,286) Someone in household participates in ROSCA 0.40 ROSCA contributions in past year (if any) 4,118 (5,471) Formal savings Has account in formal deposit-taking institution 0.09 Has account in commercial bank 0.04 Has account in post bank 0.01 Has account in village bank 0.05 Has account with MFI 0.00 Has account elsewhere 0.01 Has mobile money account 0.09

2,984 924

2,388 675

1,169 1,169 1,169 1,169 1,169 1,169 1,169

501 489 492 208

478 478 478 478 478 478 478

Source: Data from full census sample. Note: Standard deviations in parentheses. Monetary values in Kenyan shillings (KSh). Exchange rate was roughly 80 KSh to $1 US during the sample period.

Table 2.4

How do individuals save? All

No. obs.

0.41 1.50 (0.80) 6,130 (10,443) 0.15 0.08 0.02 0.05 0.00 0.02 0.19

2,605 1,079

A. All Participates in ROSCA If yes: Number of ROSCAs If yes: ROSCA contributions in past year (in KSh) Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account

1,090 2,869 2,869 2,869 2,869 2,869 2,869 2,869

B. Women Participates in ROSCA If yes: Number of ROSCAs If yes: ROSCA contributions in past year (in KSh) Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account

0.45 1.54 (0.82) 5,316 (8,272) 0.10 0.04 0.01 0.05 0.00 0.01 0.12

1608 725 723 1,640 1,640 1,640 1,640 1,640 1,640 1,640

C. Men Participates in ROSCA If yes: Number of ROSCAs If yes: ROSCA contributions in past year (in KSh) Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account

0.36 1.42 (0.74) 7,733 (13,625) 0.21 0.14 0.02 0.06 0.00 0.03 0.28

997 354 367 1,229 1,229 1,229 1,229 1,229 1,229 1,229

Source: Data from full census sample. Note: Standard deviations in parentheses. Exchange rate was roughly 80 KSh to $1 US during the sample period.

Challenges in Banking the Rural Poor

79

the relatively high penetration of such accounts in even very rural areas is potentially very promising. In contrast to the low rates of participation in formal savings, savings through informal mechanisms is quite important—53 percent of households have at least one member who participates in a ROSCA.19 A ROSCA (Rotating Savings and Credit Association) is a savings group (composed of ten to twenty members, typically) that meets on a regular basis; at each meeting, group members make a fixed, mandatory contribution that goes into a “pot” that is then assigned to one of the members. Each member gets the pot in turn. A ROSCA cycle thus requires as many meetings as there are members. Once a cycle is complete, a new cycle can start. Though the structure of ROSCAs varies from place to place, most ROSCAs in this part of Kenya use a predetermined order to allocate the savings pot. Many households (50 percent of the population) also save in animals, which can be used both as a form of savings and as productive assets. The amounts saved in ROSCAs and animals are not trivial—the average household reports saving over 7,200 KSh in ROSCAs ($76 US) over the past year and owning about 4,300 KSh ($45 US) worth of animals. These two forms of informal savings are relatively illiquid, however. Selling animals quickly in response to negative income shocks is not easy, especially if the shock is an aggregate shock at the community level (since the market may be flooded with people selling animals at that time). In the case of ROSCAs, since they typically have a predetermined order, it is impossible to access the money immediately if an emergency comes up. Thus, a more liquid savings option (such as a bank account) could still be useful to people. The breakdown by household type in panels B and C of table 2.3 shows that female-headed households are much less likely to be banked than dualheaded households (9 percent versus 25 percent). They are also less likely to use informal saving mechanisms, suggesting that their overall saving rate is lower. The individual-level means presented in table 2.4 suggest that this gap between household types is essentially driven by a gender divide: only 10 percent of women have banking accounts, compared to 21 percent of men. Similarly, only 12 percent of women have mobile money accounts, compared to 28 percent of men. There are also major differences between those who are primarily farmers and those who are not. We present the statistics disaggregated by gender and farming status in table 2A.1. Only 8 percent of farmers have savings accounts, compared to 23 percent of nonfarmers. Most striking is that only 6 percent of female farmers have accounts. Farmers are also much less likely to participate in a ROSCA or have a mobile money account. Given the low rate of banking, and the fact that the most common infor19. Besley, Coate, and Loury (1993), Anderson and Baland (2002), and Gugerty (2007) discuss various reasons why so many people in developing countries participate in ROSCAs.

80

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

mal saving alternatives are relatively illiquid, a key question is how people deal with emergencies that require immediate liquidity. To shed some light on this issue, our census survey asked people “If you absolutely needed 1,000 KSh ($10.5 US) right now, where would you get the money?” We allowed people to list as many sources as they wanted (so that the categories are not exclusive). The results are presented in table 2.5. We find that only 13 perTable 2.5

If you absolutely needed 1,000 KSh, where would you get the money? All

No. obs.

A. All Would use savings Would work more Spouse would work more Would borrow from friend/relative/neighbor Would get donations from friend/relative/neighbor Would get a loan from ROSCA Would sell household asset/animal/land Would sell business asset Would sell agricultural product Other

0.13 0.14 0.07 0.43 0.13 0.06 0.13 0.01 0.14 0.08

1,984 1,984 1,984 1,984 1,984 1,984 1,984 1,984 1,984 1,984

B. Women If you absolutely needed 1,000 KSh, where would you get the money? Would use savings Would work more Spouse would work more Would borrow from friend/relative/neighbor Would get donations from friend/relative/neighbor Would get a loan from ROSCA Would sell household asset/animal/land Would sell business asset Would sell agricultural product Other

0.08 0.12 0.09 0.45 0.16 0.06 0.12 0.01 0.15 0.08

1,221 1,221 1,221 1,221 1,221 1,221 1,221 1,221 1,221 1,221

C. Men If you absolutely needed 1,000 KSh, where would you get the money? Would use savings Would work more Spouse would work more Would borrow from friend/relative/neighbor Would get donations from friend/relative/neighbor Would get a loan from ROSCA Would sell household asset/animal/land Would sell business asset Would sell agricultural product Other

0.20 0.16 0.04 0.38 0.08 0.05 0.14 0.01 0.13 0.09

763 763 763 763 763 763 763 763 763 763

Source: Data from full census sample. Note: Respondents could give more than one answer to the question (i.e., categories are not mutually exclusive).

Challenges in Banking the Rural Poor

81

cent of people would be able to get even part of the money from savings. Most people would ask others for help, while others would have to sell a household asset or work more. Although it is conceivable that people could fully make up for a 1,000 KSh shortfall by relying on others, nearly every study of interhousehold risk coping suggests that this is unlikely. Thus, it seems likely that increasing savings would better allow people to cope with shocks. 2.5

Understanding Low Levels of Formal Banking

This section discusses factors that partially explain the low observed rate of formal banking. We start by describing our baseline survey evidence. We find that at the time our study began people knew very little about local financial institutions, suggesting that earlier marketing activities by these financial institutions, if any, had been mostly unsuccessful. We then present evidence from the randomized savings experiment. Overall, while we find that reducing the account-opening fees and minimizing the hassle of opening an account did induce a minority to start saving in the bank, we find that most people did not use their accounts. Survey evidence suggests that the major reasons people did not use the bank is that they were concerned about high withdrawal fees and poor service, and that they did not trust their money with the bank. Note that given our experimental design, it is not surprising that distance to a local banking option does not appear as a major factor, as the sample was drawn from villages within walking distance of the bank. 2.5.1

Survey Evidence: Baseline Interviews

Table 2.6 presents data from the census on knowledge of and trust in the village bank, separately by branch. At the time of the census, the village bank had been established in market center A for nearly ten years, in market center B for about eighteen months, and in market center C for ten months. Despite this, only 64 percent of household heads in markets A and B, and 51 percent in market C, had ever heard of the village bank. Even those who had heard of the bank did not know enough about it to have an opinion about it. Thus, when those who had at least heard of the village bank were asked if they would trust the bank with their money, 43 percent said they did not know enough about the bank to respond. Around 49 percent said they would trust the bank, while the remaining 8 percent said they would not. The main reasons for not trusting the bank were lack of familiarity and being worried about embezzlement of funds. Table 2.7 addresses those in the experimental sample, who were all unbanked at the time of the census, and to whom a slightly more detailed survey was administered. In this sample, we asked about knowledge of both the village bank and the commercial bank. Unsurprisingly, familiarity with

Table 2.6

Perception of village bank

A. All Have you heard of the village bank? Do you trust the village bank? Don’t know Yes No Of those who don’t trust the village bank, why? Worried that the village bank will take my money Don’t know the village bank/unfamiliar with banking Fees are high No interest Bank is unreliable Other

All

Market A

Market B

Market C

No. obs.

0.60

0.64

0.64

0.51

2,018

0.43 0.49 0.08

0.41 0.52 0.08

0.44 0.48 0.08

0.46 0.46 0.09

1,191 1,191 1,191

0.23

0.10

0.49

0.17

111

0.43 0.10 0.04 0.08 0.12

0.64 0.08 0.02 0.04 0.14

0.23 0.11 0.06 0.03 0.09

0.29 0.13 0.04 0.25 0.13

111 111 111 111 111

0.61

0.57

0.43

1,492

0.41 0.53 0.07

0.46 0.46 0.08

0.49 0.44 0.07

803 803 803

0.11

0.56

0.25

67

0.57 0.08 0.03 0.03 0.19

0.28 0.11 0.00 0.00 0.06

0.42 0.08 0.00 0.17 0.08

67 67 67 67 67

0.78

0.81

0.68

526

0.40 0.50 0.10

0.41 0.50 0.09

0.41 0.48 0.11

388 388 388

0.07

0.41

0.08

44

0.80 0.07 0.00 0.07 0.00

0.18 0.12 0.12 0.06 0.12

0.17 0.17 0.08 0.33 0.17

44 44 44 44 44

B. Women Have you heard of the village bank? 0.55 Do you trust the village bank? Don’t know 0.44 Yes 0.49 No 0.07 Of those who don’t trust the village bank, why? Worried that the village bank will take my money 0.25 Don’t know the village bank/unfamiliar with banking 0.46 Fees are high 0.09 No interest 0.02 Bank is unreliable 0.05 Other 0.13 C. Men Have you heard of the village bank? 0.75 Do you trust the village bank? Don’t know 0.41 Yes 0.49 No 0.10 Of those who don’t trust the village bank, why? Worried that the village bank will take my money 0.21 Don’t know the village bank/unfamiliar with banking 0.39 Fees are high 0.11 No interest 0.07 Bank is unreliable 0.14 Other 0.09

Source: Data is from the full census sample. Note: The village bank in market A had a freeze on withdrawals a few months prior to the survey. The bank in market C does not allow withdrawals and is often closed during business hours.

Table 2.7

Familiarity with local financial institutions among the unbanked All

A. All Distance to closest deposit-taking branch 1.63 (in km) (0.71) Distance to closest branch offering 3.04 withdrawals (in km) (2.44) Has heard of the local village bank 0.52 If yes: Knows account opening fee at local village bank 0.08 Has heard of local commercial banka If yes: Knows comm. bank accounts are free to open Would use bank account if had one 0.85 Would choose village bank over commercial bank if had choice 0.38 B. Women Distance to closest deposit-taking branch 1.60 (in km) (0.71) Distance to closest branch offering 3.01 withdrawals (in km) (2.45) Has heard of the local village bank 0.47 If yes: Knows account opening fee at local village bank 0.05 Has heard of local commercial bank If yes: Knows comm. bank accounts are free to open Would use bank account if had one 0.81 Would choose village bank over commercial bank if had choice 0.39 C. Men Distance to closest deposit-taking branch 1.67 (in km) (0.72) Distance to closest branch offering 3.08 withdrawals (in km) (2.44) Has heard of the local village bank 0.73 If yes: Knows account opening fee at local village bank 0.14 Has heard of local commercial bank If yes: Knows comm. bank accounts are free to open Would use bank account if had one 0.92 Would choose village bank over commercial bank if had choice 0.36

Market A

Market B

Market C

No. obs.

1.87 (0.69) 1.87 (0.69) 0.54

1.40 (0.77) 1.40 (0.77) 0.57

1.49 (0.52) 7.00 (0.73) 0.45

1,260

0.11 0.59

0.07

0.00

0.00 0.85

0.87

0.83

71 1,468

0.31

0.47

0.41

1,320

1.85 (0.69) 1.85 (0.69) 0.52

1.37 (0.77) 1.37 (0.77) 0.49

1.46 (0.53) 7.01 (0.73) 0.37

828

0.09 0.54

0.03

0.00

247 71

0.00 0.82

0.82

0.79

37 958

0.31

0.48

0.45

858

1.92 (0.70) 1.92 (0.70) 0.71

1.45 (0.79) 1.45 (0.79) 0.80

1.54 (0.50) 6.99 (0.73) 0.67

432

0.14 0.67

0.15

0.00

142 54

0.00 0.91

0.98

0.91

34 510

0.31

0.45

0.34

462

1,260 1,122 389 125

828 914

432 208

Note: Data consists of restricted experimental sample. Standard deviations in parentheses. a This question only asked in market A (where the commercial bank has a branch).

84

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

local financial institutions is lower among these unbanked individuals than in the full census sample. Only about half of household heads had heard of the village bank across the three market centers, though awareness was higher in markets A and B, and lower in market C (where the village bank had only recently started a deposit-only branch). Very few individuals knew the details of the services offered by the village bank, however—only 8 percent of those who had heard of the bank knew the cost of opening an account. Despite the fact that the commercial bank, located in market A, had only opened in late 2008 (eight years after the village bank), by 2009 it had the same level of name recognition as the village bank. Just as with the village bank, however, people knew very little about the products offered at the commercial bank: none of the respondents knew that accounts were free (with only a minimum balance requirement). Though people do not know much about either bank, most people tend to prefer the commercial bank (likely because it is a large, well-established bank with a national presence). When asked which institution they would prefer to have an account in, close to two-thirds of respondents said they would choose the commercial bank over the village bank. 2.5.2

Experimental Evidence

The randomized savings experiment allows us to test the extent to which eliminating opening fees, facilitating account opening, and providing information can increase access to formal banking. Table 2.8 presents figures on take-up of the experimental offer of a free bank account. A relatively large fraction of individuals elected to open an account: overall take-up was 62 percent. In market A, where both banks are available, the commercial bank was the favorite choice: 43 percent of people opened an account at the commercial bank, compared to only 17 percent at the village bank. Across the branches, take-up was lowest in market C, where the village bank only offers partial service. However, many of those who opened accounts did not actively use them. In table 2.8, we define an account as “active” if the respondent made at least two deposits in the year following the account opening date. We find that only 28 percent of opened accounts were active. Since only 62 percent of people even opened accounts, this means that the overall usage rate was only 0.28 × 0.62 = 18 percent. In table 2.9, we show the results separately for men and women (pooling all the market centers together). While women were slightly less likely to open accounts than men, they were 10 percentage points more likely to actively use the account if they opened one. Overall, the active take-up rate was thus higher among women than men (19.5 percent versus 14.3 percent), but still relatively modest among both groups. Overall, these results suggest that entry costs—be it the cost of acquiring information, the

85

Challenges in Banking the Rural Poor Table 2.8

Experimental results: Take-up and usage of free accounts among those initially unbanked

Opened an account Opened account at village bank Opened account at commercial bank If opened an account: “Active” (= at least 2 deposits) If village bank account: “Active” If commercial bank account: “Active”a If “Active”: Average number of deposits Total deposited on account (KSh) Average deposit size (KSh) Average number of withdrawals Average withdrawal size (KSh) Account joint with spouse (if married) Overall: Active take-up of free account

All

Market A

Market B

Market C

No. obs.

0.625 0.427 0.198

0.613 0.175 0.438

0.753 0.744 0.009

0.530 0.526 0.004

840 840 840

0.28 0.23

0.36 0.22 0.39 4.57 (3.46) 6,477 (13,174) 1,288 (2,901) 2.27 (4.49) 1,760 (2,269) 0.060 0.218

0.27 0.28

0.17 0.17

4.71 (3.44) 2,221 (3,836) 460 (713) 1.42 (2.67) 845 (1,063) 0.039 0.206

4.36 (3.79) 730 (1,818) 132 (228) 0

525 359 166 147

4.58 (3.48) 4,314 (10,231) 862 (2,223) 1.68 (3.71) 1,455 (1,990) 0.058 0.176

0 0.078 0.089

147 147 148 66 397

Source: Data from subset of individuals sampled for free account (among restricted experimental sample). Note: Accounts were opened in May–July 2010 and follow-up data is from May 2011 (approximately ten to twelve months after account opening). Standard deviations in parentheses. Monetary values in Kenyan shillings (KSh). Exchange rate was roughly 80 KSh to $1 US during sample period. a Accounts at the commercial bank were only offered in market A (where the commercial bank has a branch).

opening fees (including minimum balance requirement) or the administrative hassle of opening an account—explain only about one-fifth of the low banking rates observed in our study area. 2.5.3

More Survey Evidence: Debriefing Interviews

To understand what other supply factors explain the relatively low demand for formal banking we observe once the entry costs were experimentally removed, we asked respondents, in an open-ended way, what their concerns were with the various saving mechanisms available to them. We asked these questions to a random subset of our restricted experimental sample. The results are presented in table 2.10. We present the results separately for those in the control group (who did not receive information and assistance with account opening), those in the treatment group who did not actively use the account (whom we call noncompliers), and those in the treatment group who

86

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Table 2.9

Experimental results: Take-up and usage of free accounts among those initially unbanked (by gender)

Opened an account Opened account at village bank Opened account at commercial bank If opened an account: “Active” (= at least 2 deposits) If village bank account: “Active” If commercial bank account: “Active”a If “Active”: Average number of deposits Total deposited on account (Ksh) Average deposit size (Ksh) Average number of withdrawals Average withdrawal size (Ksh) Account joint with spouse (if married) Overall: Active take-up of free account Observations

Women

Men

0.611 0.415 0.195

0.649 0.448 0.201

0.32 0.27 0.41 4.29 (3.26) 1,966 (3,955) 480 (1,144) 0.96 (1.56) 1,059 (1,658) 0.071 0.197

0.22 0.16 0.34 5.27 (3.91) 9,637 (16,421) 1,727 (3,506) 3.31 (6.03) 2,148 (2,345) 0.045 0.140

532

308

Source: Data from subset of individuals sampled for free account (among restricted experimental sample). Notes: Accounts were opened in May–July 2010 and follow-up data is from May 2011 (approximately ten to twelve months after account opening). Standard deviations in parentheses. Monetary values in Kenyan shillings (KSh). Exchange rate was roughly 80 KSh to $1 US during sample period. a Accounts at the commercial bank were only offered in market A (where the commercial bank has a branch).

did use the accounts (whom we call compliers). We present results for the three formal and quasi-formal banking options available: the commercial bank, the village bank, and mobile money. The main concerns raised with formal banks are transaction fees, unreliability, and risk of embezzlement. The relative importance of these concerns varies substantially between the two financial institutions in our study area. Transaction fees are the primary concern with the commercial bank, which charges a flat fixed fee of 30 KSh per withdrawal at the ATM, making it very costly to make small withdrawals. While these withdrawal fees could act as a commitment device to not withdraw money until a relatively large lump sum has been saved, they can also deter people from saving in the account if they anticipate needing small sums to deal with emergencies as they arise. This is in line with a related study we conducted in this part of Kenya, in

Challenges in Banking the Rural Poor Table 2.10

87

Concerns with local financial institutions among those initially unbanked Commercial bank

A. Control group (no account voucher) Concerns with savings option: Fees 0.34 Unreliable 0.16 Distance 0.11 Risk of embezzlement 0.06 Agent can’t always handle transactionsa — Requires phone — Observations 283

Village bank

Mobile money

0.15 0.32 0.02 0.17 — — 294

0.11 0.01 0.02 0.00 0.29 0.36 292

B. Noncompliers (offered account voucher but did not open account or is not actively using account) Concerns with savings option: Fees 0.39 0.21 Unreliable 0.15 0.37 Distance 0.19 0.03 Risk of embezzlement 0.07 0.24 Agent can’t always handle transactions — — Requires phone — — Observations 285 284

0.05 0.02 0.02 0.00 0.33 0.38 284

C. Compliers (offered account voucher, opened account and actively using account) Concerns with savings option: Fees 0.46 0.16 0.09 Unreliable 0.17 0.43 0.01 Distance 0.11 0.02 0.01 Risk of embezzlement 0.06 0.21 0.00 Agent can’t always handle transactions — — 0.22 Requires phone — — 0.35 Observations 79 82 82 Source: Data from restricted experimental sample. a If customers make a large number of withdrawals on a given day, the agent may run out of liquidity. If customers make a large number of deposits, he may run out of e-float. See text for more details.

which we find that the cost of limiting liquidity exceeds its benefit for many people (Dupas and Robinson 2011). For the village bank, while fees remain a major concern, substantial fractions of people also report unreliability and risk of embezzlement as problems. Among the noncompliers, 37 percent cite unreliability and 24 percent cite risk of embezzlement, suggesting that many of those who did not actively take up village banking thought service quality was poor or lacked trust in the institution. In regard to mobile banking, the most common concerns are that it requires owning a cell phone and that there are network or liquidity issues

88

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

(i.e., the agent runs out of “e-float,” to pay out withdrawals). Fees are less of a concern for mobile money banking than for formal banks, even though in practice the fees associated with mobile money fall somewhere in between the fees charged at the village bank and those charged at the commercial bank. Notably, trust in mobile banking is extremely high (another promising sign if mobile money is eventually to be mobilized for savings as well as transfers). Table 2.11 provides further qualitative evidence on these issues by disaggregating results by market center. Recall that there was a withdrawal freeze in the wake of an embezzlement scandal in market A and that the service in market C is spotty, so we might expect people to trust the village bank least in market A and find it most unreliable in market C. Interestingly, table 2.11 shows that this is true only of people who actively used the accounts. Though the sample of people who use their account is obviously selected, one interpretation of these findings is that people’s experience with the village bank reinforced the mistrust in the institution. Table 2.11 also reports responses to a question in which we asked people for their preferred savings options. All in all, when asked what their preferred savings mechanism would be if they could choose, over 40 percent of respondents answered “a commercial bank.” A sizable fraction also reported the village bank. As expected, this share is higher in the control and compliers groups than in the noncompliers group. Somewhat surprisingly, mobile money banking was the least favorite mechanism, behind grain storage and ROSCA participation. In fact, almost 40 percent of the control group reported informal options (animals, ROSCAs, or saving in grain) as their preferred saving tool. Given the risks associated with these informal saving mechanisms, the fact that they remain preferred is suggestive that the formal products being offered are insufficient for many people. This is consistent with the finding that close to a quarter of respondents said they had been discouraged to open a bank account by a friend. 2.5.4

Open Questions

The evidence presented thus far has focused on supply issues. However, these issues only partially explain the low formal-savings rates we observe in our experiment. Many of those in the treatment group who do not list trust, fees, or reliability as concerns still do not use the accounts. When asked directly what keeps them from saving, many of them say that their expenses are too high or that their income is simply too low for them to save at all. However, it is hard to know how to interpret these responses. Existing evidence strongly suggests that even extremely poor people can save. For instance, research in Gambia (Shipton 1990) and Bangladesh, India, and South Africa (Collins et al. 2009) demonstrates that poor households do find ways to save, albeit often through informal mechanisms. Moreover, Banerjee and Duflo (2007) find that even among the poorest households—those living

Table 2.11

Concerns with village bank by market center All

A. Control group (no account voucher) Concerns with village bank Fees 0.15 Unreliable 0.32 Distance 0.02 Risk of embezzlement 0.17 Preferred banking option Commercial bank 0.43 Village bank 0.07 M-Pesa 0.06 Animals 0.21 ROSCA 0.11 Grain 0.07 Have you ever been discouraged to open an account by a friend or relative? 0.17 Observations 294

Market A

Market B

Market C

0.19 0.33 0.01 0.19

0.07 0.21 0.03 0.14

0.17 0.40 0.01 0.16

0.40 0.05 0.08 0.22 0.11 0.07

0.40 0.10 0.07 0.25 0.06 0.10

0.49 0.08 0.02 0.15 0.15 0.05

0.19 138

0.16 73

0.14 83

B. Noncompliers (offered account voucher but did not open account or is not actively using account) Concerns with village bank Fees 0.21 0.23 0.27 0.12 Unreliable 0.37 0.34 0.36 0.40 Distance 0.03 0.02 0.03 0.05 Risk of embezzlement 0.24 0.26 0.21 0.23 Preferred banking option Commercial bank 0.36 0.39 0.30 0.35 Village bank 0.17 0.06 0.30 0.23 M-Pesa 0.04 0.05 0.04 0.02 Animals 0.22 0.22 0.19 0.25 ROSCA 0.07 0.10 0.03 0.07 Grain 0.08 0.09 0.08 0.07 Have you ever been discouraged to open an account by a friend or relative? 0.24 0.24 0.27 0.21 Observations 284 125 70 89 C. Compliers (offered account voucher, opened account and actively using account) Concerns with village bank Fees 0.16 0.11 0.32 Unreliable 0.43 0.41 0.26 Distance 0.02 0.00 0.11 Risk of embezzlement 0.21 0.28 0.11 Preferred banking option Commercial bank 0.49 0.67 0.20 Village bank 0.29 0.13 0.55 M-Pesa 0.02 0.02 0.05 Animals 0.12 0.09 0.10 ROSCA 0.04 0.02 0.10 Grain 0.04 0.07 0.00 Have you ever been discouraged to open an account by a friend or relative? 0.31 0.26 0.40 Observations 82 46 19

0.12 0.65 0.00 0.12 0.35 0.41 0.00 0.24 0.00 0.00 0.31 17

Source: Data from restricted experimental sample. Note: The village bank in market A had a recent freeze on withdrawals. The bank in market C does not allow withdrawals and was often closed during business hours.

90

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

at or under $1 per person per day—the majority do not exhaust all their income on basic necessities. Ultimately, the low take-up rate in this study begs the question: Is a savings account in a bank relatively far from home well tailored for people who can only save in very small increments? Providing a more convenient place to save, or stronger incentives to make deposits, may be more effective. For example, in previous work we find that people save quite readily if provided with a lock box and key that they can keep at home (Dupas and Robinson 2011). Furthermore, providing people with a credit incentive to make deposits, and social pressure to continue making them, was extremely effective in mobilizing savings. Similarly, Duflo, Kremer, and Robinson (2011) find that providing people with small incentives to set aside money for fertilizer when people have money (after harvest) increases fertilizer investment. Other recent papers have shown how prompting people to save (Atkinson et al. 2010) or providing people with reminders to save can also be quite effective (Kast, Meier, and Pomeranz 2012; Karlan et al. 2010). Indeed, in countries like the United States, where many transactions are conducted electronically, it is has been well documented that savings can be most effectively mobilized when they are “unseen”; for example, when wage increases are automatically put into a savings account (Thaler and Benartzi 2004) or when people are automatically opted in to a certain 401(k) savings level (Carroll et al. 2009). Designing such products in a much more cashbased economy may be difficult, but is worth exploring. In fact, the value of mobile money (such as M-Pesa in Kenya) may be largely in making savings more electronic; it might be less costly, both in terms of transportation and time, to transfer an electronic balance to a linked savings account than to physically take cash to the bank during operating hours. It may also be less painful psychologically to devote electronic money to savings rather than to physically put cash into a savings account. 2.6

Understanding Low Borrowing Rates

While much of our focus has been on savings, the flipside of savings is credit. Returns to capital have been estimated to be quite high in the study area (as discussed in footnote 12), higher than the APR on loans offered by the two financial institutions (which vary from 16–19.5 percent APR). What keeps people from taking out such loans and reaping high returns? We examine this issue in tables 2.12–2.14. To start, table 2.12 describes familiarity with local credit options among our restricted experimental sample of unbanked households. As with savings, people have very limited information about credit options. Only 64 percent think there is a local credit option and only 38 percent (41 percent) correctly identified the village bank (commercial bank), respectively, as a credit option. Only 15 percent said that

91

Challenges in Banking the Rural Poor Table 2.12

Baseline knowledge of local credit opportunities and interest in loans among those initially unbanked

Is there a local institution in which you can get loans? Yes No Don’t know Correctly identified village bank as local credit option Correctly identified commercial bank as local credit optiona Says knows procedure for loan Really knows procedure for loan Interested in loan at 1.5% monthly interest without collateral Interested in loan at 1.5% monthly interest with full collateral Do you think you could qualify for a loan? Yes No Don’t know Has ever applied for loan Has ever gotten loan

All

No. obs.

0.64 0.11 0.26 0.38 0.41 0.15 0.07 0.74 0.32

665 665 665 667 311 660 658 645 643

0.37 0.22 0.41 0.06 0.05

664 664 664 537 536

Source: Data from random subset of restricted experimental sample. a This question was only asked in market A, where the commercial bank has a branch.

they knew the procedure to get a loan; interestingly, only 7 percent could correctly describe the procedure when asked.20 To gauge potential interest in loans, we asked people if they were interested in a loan at 1.5 percent monthly interest, with or without collateral requirements. While 74 percent said that they were interested if no collateral was required, this dropped to only 32 percent with collateral. We also asked people if they thought that they could qualify for a loan, and 37 percent said yes. Yet only 6 percent had ever applied for loan. Given this low level of information, we implemented an intervention to improve information and access to credit (the details are presented in the Credit Experiment section). We present two sets of results of this intervention. First, in table 2.13, we report the answers to questions we asked participants immediately after they received detailed information about local credit opportunities. While a majority of people were interested in a loan at the village bank, far fewer were interested at the commercial bank. This is likely primarily due to the fact that the commercial bank only lends to people with an existing business, and as shown in table 2.2, only 15 percent of women and 36 percent of men in our restricted experimental sample had a business at baseline (farming is not considered a business by the bank). What’s more, most of those have a very small market vending business with 20. As shown in table 2A.2, these levels are even lower among farmers.

92

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Table 2.13

Interest in loans among those initially unbanked, immediately after receiving information on local credit opportunities

Interested in loan at village bank Interested in loan at commercial bank If interested in a loan: Purpose of loana Farm inputs Farm equipment Start business Business inventory Business equipment Home construction Home repair Furniture School fees Health care Wedding Land Debts Other If interested in a loan: Desired loan amount (in KSh) Mean Median Standard deviation Months needed before can make first repayment Percent say would be able to make first payment within 1 month Percent say would be able to make first payment within 2 months

All

No. obs.

0.60 0.35

645 644

0.11 0.02 0.77 0.19 0.09 0.04 0.00 0.01 0.08 0.00 0.00 0.01 0.00 0.11

98 98 98 98 98 98 98 98 98 98 98 98 98 98

18,878 10,000 31,813 2.15 0.41 0.70

95 95 95 95 95

Note: Results restricted to those who received the credit intervention. a Due to problem in the skip code on a version of the survey, this question was only asked for a subset of people who were interested in a loan.

very low levels of working capital, and they would most likely not quality for a loan from the commercial bank in any case. It is also likely that people perceive it as more difficult to qualify for a loan from the commercial bank, even aside from the business requirement. Among those interested in borrowing, we asked what they would want to borrow for. Few people were interested in loans for small investments like agricultural inputs. People were much more likely to report wanting to start a business or adding to business inventory with a loan. Whether these types of goals are feasible for such borrowers is an open question. The second set of results from the experimental credit intervention is presented in table 2.14, and concerns the take-up of the “share capital” voucher we gave to a random subset of those who had already been offered an account voucher. By redeeming this voucher, people would be credited with one village bank share (valued at 300 KSh) and thus be eligible to bor-

Table 2.14

Experimental results: Impacts of credit intervention All

A. Share voucher + information Intervention All Accepted voucher 0.87 Redeemed voucher 0.40 Inquired about loan at village bank 0.028 Completed loan training at village bank 0.011 Formed group at village bank 0.014 Got loan at village banka 0.003 Women Accepted voucher 0.85 Redeemed voucher 0.37 Inquired about loan at village bank 0.043 Completed loan training at village bank 0.017 Formed group at village bank 0.021 Got loan at village bank 0.004 Men Accepted voucher 0.91 Redeemed voucher 0.46 Inquired about loan at village bank 0.00 Completed loan training at village bank 0.000 Formed group at village bank 0.000 Got loan at village bank 0.000 B. Information only intervention All Inquired about loan at village bank 0.00 Completed loan training at village bank 0.00 Formed group at village bank 0.00 Got loan at village bank 0.00 Women Inquired about loan at village bank 0.00 Completed loan training at village bank 0.00 Formed group at village bank 0.00 Got loan at village bank 0.00 Men Inquired about loan at village bank 0.00 Completed loan training at village bank 0.00 Formed group at village bank 0.00 Got loan at village bank 0.00

No. obs.

358 358 358 358 358 358 233 233 233 233 233 233 125 125 125 125 125 125

296 296 296 296 196 196 196 196 100 100 100 100

Notes: For data, see text for detailed description of interventions. Loan take-up is updated through August 31, 2011. Respondents in panel A received both a voucher for one share at the bank and information on how to apply for a loan. Respondents in panel B received information only. Information on loan take-up is from the village bank only. As of August 31, 2011, no respondents had applied for a loan at the commercial bank. a Exactly one person had qualified for a loan by August 31, 2011.

94

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Table 2.15

What factors might prevent you from getting a loan?

Don’t need the money Afraid bank will seize collateral Too risky Don’t trust the bank Don’t like the idea of being in debt Have too much other debt Too much hassle I don’t have a business, which is required for loan I can’t pay immediately Other

All

No. obs.

0.14 0.51 0.45 0.09 0.08 0.01 0.12 0.27 0.18 0.38

179 179 179 179 179 179 179 179 179 179

Note: Results restricted to those who received the credit intervention.

row up to four times the value of the share (1,200 KSh). Note, however, that this is much smaller than the median desired loan size listed in table 2.13, which was 10,000 KSh. Table 2.14 shows that, while 87 percent accepted the voucher when it was given to them, only 40 percent redeemed it, and as of the time of writing (six months after vouchers had been distributed), only 3 percent of individuals had started the process of applying for a loan by making an inquiry, and only one person (out of 358) had applied for and been granted a loan. While we have not yet followed up with these individuals directly to ask why they did not end up applying for a loan, we did ask people about concerns about taking out loans at the time the vouchers (and information) were given out. These results are reported in table 2.15. Overwhelmingly, people report that they are afraid of losing collateral or that taking out a loan is risky. Thus, even at relatively low interest rates, the fear of losing assets overwhelmed loan demand in our study area. Clearly, this creates a serious problem in generating a market for credit, since the interest rate would have to be much higher if loans were not collateralized. 2.7

Conclusion

Without a safe place to save up money, it may be very difficult for people to take advantage of high-return investments of many types. Likewise, without a safe place to keep an emergency cash buffer, vulnerability to shocks might be very high. Recognizing this, policymakers and international aid organizations have begun to devote attention to expanding access to financial services in developing countries, especially in rural areas where access continues to be extremely limited. This chapter shows that unless serious attention is paid to the reliability and quality of financial services offered, simply expanding

Challenges in Banking the Rural Poor

95

access by reducing monetary or time costs will fail to effectively achieve financial inclusion. Our analysis generates several important insights that bear on policies designed to expand financial services to the poor. First, trust is an important reason that people do not use current banking services. Providing stronger consumer protection through tighter regulation and deposit insurance could be very important. Second, many people are uninformed about banking options (in part because they have little or no experience with them). Better marketing from the banks themselves might be important for raising the use of financial services. Finally, more attention should be paid to the types of products that banks provide. While basic savings accounts do appear to be useful to a minority, more sophisticated products might be necessary for others (just as they are for many people in developed countries). For example, many people in Kenya save to deal with health emergencies, which are very common. For them, putting money into a bank that does not offer withdrawal services at night or on weekends and that has big withdrawal fees might not be very attractive. Similarly, people whose income is seasonal (such as farmers, who make up the great majority of the rural poor in subSaharan Africa) might benefit from products that provide stronger incentives to save as soon as they have money. In this sense, it is good news that as many as 18 percent of people in our poor, rural sample took up and actively used basic savings accounts when they could access them for free. However, the evidence we presented suggests that this is a lower bound on potential demand for formal saving products. Serious attention should be paid to improving the delivery of financial services—doing so could improve the lives of millions of people.

96

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Appendix Table 2A.1

Savings for farmers and nonfarmers Female

Male

No. obs.

0.39

0.42

0.32

1,576

1.43 (0.74) 4,362 (7,245)

1.46 (0.77) 4,205 (6,147)

1.35 (0.60) 4,881 (10,051)

0.08 0.04 0.01 0.04 0.00 0.01 0.13

0.06 0.02 0.01 0.03 0.00 0.01 0.09

0.15 0.09 0.02 0.05 0.00 0.01 0.24

1,657 1,657 1,657 1,657 1,657 1,657 1,657

0.45

0.53

0.39

1,014

1.60 (0.87) 8,731 (13,451)

1.72 (0.89) 7,797 (11,302)

1.46 (0.82) 9,704 (15,340)

0.23 0.15 0.02 0.07 0.00 0.03 0.26

0.19 0.09 0.02 0.10 0.00 0.02 0.19

0.25 0.18 0.02 0.06 0.00 0.04 0.31

All A. Farmers Informal savings Participates in ROSCA Number of ROSCAs (for those who participate) ROSCA contributions in past year (for those who participate) Formal savings Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account

620 643

B Nonfarmers Informal savings Participates in ROSCA Number of ROSCAs (for those who participate) ROSCA contributions in past year (for those who participate) Formal savings Has account in formal deposit-taking institution Has account in commercial bank Has account in post bank Has account in village bank Has account with MFI Has account elsewhere Has mobile money account Source: Data from full census sample.

455 443

1,197 1,197 1,197 1,197 1,197 1,197 1,197

Table 2A.2

Baseline knowledge of local credit opportunities and interest in loans among those initially unbanked (farmers vs. nonfarmers)

A. Farmers Is there a local institution in which you can get loans? Yes No Don’t know Correctly identified village bank as local credit option Correctly identified commercial bank as local credit option Says knows procedure for loan Really knows procedure for loan Has ever applied for loan Has ever gotten loan Do you think you could qualify for a loan? Yes No Don’t know Interested in loan at 1.5% monthly interest without collateral Interested in loan at 1.5% monthly interest with full collateral B. Nonfarmers Is there a local institution in which you can get loans? Yes No Don’t know Correctly identified village bank as local credit option Correctly identified commercial bank as local credit option1 Says knows procedure for loan Really knows procedure for loan Has ever applied for loan Has ever gotten loan Do you think you could qualify for a loan? Yes No Don’t know Interested in loan at 1.5% monthly interest without collateral Interested in loan at 1.5% monthly interest with full collateral

All

Female

Male

No. obs.

0.59 0.12 0.29 0.34

0.55 0.13 0.33 0.30

0.73 0.08 0.19 0.47

451 451 451 452

0.37 0.12 0.05 0.05 0.04

0.31 0.11 0.04 0.04 0.03

0.55 0.17 0.09 0.08 0.04

212 446 444 360 359

0.34 0.21 0.45

0.30 0.24 0.46

0.46 0.14 0.40

449 449 449

0.74

0.71

0.85

424

0.27

0.22

0.47

423

0.74 0.09 0.17 0.47

0.62 0.11 0.28 0.34

0.84 0.07 0.09 0.58

196 196 196 197

0.47 0.19 0.09 0.09 0.06

0.38 0.12 0.06 0.06 0.03

0.56 0.24 0.11 0.11 0.08

95 196 196 160 160

0.43 0.24 0.34 0.73

0.30 0.30 0.40 0.67

0.52 0.19 0.29 0.77

197 197 197 204

0.41

0.29

0.50

203

Source: Data from random subset of restricted experimental sample. 1 This question was only asked in market A, where the commercial bank has a branch.

98

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

References Allen, Franklin, Elena Carletti, Robert Cull, Jun “QJ” Qian, Lemma Senbet, and Patricio Valenzuela. 2011. “Improving Access to Banking: Evidence from Kenya.” Working Paper, Financial Institutions Center, Wharton School, University of Pennsylvania. Anderson, Siwan, and Jean-Marie Baland. 2002. “The Economics of Roscas and Intrahousehold Resource Allocation.” Quarterly Journal of Economics 117:963–95. Aportela, Fernando. 1999. “Effects of Financial Access on Savings by Low-Income People.” Working Paper, Banco de México. Armendáriz, Beatriz, and Jonathan Morduch. 2007. The Economics of Microfinance, 2nd ed. Cambridge, MA: MIT Press. Atkinson, Jesse, Alain de Janvry, Craig McIntosh, and Elisabeth Sadoulet. 2010. “Creating Incentives to Save among Microfinance Borrowers: A Behavioral Experiment from Guatemala.” Working Paper, BASIS, University of California, Davis. Banerjee, Abhijit, and Esther Duflo. 2007. “The Economic Lives of the Poor.” Journal of Economic Perspectives 21:141–67. Banerjee, Abhijit, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan. 2010. “The Miracle of Microfinance? Evidence from a Randomized Evaluation.” Working Paper, MIT Department of Economics. BBC. 2008. “Tanzania in Bank Scandal Sacking.” BBC News, January 10. http:// news.bbc.co.uk/2/hi/africa/7181065.stm. Besley, Timothy, Stephen Coate, and Glenn Loury. 1993. “The Economics of Rotating Savings and Credit Associations.” American Economic Review 83 (4): 792–810. Bill and Melinda Gates Foundation. 2010. “Melinda Gates Challenges Global Leaders: Create Savings Accounts and Bring Financial Security to the World’s Poorest.” Press Release, November 16. http://www.gatesfoundation.org/press-releases /Pages/melinda-gates-at-global-savings-forum-101116.aspx. Brownbridge, Martin. 1998. “The Causes of Financial Distress in Local Banks in Africa and Implications for Prudential Policy.” UNCTAD Discussion Paper Series no. 132, United Nations Conference on Trade and Development. http://unctad .org/en/docs/dp_132.en.pdf. ———. 2002. “Resolving Bank Failures in Uganda: Policy Lessons from Recent Bank Failures.” Development Policy Review 20 (3): 279–29. Bruhn, Miriam, and Inessa Love. 2009. “The Economic Impact of Banking the Unbanked: Evidence from Mexico.” Policy Research Working Paper no. 4981, Washington, DC, World Bank. Burgess, Robin, and Rohini Pande. 2005. “Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment.” American Economic Review 95 (3): 780– 95. Caprio, Gerard, Jr., and Daniela Klingebiel. 1997. “Bank Insolvency: Bad Luck, Bad Policy, or Bad Banking?” In Annual World Bank Conference on Development Economics, edited by M. Bruno and B. Pleskovic, 79–104. Washington, DC: World Bank. Carroll, Gabriel, James J. Choi, David Laibson, Brigitte Madrian, and Andrew Metrick. 2009. “Optimal Defaults and Active Decisions: Theory and Evidence from 401(k) Saving.” Quarterly Journal of Economics 124 (4): 1639–74. Central Bank of Kenya. 2009. “About DPF [Deposit Production Fund].” Accessed August 30, 2011. http://www.centralbank.go.ke/dpfb/Default.aspx.

Challenges in Banking the Rural Poor

99

Chaia, Alberto, Aparna Dalal, Tony Goland, Maria Jose Gonzalez, Jonathan Morduch, and Robert Schiff. 2009. “Half the World is Unbanked.” FAI Framing Note, Financial Access Initiative, NYU Wagner. http://wagner.nyu.edu/faculty /publications/half-world-unbanked. Collins, Daryl, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven. 2009. Portfolios of the Poor: How the World’s Poor Live on $2 a Day. Princeton, NJ: Princeton University Press. Crepon, Bruno, Florencia Devoto, Esther Duflo and William Pariente. 2011. “Impact of Microcredit in Rural Areas of Morocco: Evidence from a Randomized Evaluation.” Working Paper, MIT Department of Economics. Daumont, Roland, Françoise Le Gall, and François Leroux. 2004. “Banking in Sub-Saharan Africa: What Went Wrong?” IMF Working Paper no. WP/04/55, International Monetary Fund. de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital in Microenterprises: Evidence from a Field Experiment.” Quarterly Journal of Economics 123 (4): 1329–72. ———. 2011. “Getting Credit to High Return Microentrepreneurs: The Results of an Information Intervention.” World Bank Economic Review 25 (3): 456–85. Demirgüç-Kunt, Asli, Robert Cull, and Jonathan Morduch. 2009. “Microfinance Meets the Market.” Journal of Economic Perspectives 23 (1): 167–92. Demirgüç-Kunt, Asli, Baybars Karacaovali, and Luc Laeven. 2005. “Deposit Insurance around the World: A Comprehensive Database.” World Bank Working Paper Series no. 3628, Washington, DC, World Bank. Duflo, Esther, Michael Kremer, and Jonathan Robinson. 2011. “Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya.” American Economic Review 101 (6): 2350–90. Dupas, Pascaline, and Jonathan Robinson. 2009. “Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya.” NBER Working Paper no. 14693, Cambridge, MA. ———. 2011. “Why Don’t the Poor Save More? Evidence from Health Savings Experiments.” NBER Working Paper no. 17255, Cambridge, MA. Economist Intelligence Unit. 1995. Country Report–Kenya, fourth quarter. Nairobi. Fafchamps, Marcel, David McKenzie, Simon Quinn, and Christopher Woodruff. 2011. “When Is Capital Enough To Get Female Microenterprises Growing? Evidence from a Randomized Experiment in Ghana.” Working Paper, Department of Economics, University of Oxford. http://www.economics.ox.ac.uk/Centre-for -the-Study-of-African-Economies-Series/when-is-capital-enough-to-get-female -microenterprises-growing-evidence-from-a-randomized-experiment-in-ghana. Ferreira, Francisco, and Norbert Schady. 2009. “Aggregate Economic Shocks, Child Schooling, and Child Health.” World Bank Research Observer 24 (2): 147–81. Gugerty, Mary Kay. 2007. “You Can’t Save Alone: Commitment in Rotating Savings and Credit Associations in Kenya.” Economic Development and Cultural Change 55:251–82. Habyarimana, James. 2005. “The Benefits of Banking Relationships: Evidence from Uganda’s Banking Crisis.” Working Paper, Georgetown University. Honohan, Patrick. 2008. “Cross-Country Variation in Household Access to Financial Services.” Journal of Banking and Finance 32 (11): 2493–500. Jack, William, and Tavneet Suri. 2011. “Mobile Money: The Economics of M-PESA.” Working Paper, Georgetown University and MIT Sloan. http://www .mit.edu/~tavneet/M-PESA-Final2.pdf. Johnston, Don Jr., and Jonathan Morduch. 2008. “The Unbanked: Evidence from Indonesia.” World Bank Economic Review 22 (3): 517–37.

100

Pascaline Dupas, Sarah Green, Anthony Keats, and Jonathan Robinson

Karlan, Dean, Margaret McConnell, Sendhil Mullainathan, and Jonathan Zinman. 2010. “Getting to the Top of the Mind: How Reminders Increase Saving.” NBER Working Paper no. 16205, Cambridge, MA. Karlan, Dean, and Jonathan Zinman. 2009. “Expanding Microenterprise Credit Access: Using Randomized Supply Decisions to Estimate the Impacts in Manila.” Working Paper, Dartmouth College and Yale University. Kast, Felipe, Stephan Meier, and Dina Pomeranz. 2012. “Under-Savers Anonymous: Evidence on Self-Help Groups and Peer Pressure as a Savings Commitment Device.” HBS Working Paper no. 12-060, Harvard Business School. http:// www.hbs.edu/faculty/Publication%20Files/12-060_25eb4ab1-7f56-470b -b3a9-8e039b44118e.pdf. Kendall, Jake, Nataliya Mylenko, and Alejandro Ponce. 2010. “Measuring Financial Access around the World.” Policy Research Working Paper no. 5253, Washington, DC, World Bank. Kneiding, Christoph, and Richard Rosenberg. 2008. “Variations in Microcredit Interest Rates.” CGAP Brief, July. Washington, DC, Consultative Group to Assist the Poor. http://www.cgap.org/publications/variations-microcredit-interest-rates. Kremer, Michael, Jean Lee, Jonathan Robinson, and Olga Rostapshova. 2011. “The Return to Capital for Small Retailers in Kenya: Evidence from Inventories.” Working Paper, Department of Economics, Harvard University. http://www.cid .harvard.edu/neudc07/docs/neudc07_s5_p09_kremer.pdf. Kumar, Anjali, Ajai Nair, Adam Parsons, and Eduardo Urdapilleta. 2006. “Expanding Bank Outreach through Retail Partnerships: Correspondent Banking in Brazil.” World Bank Working Paper no. 85, Washington, DC, World Bank. Mas, Ignatio, and Dan Radcliffe. 2010. “Mobile Payments Go Viral: M-PESA in Kenya.” Working Paper, World Bank. http://documents.worldbank.org/curated /en/2010/03/12180031/mobile-payments-viral-m-pesa-kenya. Mbiti, Isaac, and David Weil. 2011. “Mobile Banking: The Impact of M-Pesa in Kenya.” NBER Working Paper no. 17129, Cambridge, MA. McKenzie, David, and Christopher Woodruff. 2008. “Experimental Evidence on Returns to Capital and Access to Finance in Mexico.” World Bank Economic Review 22 (3): 457–82. McKinsey and Company. 2010. “Global Financial Inclusion.” Accessed August 29, 2011. http://www.mckinsey.com/clientservice/Social_Sector/our_practices/Eco nomic_Development/ Knowledge_Highlights/ ~/ media/ Reports/ SSO/ GFI_v13 _complete.ashx. Ministry of Co-Operative Development and Marketing [Kenya]. 2009. Report of the Taskforce on Pyramid Schemes. Nairobi: Government of Kenya. Morduch, Jonathan. 1999. “The Microfinance Promise.” Journal of Economic Literature 37 (4): 159–614. Nguyen Tien Hung, G. 2004. “Bank on Wheels.” Finance & Development Nbr. 41-2, June. Opiyo, Dave. 2010. “Banking via Phone Made Possible.” Daily Nation, May 18. Prina, Sylvia. 2011. “Do Simple Savings Accounts Help the Poor to Save? Evidence from a Field Experiment in Nepal.” Working Paper, Case Western Reserve University. Robinson, Jonathan, and Ethan Yeh. 2011. “Transactional Sex as a Response to Risk in Western Kenya.” American Economic Journal: Applied Economics 3 (1): 35–64. Rosenzweig, Mark, and Kenneth Wolpin. 1993. “Credit Market Constraints, Consumption Smoothing, and the Accumulation of Durable Production Assets in Low-Income Countries: Investments in Bullocks in India.” Journal of Political Economy 101 (2): 223–44.

Challenges in Banking the Rural Poor

101

Ruiz, Claudia. 2010. “From Pawn Shops to Banks: The Impact of Formal Credit on Informal Households.” Working Paper, University of California, Los Angeles. Shipton, Parker. 1990. “How Gambians Save and What Their Strategies Imply for International Aid.” Policy Research Working Paper Series no. WPS 395, Washington, DC, World Bank. Stuart, Guy, Michael Ferguson, and Monique Cohen. 2011. “Cash In, Cash Out: Financial Transactions and Access to Finance in Malawi.” Financial Services Assessment Report. Accessed August 29, 2011. http://www.fsassessment.umd.edu. Suri, Tavneet. 2011. “Selection and Comparative Advantage in Technology Adoption.” Econometrica 79 (1): 159–209. Thaler, R., and S. Benartzi. 2004. “Save More Tomorrow: Using Behavioral Economics to Increase Employee Saving.” Journal of Political Economy 112:S164–87. Thyagarajan, S., and Jayaram Venkatesan. 2008. “Cost-Benefit and Usage Behavior Analysis of No Frills Accounts: A Study Report on Cuddalore District.” Working Paper, Centre for Micro Finance. http://www.cab.org.in/Lists/Knowledge%20 Bank/Attachments/100/Nofrills%20in%20Cuddalore%20%20Report.pdf. Warutere, P. 2005. “The Goldenberg Conspiracy: The Game of Paper Gold, Money and Power.” ISS Paper no. 117, Institute for Security Studies. https://www.issafrica .org/uploads/Paper117.pdf.

3

The Financial Sector in Burundi An Investigation of Its Efficiency in Resource Mobilization and Allocation Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

3.1

Introduction

The postindependence period in Burundi has been characterized by low and volatile growth, which has made it difficult for the country to achieve national development goals, especially poverty reduction. Factors that account for the sluggish and volatile growth range from physical constraints (e.g., Burundi is landlocked) that raise the costs of production and trade, to political instability (Nkurunziza and Ngaruko 2008). The country’s political and economic instability has constrained the mobilization of public and private domestic resources, thus limiting investment, entrepreneurship, and growth in productivity. Yet private investment and enterprise development are important drivers of employment creation, poverty reduction, long- term growth, and economic resilience through diversification and expansion of the growth base. Janvier D. Nkurunziza is chief, Commodity Research and Analysis Section, Special Unit on Commodities, UNCTAD, Geneva. Léonce Ndikumana is professor of economics in the economics department and Director of the African Policy Program at the Political Economy Research Institute (PERI) at the University of Massachusetts Amherst. Prime Nyamoya is the CEO of Organisation et Gestion Industrielle (OGI), Bujumbura, and former CEO of Banque de Crédit de Bujumbura (BCB), Burundi’s oldest bank. The authors thank the NBER Africa Project for funding support. The authors are grateful for constructive comments and suggestions from Lant Pritchett, Sebastian Edwards, and Simon Johnson (as lead discussants) and other participants at the NBER Africa Project meetings on December 11– 12, 2009, in Cambridge, MA, and on July 17, 2010, in Accra, Ghana. The authors also thank Lionel R. Ngenzebuke and Christophe Niyonganji for excellent research assistance, and Elisa Pepe for diligently copy editing the paper. Thanks are also extended to the institutions and individuals who agreed to share their data with us. The opinions expressed in this chapter are solely those of the authors and do not represent those of their institutions of affiliation. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13359.ack.

103

104

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Access to finance affects poverty through direct and indirect channels. The direct channel relates to income generated by job creation following new investments. The indirect channel is the wealth effect of economic growth resulting from an increase in investment. This study investigates the extent to which the financial sector contributes to the process of growth and poverty reduction through financial resource mobilization and efficient allocation of resources to activities and sectors with the highest economic and social returns.1 The study acknowledges the importance of demand- side factors as impediments to accessing finance, especially sluggish growth and the resulting stagnation of incomes. Due to the complexity of the topic, the study adopts an eclectic approach, drawing from macroeconomic analysis, political economy, and industrial organization. On the one hand, the chapter discusses the impact of economic performance, and shocks to economic activity, on financial intermediation. On the other hand, the chapter discusses the failure of the financial sector to fuel an increase in investment and growth. The study examines the sectoral and temporal allocation of resources and attempts to explain banks’ revealed preference for short- term credit and specific sectors. By comparing the distribution of resources by sector and term structure to the risk profile measured by default rates, a risk- efficiency index is constructed to assess the efficiency of resource allocation from a risk perspective. The main finding of the study is that poor governance and political instability have affected the behavior and performance of financial intermediaries. In particular, the chapter shows that the highly centralized and monolithic regimes that have ruled the country over the past five decades used the financial sector and the economy in general for rent seeking and consolidation of power rather than economic development (Ndikumana 1998, 2005; Nkurunziza and Ngaruko 2000, 2008). As a result, the need to control economic and political power determined or influenced the ownership of banks, undermined the independence and regulatory capacity of the central bank, impeded the management of public financial institutions, and perpetuated inefficiencies in the allocation of resources, particularly of credit. Hence, the credit market is segmented between “insiders” who benefit from preferential credit terms and “outsiders” who are relatively penalized. The inability and unwillingness of the central bank to enforce appropriate regulation and supervision is considered one of the main factors that explain the weakness of the financial sector. The study contributes to the literature on the effect of political economy factors on economic performance. Most studies on the effect of political factors on firm and industry performance in developing countries have focused on Asia (Wurgler 2000; Khwaja and Mian 2005; Hsieh and Klenow 2009; 1. One key indicator of social returns to investment in a poor country like Burundi is the extent to which it reduces poverty.

The Financial Sector in Burundi

105

Fisman 2001; Leuz and Oberholzer-Gee 2003).2 Faccio (2006) provides a cross- country analysis of forty- seven countries with only two from Africa (South Africa and Zimbabwe) and the analysis is not disaggregated enough to provide detailed evidence on the situation in these countries. This chapter is one of the few detailed studies on the subject in Africa. Moreover, unlike most countries where privatization in the 1980s and 1990s reduced the influence of the public sector in the management of financial institutions, the state in Burundi is still very influential in the financial sector despite some progress toward its liberalization. As Shleifer and Vishny (1994) show, the state can still exert a high degree of control over privatized firms in cases of privatization without commercialization. In Burundi, most financial firms have only been partially privatized, and the state is still the most important shareholder either directly or through state- owned companies. The state has representatives on the boards of almost all major financial institutions. The problem is that the loyalty of state representatives lies with their political backers, not the firms they are supposed to manage. The influence of the state in the financial sector is also transmitted through the state’s pervasive presence in other sectors of the economy. The state has never favored the emergence of a strong private sector, so it remains a dominant economic actor. As a result, there are private enterprises—including in the financial sector—that owe their existence to government contracts and hence political connections. Given that transactions with state entities are not necessarily based on competitive bidding, privileged firms often collapse when the country’s political leadership changes (Nkurunziza and Ngaruko 2008). In this environment, many firms strive to be in state representatives’ good graces in order to survive. Similar to Khwaja and Mian (2005), the present study uses disaggregated data to identify some of the reasons why the financial sector in Burundi has had a very limited effect on development. The next section provides the context of the study by presenting the political economy of growth and poverty reduction. Section 3.3 begins with a brief overview of financial liberalization and proceeds to examine the structure and characteristics of the financial sector. Section 3.4 examines the term structure and sectoral credit allocation and the contribution of banks to economic activity. This leads, in section 3.5, to an analysis of the performance of the sector both at the firm level (bank profitability) and sectoral level. The section discusses the fragility of financial intermediaries and the risk of bank failure. The analysis of a unique microeconomic data set uncovers features of credit allocation that are motivated by political objectives rather than profit maximization. The section ends with a discussion of the role of the financial sector in poverty reduction, which is probably the 2. According to Hsieh and Klenow (2009), capital misallocation due to political factors could explain 30 percent to 50 percent of TFP difference between Chinese and American firms, and 40 percent to 60 percent of TFP difference between Indian and American firms.

106

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

most pertinent dimension of financial- sector efficiency in a poor country like Burundi. Section 3.6 concludes the chapter with a summary of key findings. 3.2 3.2.1

The Political Economy of Growth and Poverty A History of Political Instability and Poor Governance

Knowledge of Burundi’s political economy is essential for understanding the performance of the country’s financial sector over the past four decades.3 Burundi’s postindependence history has been marked by political violence. Since independence in 1962, the country has experienced five episodes of civil war: 1965, 1972, 1988, 1991, and from 1993 to 2003. These civil wars claimed more than half a million lives and generated more than one million refugees (Nkurunziza and Ngaruko 2005). These conflicts have been characterized as opposing the country’s two main groups, the Hutu and Tutsi, despite the fact that it is difficult to objectively delineate the boundaries separating them. These two groups share the same language and culture, and live in mixed communities across the country.4 A deeper analysis of conflict in Burundi shows that political violence resulted from a combination of three key factors (Nkurunziza and Ngaruko 2005, 2008; Ntibazonkiza 1993; Ndikumana 1998; Lemarchand 1994). The first factor relates to the divisive policies instituted by the Belgian colonial authority in the 1920s and 1930s, which opposed Hutu and Tutsi, destroying a regulatory system that had been used to keep a balance between the two groups. The second factor is the 1959 social revolution in Rwanda, a neighboring country with a similar ethnic configuration to that of Burundi. The Hutu in Rwanda captured power through a violent rebellion against the traditional Tutsi monarchs. This had a negative demonstration effect on Burundi where Hutu elites made several attempts to emulate the Rwandan revolution, arguing that their majority should guarantee them de facto control of political power. Tutsi elites responded by deploying their political, military, and diplomatic power to prevent a Rwandan- type revolution from happening. This created and perpetuated tension between Hutu and Tutsi elites. The third factor is the failure of postindependence elites to put policies in place that could reestablish and safeguard mechanisms that guarantee peaceful coexistence between the different groups. Most particu3. This section draws from Nkurunziza and Ngaruko (2005, 2008), Nkurunziza (2010a, 2010b), and Ndikumana (1998, 2005). Also see Ntibazonkiza (1993) and Lemarchand (1994) for more detailed historical accounts of ethnic conflicts in Burundi. 4. There are two additional small groups: the Twa and the Ganwa. The former are loosely integrated into Burundi’s society. The latter are mostly descendants of Burundi kings who ruled the country until the monarchy was abolished in 1966. The Hutu are considered to be by and large the majority of the population and the Tutsis a very large minority.

The Financial Sector in Burundi

107

larly, the unwillingness to punish the state officials who were responsible for masterminding the bloody civil wars instilled a sense of mistrust of the state by the majority of the population. Indeed, Burundi’s postcolonial political elite have invested in instituting a political system aimed at capturing the state in order to conserve power. We discuss this latter factor in some detail as it has had the most profound effect on the country’s economy, including the performance of the financial sector. From the mid- 1960s until 2005, political power and, to a large extent, economic power, were controlled by a Tutsi- dominated group from the southern province of Bururi. The three presidents who ruled Burundi from 1966 to 2003, except for the period between July 1993 and July 1996, took power through a military coup d’état. They were Tutsis from the Rutovu commune in the Bururi province. Through a system of patronage, they ruled the country with an iron fist, under the protection of a strong army and security apparatus totally dedicated to their protection and that of their associates. Indeed, security was fully controlled by Bururi natives (Nkurunziza and Ngaruko 2005). The political elite and their private- sector associates enjoyed all the spoils of power while excluding the majority of the population from political and economic participation. In addition to dissuasive methods used by the army and state security to ensure that power remained fully controlled by the ruling elite, several other tactics were used. For example, Bururi had the best schools that prepared the future elites. Entry to the state- controlled University of Burundi—which until recently was the country’s single institution of higher education—was filtered to limit the number of non-Bururi students, who were seen as future potential political challengers (Nkurunziza and Ngaruko 2008; Ndikumana 1998). In order to extract and appropriate maximum rents from the population, several methods were employed. For example, the state controlled the economy directly through pervasive interventionist measures, and indirectly through a large number of state enterprises, including major banks and other financial institutions. The allocation of employment in these enterprises was an important source of rents as the best jobs were reserved for the members of the ruling class. Most of these firms were poorly managed and were only kept afloat by large government subsidies. For example, state firms produced only 9 percent of Burundi’s GDP in 1990 but accounted for 31 percent of formal- sector employment, 25 percent of outstanding domestic credit, and benefited from 3.4 percent of GDP in subsidies from the government. In 1995, equity capital of thirty- six firms with majority state participation amounted to 20 percent of the country’s GDP. In the same year, these firms recorded a net loss equivalent to 6 percent of GDP or 14 percent of government revenue, excluding grants (Ngaruko and Nkurunziza 2006). Essentially, the losses in state- owned enterprises (SOEs) represented appropriation of public resources by firm managers and their employees.

108

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Subsequent subsidies to such loss- making firms, which were not conditional on good management, could be considered taxes levied on the general public and transferred to members of the elite. Another indirect way of levying rents from the population was the pricing of cash crops, especially coffee, the country’s main export, which was largely produced in the north of the country. Traditionally, the government allocated a small portion of the international price to producers (only 40 percent in the 1970s and 1980s), transferring the remainder to government- controlled institutions and wellconnected intermediaries. Economic controls such as import licenses and access to foreign currency were also used to favor selected individuals and enterprises.5 As could be expected, policies based on exclusion, state capture by a small group, and rent extraction created economic and social inequalities, breeding resentment from the excluded. It is not surprising that these politically motivated inequalities caused violent reactions from the excluded segments of the population, and, ultimately, a cycle of civil wars. Hence, rather than alleged ethnic antagonisms, these wars are best characterized as the outcome of distributional conflict (Ndikumana 2005). In 1993, pressure from the international community forced the ruling elite to organize democratic elections, which were overwhelmingly won by Melchior Ndadaye, who was inaugurated on July 10, 1993. This was the first time since the first military coup in 1966 that a non-Bururi civilian and Hutu became president. Less than four months into his term, President Ndadaye was assassinated by members of the old elite. The reaction of the Hutu population was immediate: they engaged in blind killings of innocent Tutsis to “avenge” the assassination of “their” president, sparking a tenyear- long civil war that claimed about 300,000 lives and generated a large number of refugees. The country and its economy are yet to recover from the devastating effects of this war. The year 2005 brought a fundamental change to Burundian politics. The CNDD-FDD (Conseil pour la Défense de la Démocratie-Force de Défense de la Démocratie), the main Hutu- dominated rebel group that waged the ten- year rebellion following the assassination of Melchior Ndadaye in 1993, forced the ruling elite to negotiate a political settlement in order to stop the war. It is in this context that elections were organized in 2005 and won by CNDD-FDD. This victory was celebrated by most Burundians as heralding a new era of a more democratic and inclusive political system. The new leaders were among the prominent victims of exclusionary policies under previous regimes, so these new leaders were expected to have a higher sense of responsibility and adopt a better governance system than their predeces5. Today, foreign currency allocation is mostly liberalized and the coffee sector is in the process of liberalization. As a result, in 2009 and 2010, producer prices substantially increased.

The Financial Sector in Burundi

109

sors. After the first five- year term, the same team from CNDD-FDD won another term as a result of elections organized in June-August 2010.6 Having won the elections two consecutive times, CNDD-FDD is now challenged to fulfill the promise of improvement in the political system as has been expected since 2005. The 2005 transition mainly consisted of a power transfer from the southern Tutsi elite to the Hutu elite, but so far there has been little positive impact on the population. The current leadership is a replica of the old one, with the difference being that rents are now controlled and channeled through CNDD-FDD party structures, a mirror image of the old one- party system.7 Despite this important change in Burundian politics, most members of the traditional elite, particularly those in the private sector, are still better off than other segments of the population. The assets they were able to accumulate over several decades under monolithic regimes are shielding them from the deep poverty that afflicts the majority of Burundians. Most bank executives and employees belong to or are connected to this group. 3.2.2

Effect of Poor Governance on the Financial Sector, Economic Growth, and Poverty Reduction

Bad governance has had direct nefarious effects on the financial sector. Traditionally, the state in Burundi either fully owned or had controlling shares in most Burundian financial institutions. As was the case with other state- owned enterprises (SOEs), these institutions were used as sources of rents. Managers and employees were carefully selected to ensure that financial institutions remained in the hands of the ruling elite. Bank credit was often issued on the basis of political connections rather than projects’ expected returns. A large fraction of bank loans were never paid back and poor management led many banks to the brink of collapse, prompting the central bank to bail them out. This was indirect resource transfer from the public to the defaulting customers, bank managers, and employees. In many cases, banks that were bailed out kept their management team who continued to use these institutions to extract rents. In other cases, mismanagement led to the collapse of several financial institutions (see discussions in section 3.6). To date the financial sector in Burundi remains fragile, partly as a result of political interference. Political instability and poor governance affected the financial sector through supply and demand factors. High inflation, for example, increased the average nominal interest rates in order to maintain high real interest 6. The CNDD-FDD’s victory was largely contested by the opposition, with all but one opposition party pulling out of the presidential elections in protest against claims of vote rigging in the local (commune- level) elections. 7. For example, the government is beset by several cases of alleged corruption that have never been credibly investigated.

110

Fig. 3.1

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Trends of key macroeconomic variables

Source: Based on data from BRB.

rates. Figure 3.1 shows that during the war period (especially 1994– 1996) there was a jump in inflation and lending interest rates and the two variables remained persistently high. For most of the period between 1985 and 2009, inflation remained over 10 percent per year while the average lending rate reached 19 percent in the early years of the twenty- first century. As expected, the high lending rate restricted access to credit. The demand for credit was also negatively affected by the erosion of consumers’ purchasing power due to high inflation and steady decline in real income per capita. In constant

The Financial Sector in Burundi

111

2000 US dollars, GDP per capita declined from about $150 in the mid- 1980s to $101 in 2007. The collapse in GDP per capita was associated with patronage and the policies of exclusion that hindered economic growth and poverty reduction. Since independence in 1962, annual real GDP growth has reached the 6 percent mark only nine times, and GDP growth was negative for a large part of the 1990s. In the first decade of the twenty- first century, GDP growth turned positive, albeit at low values, but its volatility increased. As a result, per capita GDP growth rates remained low given the high population growth rate of about 3 percent per year. Slow growth of GDP has prevented meaningful reduction in poverty as will be discussed below. The empirical literature shows that robust and sustained investment is a fundamental driver of long- term economic growth (Barro 1991). In Burundi, gross capital formation has remained below 15 percent of GDP for most of the postindependence period, dropping below 5 percent over the war period between 1994 and 2003. Burundi’s investment performance is much below the sub-Saharan average and below its peers in the East African Community (figure 3.2). In addition to the failure to mobilize domestic savings, Burundi has also attracted little private capital and continues to depend heavily on official development assistance.8 While Burundi has managed an unprecedented political transition with institutionalized, albeit imperfect, mechanisms for power sharing that was intended to help alleviate the risks of ethnic antagonisms (Bertelsmann Stiftung 2009), it still faces the critical challenge of initiating and sustaining a robust postconflict economic recovery. The failure to mobilize private finance for long- term investment has prevented Burundi from exploiting its growth potential.9 In particular, mineral resources have not been exploited due to the inability to mobilize the needed $4.6 billion for infrastructure investments. The country also lacks a comprehensive natural resource development plan, which must include a major scaling up of energy supply.10 According to the AfDB (2009), the country could achieve a real GDP growth 8. The government has been receiving about $450 million of aid per year from all sources (AfDB Database). The African Development Bank report on infrastructure in Burundi (AfDB 2009) estimates that an additional 30 percent of that amount would be needed to complement financing from government and the private sector to fill the country’s infrastructure financing deficit. 9. The country possesses substantial mineral resources including nickel, cassiterite, and columbo- tentalie (coltan), and reasonable amounts of other minerals, notably phosphate and gold, as well as potential for substantial production of ceramics (from kaolinite and feldspar) and cement (from carbonate rocks). Nickel is the largest mineral resource, with about 284 million tons in Musongati, Waga, Nyabikere, and Murera, which represent some of the world’s largest nickel deposits (AfDB 2009). It is believed that the actual reserves could be even higher. 10. Despite a dense hydrographic network that could supply up to 6,000 GWh/year of hydroelectric power, electricity consumption in Burundi (20 kWh per capita per year) is among the lowest in the developing world. Only 2 percent of the population has access to electricity, compared to 16 percent in sub-Saharan Africa, and 41 percent in low- income countries.

112

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Fig. 3.2 Gross capital formation in Burundi compared to other EAC countries and SSA average (% of GDP) Source: World Bank, World Development Indictors. Note: SSA = sub-Saharan Africa.

rate of about 7.4 percent over the 2010– 2030 period by implementing an infrastructure investment program that would allow, among other things, full exploitation of Burundi’s nickel mines. Moreover, patronage policies that allocated economic resources based on political considerations rather than efficiency have contributed to the erosion of the country’s productivity, lowering the rate of growth. Taking 1960 as the base year, total factor productivity—a measure of efficiency—had divided by 25 in 1997 while GDP per capita had declined by 40 percent. This is despite the fact that physical capital per capita had multiplied by 58 from an extremely low initial stock. In fact Burundi recorded, on average, negative productivity growth in the period from 1960 to 1997. From 1989 to 1997, a period of high political instability, negative productivity growth (– 5.25 percent annually, on average) was the main cause of the negative average economic growth rate of – 2.8 percent per year (Nkurunziza and Ngaruko 2008). Low productivity prevented a structural transformation of the economy that could have generated higher and stable levels of economic growth. While the share of services in GDP has increased substantially, the Burundian economy is still heavily dependent on the primary sector, which is dominated by rain- fed agriculture (figure 3.3), explaining the high volatility of growth. Low productivity in the agriculture sector is illustrated by the steep decline in the contribution of the primary sector to GDP without any decline in the share of the population dependent on primary sector activities. Other constraints to growth in Burundi range from physical factors

The Financial Sector in Burundi

Fig. 3.3

113

Structure of Burundi’s economy

Source: Authors’ calculations based on data from the Central Bank of Burundi, annual reports.

including unfavorable geography, poor infrastructure, and high production and transport costs, as well as policy and institutional constraints. The country is landlocked and depends on poor regional infrastructure and logistics networks, resulting in high production and transportation costs. It is estimated that transport costs account for 30 percent of import prices and as much as 40 percent of export prices for agricultural products in Burundi (AfDB 2009). Slow growth and low productivity, particularly in the agriculture sector, which is the primary source of employment and livelihood for the majority of the population, have resulted in persistently high levels of poverty.11 To date, Burundi has one of the highest levels of poverty incidence in the world. The proportion of the population below the $1.25/day poverty line declined only slightly from 84.5 percent to 81 percent between 1990 and 2005. By comparison, the poverty rate in Uganda declined from 69 percent 11. Agricultural value added per worker declined from $166.7 (in constant 2000 dollars) in 1992 to $99 in 2005 (World Bank 2005).

114

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

to 51 percent during the same period. In terms of human development, Burundi in 2010 was ranked as 166th out of 169 countries (UNDP 2010). The development of human capital was hindered by the exclusionary policies that marginalized the majority of the population, in addition to slow growth. By keeping large sections of society out of the reach of the educational system and other socioeconomic opportunities, Burundian leaders ultimately engineered a form of politically induced poverty. Wealth and employment creation to reduce poverty will require a change in the mindset of the leadership. To some extent, the latter will have to accept losing some control over the economy by encouraging the expansion of the private sector and empowerment of private entrepreneurs because the public sector seems to have reached its job- generation capacity. In turn, private- sector development will require a massive mobilization of financial resources to support investment. Indeed inadequate access to finance has been identified by firms as a key constraint to investment in business environment surveys (see table 3B.1 in appendix B). This chapter argues that despite the high levels of poverty in the country, there is substantial untapped potential in savings mobilization that may be harnessed to the extent that the financial system is efficiently organized, managed, and regulated. The issue of efficiency of the financial system is therefore central to the objectives of increasing growth and reducing poverty in Burundi. 3.3

Structure and Characteristics of the Financial Sector

The Burundian financial sector is dominated by commercial banks and includes a handful of formal nonbank financial institutions, mainly development banks and a growing microfinance network. The insurance and pension sector is underdeveloped, which is an impediment to resource mobilization and maturity transformation. As the country does not have a stock market, this section focuses on commercial banks, development banks, and microfinance institutions. The discussion begins with a brief account of the financial liberalization experience of the 1980s. 3.3.1

The Financial Liberalization Experiment

In an effort to improve the efficiency of the financial system, the government initiated its liberalization in 1987 in the context of the second phase of the Structural Adjustment Program (SAP).12 The period leading to the adjustment program was characterized by strict controls of interest rates and credit allocation across sectors. In addition, some financial institutions 12. The first phase of the SAP, implemented in July 1986, focused on: (a) trade and industrial policy, (b) privatization and restructuring of state- owned enterprises, and (c) agricultural policy. The second phase extended the liberalization program to other activities: export promotion, the labor market, and the financial sector. In addition, the program began to consider the social dimensions of the adjustment program.

The Financial Sector in Burundi

115

were granted monopoly in the mobilization of deposits (especially from parastatals) and the allocation of credit (especially for coffee- trade financing). Details on the key controls prior to financial liberalization and the chronology of the reforms are provided in appendix A (boxes 3A.1 and 3A.2; also see Nyamoya [2004]). The liberalization of the financial system was meant to correct market distortions and create a level playing field, improve resource mobilization, and achieve efficient allocation of resources across sectors. The results fell below expectations. On the positive side, liberalization succeeded in opening up the sector to entry of new banks. In 1987, the financial system was comprised of only three commercial banks, four nonbank financial institutions (including Caisse d’Épargne du Burundi [CADEBU]), and a nascent network of microfinance institutions. The sector was dominated by the state and public enterprises that held the lion’s share of the assets. Following liberalization, new commercial banks were established, and some of the previously protected institutions succumbed to competition (CADEBU, CAMOFI). Over time commercial banks were partially privatized, though the government kept a relatively strong influence in their management. Despite these positive developments, the effects of financial liberalization remained limited. Until 2004, the central bank continued to exercise control over credit allocation by imposing ceilings on credit disbursed by each institution and on credit supply to selected activities such as trade of coffee, tea, and cotton. At the same time, monetary policy was marked by lax control vis-à-vis prudential regulation and reserve requirements. There was also a lack of coordination in the management of liquidity and foreign exchange (IMF 2005). The reforms undertaken starting in mid- 2004 included the abolition of credit ceilings, the abolition of the discount rate as a tool of monetary policy in April 2005, and the adoption of a systematic method of liquidity management as a means of controlling money supply. In addition, the central bank began to strengthen prudential regulation and banking supervision. Overall, financial liberalization failed to correct the underlying structural deficiencies in the system. The increase in the number of institutions did not translate into an increase in savings and lending nor did it reduce the interest rate margin. While both the lending and deposit interest rates increased during the liberalization period, the lending rate rose faster, resulting in higher interest rate margins. Moreover, when interest rates declined in the early twenty- first century, the deposit rate declined faster than the lending rate, resulting in an increase in the spread (figure 3.4). Furthermore, despite the removal of interest rate controls, credit allocation did not improve. Actual interest rate setting showed preference for high turnover activities and the bulk of credit continued to go to import and export activities and the public sector. Several factors contributed to the limited effectiveness of financial liberal-

116

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Fig. 3.4 Interest rates in Burundi in the postliberalization period: Deposit, lending, and spread Source: Central Bank of Burundi database.

ization. They mainly relate to institutional deficiencies, structural features of the economy, and inappropriate implementation of the reforms. One important shortcoming was that the reforms were implemented over a very short period (eighteen months), which did not allow for the development of the institutional capacity to manage the new system. It was impossible for the existing monetary institutions to absorb such wide- ranging reforms in such a short period of time. As a result, there remained very strong temptations to revert to the prereform system. As recently as June 2010, for example, the central bank changed its relatively liberal foreign exchange policy adopted several years ago by reimposing new controls.13 The measure was revoked after a public outcry. The failure of financial- sector reforms reflected the failure of state institutions, which were serving interest groups rather than national interests. Financial- sector liberalization was expected to drastically reduce the pervasive presence of the government in the sector and its associated rents. Excessive controls over credit and foreign exchange allocation, interest rate repression, and high default rates on loans to the public sector and to politically connected individuals were channels used to extract rents from the sector. As liberalization would naturally remove these privileges, it is not surprising that the reforms were resisted and did not produce the intended results (Ngaruko and Nkurunziza 2006). 13. For details, see: http://www.brb- bi.net/se/docs/rglt_chge_scn.pdf.

The Financial Sector in Burundi

117

The limited and delayed results of financial- sector reforms were also due to the economic downturn in the wake of the debt crisis of the 1980s and 1990s. Financial intermediaries retreated even further from long- term lending and borrowers shied away from long- term investment. While inefficient financial intermediation undermined economic activity, weak economic activity also constrained financial deepening. 3.3.2

The Commercial Banking Sector

Currently, the banking sector comprises eight commercial banks, including Diamond Trust Bank, which opened in 2009. Table 3.1 gives the key characteristics of the banks. The banking sector is highly concentrated with the two mature banks, the Banque de Crédit de Bujumbura (BCB) and the Banque Commerciale du Burundi (BANCOBU) accounting for a commanding share of the market. These two banks account for 43 percent of deposits, 42 percent of total assets, and 42 percent of credit allocated in 2008. Together with the Interbank Burundi (IBB) created in 1992, the three largest banks represented 76 percent of total assets, 74 percent of credit, and 79 percent of deposits in 2008, as well as most bank branches in the country. State ownership in the banking sector is low, representing only 3.6 percent of total capital of commercial banks. However, the government still has substantial influence in the banking sector through its public entities that own up to 31.6 percent of the capital of all banks combined. The government is also a majority shareholder in two out of the three most important banks (BANCOBU and BCB). Hence, the government is still able to influence the management of banks through the nomination of its representatives to the board of directors. The government’s presence also has implications on the allocation of credit, directly through borrowing by state entities and indirectly through political pressure on bank management. 3.3.3

Nonbank Sector: Development Banks and Microfinance Institutions

Development Banks Created in 1964, the Banque Nationale pour le Développement Economique (BNDE) is the only genuine development bank with a statutory mandate to finance economic development. In particular, the BNDE contributes to the financing of small and medium enterprises and microfinance operations. However, several constraints hamper BNDE’s ability to accomplish its mission. The most important constraint is the shortage of stable long- term resources. As a public institution, BNDE relies primarily on donor funding through the government. Consequently, BNDE’s lending capacity is adversely affected by volatility and unpredictability of donor funding. In the past, BNDE also relied on direct refinancing via an

4 24

2002 1992

 

64

3

8

1996

1983

10 10

1964 1964

0

0

0

0

3 10.6

4

State’s share (%)

35.6

0

0

0

75 45

47.9

Public share (%)*

559,069 471.8

28,657

187,630

41,516

32,723

94,161 143,122

31,703

Total assets (million BIF)

Note: The 8th bank is Diamond Trust Bank, which was created in 2009; no data could be obtained on this bank. * Public share: Shares owned by state-owned firms (public enterprises).

TOTAL (million BIF) TOTAL (equivalent in USD)

7. SBF—Société Burundaise de Financement (ECOBANK starting in 2008)

5

Branches

1988

 Year of creation

Characteristics of commercial banks in Burundi, 2008

1. BBCI—Banque Burundaise pour le Commerce et l’Investissement 2. BANCOBU—Banque Commerciale du Burundi 3. BCB—Banque de Crédit de Bujumbura 4. BGF—Banque de Gestion et de Financement 5. FINBANK—Finalease Bank (taken over by Access Bank Nigeria) 6. IBB—Interbank Burundi

 

Table 3.1

415,746 350.6

17,335

148,084

28,943

22,703

66,028 114,548

18,103

Deposits (million BIF)

280,106 226.8

13,259

90,262

21,440

20,141

47,840 69,322

17,840

Loans (million BIF)

67

76

61

74

89

72 61

99

Loan /deposits (%)

1,409

107

350

93

146

275 286

152

Employment

The Financial Sector in Burundi

119

automatic rediscount facility at the central bank, which was abolished in the context of monetary policy and financial- sector reforms. The lack of stable long- term resources forces BNDE to concentrate on short- term and medium- term lending, and on commerce to the disadvantage of agriculture and industry. This has induced BNDE to focus on the same market segments as commercial banks. The other financial institution that participates in development financing is the Fonds de Promotion de l’Habitat Urbain (Fund for the Promotion of Urban Housing [FPHU]) which specializes in urban housing. Although it supplies a larger amount of loans than BNDE (table 3.2), FPHU is also a public institution confronted with the same challenge of limited access to long- term stable resources. Thus, FPHU is unable to meet the needs of the expanding urban population. Another constraint that limits lending by development finance institutions is the low purchasing power of borrowers. This is primarily due to the stagnation of nominal wages combined with drastic increases in the cost of inputs, especially construction material. Recently the government raised salaries of civil servants in some line ministries, including justice, education, and state inspection. But the wage increases remain inadequate to catch up with the rise in the cost of living and construction costs. To illustrate the deterioration of workers’ purchasing power and hence their limited access to financial resources, we consider the case of a married couple of two university- degree holders employed in the civil service. We assess their ability to service a fifteen- year mortgage at the mortgage interest rate of 18 percent for a modest 10 m × 12 m house. The salary in the civil service for a university degree laureate increased from about 30,000 Burundi francs (BIF) in 1993 to 100,000 BIF in 2010.14 This amounts to a 233 percent increase in nominal wage, but a 58 percent decline in purchasing power, adjusting for inflation. In 2010, housing construction costs in middle- income suburbs of Bujumbura (e.g., Kanyosha) were about six times higher than in 1993. The calculations in table 3.3 show that while the couple labored to cover the mortgage with 80 percent of their combined salary in 1993, in 2010 the mortgage payment is completely out of reach, representing 178 percent of the couple’s combined nominal monthly salary! These simulations show that today the Burundian workers face a double tragedy: they qualify for less credit and the little credit they can secure buys them even less on the market. While the observed recent decline in interest rates is desirable, real improvement in access to finance would require a sizable increase in workers’ income. 14. The example considered here overestimates the real repayment capacity of the household. In particular the maturity of mortgage loans is typically less than fifteen years, as is assumed here. Moreover, only some sectors in the civil service offer a monthly salary of 100,000 BIF following recent wage increases (education, justice, and state inspection).

100 100

1991 1980  

100

1962

 

82

1990

Sources: Annual reports of these financial institutions and the central bank. Note: For RNP: 2006 figures; INSS: 2007 figures; n/a = not applicable.

TOTAL

65.5

State share (%)

1967

Year of creation

Characteristics of nonbank financial institutions in Burundi, 2008

1. BNDE Banque Nationale pour le Développement Economique 2. FPHU Fonds de Promotion de l’Habitat Urbain 3. INSS Institut National de Sécurité Sociale 4. RNP Régie Nationale des Postes 5. MFP Mutuelle de la Fonction Publique

 

Table 3.2

80,762

17,933

6,011

22,572

13,137

21,109

Total assets (million BIF)

31,286

n/a

n/a

0

20,486

10,800

Loans (million BIF)

n/a

n/a

0

156

51

Loans/assets (%)

4,992

524

800

3,488

52

77

Employment

121

The Financial Sector in Burundi Table 3.3

Cost of housing construction versus civil service wages: precrisis compared to 2010 2010

Elements of housing costs Lending interest rate (annual %) Unit cost of construction (per square meter) Cost of a 8 m × 10 m house (BIF) Income and mortgage payment capacity Monthly payment (BIF) Salary of couple of two BA holders (BIF) Monthly payment/salary (%)

Change (%)

1993

Nominal

Real

Nominal

Real

14 30,000 3,600,000

18 184,000 22,080,000

9.7 23,018 2,762,141

28.6 513.3 513.3

–30.7 –23.3 –23.3

47,942 60,000 79.9

355,581 200,000 177.8

44,482 25,019 177.8

641.7 233.3 122.5

–7.2 –58.3 122.5

Source: The information on housing costs is from the Fonds de Promotion de l’Habitat Urbain (the 2010 value is obtained by applying the inflation rate of 8.3 percent, a conservative assumption, to the 2009 value of BIF 170,000/square meter). Information on the interest rate and the price index is from the Central Bank of Burundi. At the 1993 base, the implicit consumer price index used to deflate nominal values to real values in 2010 is 799.38.

Microfinance, a Relatively New Phenomenon Microfinance is relatively new in the Burundian financial system. Apart from BNDE’s microfinance operations dating from the 1960s, genuine microfinance began with the creation of the savings and credit cooperatives (Coopératives d’Epargne et de Crédit [COOPECs]) in 1985. However, starting from the mid- 1990s, many institutions were created with diverse legal status, ranging from non- governmental organizations (NGOs) to cooperatives. In addition to COOPECs, as many as nineteen organizations have been created since 2000, of which five were created in 2005 alone. Microfinance institutions (MFIs) have experienced rapid growth over recent years (table 3.4). The increase in the cost of living and the deterioration of purchasing power due to the war and the economic crisis have made it increasingly difficult for people to survive on regular wage incomes. At the same time, formal banking services have become increasingly inaccessible. The explosion of microfinance can therefore be interpreted as an attempt to fill a financial intermediation vacuum. The Finance Ministry adopted the microfinance law in 2006, whose objective is to protect savers and borrowers while minimizing risk taking by MFIs. A clearly defined legal framework is indispensable for the development of microfinance. Nonetheless, evidence from countries that have been successful in this area indicates that, more than the formal legal framework, it is the ability of MFIs to create an environment of trust between institutions and clients that determines the success of MFIs. A well- known example is the case of the Grameen Bank (see Yunus 2003). While increasing access to financial services for clients, such a strategy also contributes to financial sustainability of the MFIs by improving loan recovery.

122

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Table 3.4

Summary indicators of microfinance institutions in Burundi, 2004–2009

Indicators Members/clients Loans issued (BIF) Outstanding loans (BIF) Active borrowers Average loan (BIF) Savings (BIF) Number of depositors Service posts Employees

12/31/2004

12/31/2009

Percent change 2004–2009

272,340 9,603,149,000 13,897,427,000 52,955 181,345 12,067,087,787 n/a 138 352

430,842 40,632,884,853 41,270,650,703 166,366 244,238 33,282,113,196 384,609 184 926

58.2 323.1 197.0 214.2 34.7 175.8 n/a 33.3 163.1

Source: Réseau des Institutions de Microfinance (RIM). Note: BIF = Burundi franc.

The success of microfinance rests on the ability to navigate the complexity of the “triangle of microfinance,” which calls for attention to not only outreach to the poor (both breadth and depth of outreach), but also financial sustainability of the institutions as well as impact on growth and poverty (Zeller and Meyer 2002; Robinson 2001). Burundian microfinance institutions face several constraints in their attempts to reach this triple objective. The key constraint is the lack of stable resources, forcing MFIs to both ration credit and charge high interest rates, hence making it difficult to achieve sufficient outreach. Interviews with BNDE senior management reveal dwindling support from donors, especially since the early 1990s at the beginning of the civil war. The MFIs also face critical capacity constraints due to the shortage of experienced experts in the field. This exposes MFIs to credit risk, notably due to inefficient credit assessment and weak loan recovery mechanisms. Nonetheless, the sector has substantial potential to increase access to finance for a large part of the population and thus contribute to poverty reduction. 3.4

Credit Allocation and the Contribution of Banks to Economic Activity

This section analyzes credit allocation to the private and public entities and across economic sectors. It also discusses the term structure of credit and derives some indicators of inefficiency of sectoral and temporal allocation of credit. 3.4.1

Credit by the Banking Sector

Credit from the banking sector to the economy is very limited in Burundi. As figure 3.5 shows, over the period 1980– 2008, domestic credit from the banking sector represented 27 percent of GDP on average per year, which is

The Financial Sector in Burundi

Fig. 3.5

123

Ratio of domestic credit to GDP in Burundi and SSA, 1980–2008

Source: Based on data from the World Bank, World Development Indicators (2010).

less than half of the average for sub-Saharan Africa (67.5 percent of GDP) and only 17 percent of the ratio in high- income Organisation for Economic Co- operation and Development (OECD) countries (160 percent of GDP).15 Credit to the private sector is even smaller. At 17 percent of GDP, it represents one- third of the average ratio for sub-Saharan Africa (51 percent of GDP); the latter compares very poorly with the figure for high- income OECD countries at 126 percent of GDP (World Bank 2010).16 These statistics suggest that even by the poor African standards, the contribution of Burundi’s banking sector to economic activity in terms of credit provision is very limited. In addition to the relatively small amount of credit available, financial resources are not efficiently allocated in the sense of meeting the needs of the economy. First, a relatively important share of credit is allocated to the government. Second, the sectoral allocation of credit does not reflect the economic importance of the sectors of the economy. Third, there is a mismatch between the term structure of bank loans and investment demand. These issues are elaborated further below. 15. The apparent increase in the credit/GDP ratio since the mid- 1990s is not necessarily an indication that the amounts of credit to the economy increased. Most of the period post- 1993 was characterized by negative economic growth rates (as discussed earlier), which could explain the increase in the ratio to GDP without an increase in the flows of credit to the economy. 16. The drop in 2008 is most probably the result of the global economic and financial crises.

124

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

3.4.2

Credit to the Government

Burundian banks allocate relatively more credit to the central government than in the rest of sub-Saharan Africa. On average, bank credit to the government represents about 38 percent of total credit, compared to 25 percent in the rest of Africa. Although governments are not necessarily wasteful,17 cross- country evidence has shown that higher state ownership of the economy—as is the case in Burundi—is positively associated with high capital misallocation (Wurgler 2000; Khwaja and Mian 2005). In turn, capital misallocation leads to low total- factor productivity and output per worker, as is typically the case in developing countries (Hsieh and Klenow 2009; Bartelsman, Haltiwanger, and Scarpetta 2009).18 In Burundi, as discussed in the previous sections, financial resources were allocated to SOEs on the basis of political considerations, resulting in very low productivity. The managers of the SOEs drove some of them to collapse, and often used assets stolen from these firms to create their own private companies. While credit to public institutions is misused, private firms are severely credit constrained. According to the World Bank’s Doing Business report, about half of Burundian firms identify finance as a major constraint, ranking second only to the lack of electricity (table 3B.1 in appendix B). Evidence from a detailed firm survey carried out in the 1990s also shows that access to credit is a constraint to firm growth and investment in Burundi (Bigsten et al. 2003). Hence, the crowding out of the private sector and inefficient use of credit by the government has negatively affected the performance of the private sector. The financing of inefficient state- owned enterprises displaced valuable resources from productive investment. In the 1980s, as much as 30– 35 percent of gross domestic investment went into state- owned enterprises, but the sector accounted for less than 10 percent of total output (Ngaruko and Nkurunziza 2006).19 3.4.3

Allocation of Credit to Economic Sectors

Sectoral misallocation of credit is the second source of inefficiencies. Given the importance of agriculture in terms of employment creation, food 17. In India, for example, an analysis covering the period 1986– 2000 found that although private banks were more productive than public banks due to technical progress, the latter were more efficient than the former (Sensarma 2006). 18. According to Hsieh and Klenow (2009), capital misallocation could explain 30 percent to 50 percent of TFP difference between Chinese and American firms, and 40 percent to 60 percent of TFP difference between Indian and American firms. 19. The link between capital misallocation and slow economic growth is observed in other developing countries as well. In Pakistan, for example, politically connected firms borrow 45 percent more than other firms and their default rates are 50 percent higher. Such preferential treatment, practiced solely by government- controlled banks, costs the economy between 0.3 percent and 1.9 percent of GDP every year (Khwaja and Mian 2005).

The Financial Sector in Burundi

125

Fig. 3.6 Sectoral allocation of commercial banks’ credit in Burundi (percentage of total) Source: Based on data from BRB, annual reports. Note: Banks did not provide data on the sectoral distribution of their credits from 1995 until 2002, probably as a result of the war that raged in this period. Since 2003, the data provided follows a different classification, but our sectors of interest—namely agriculture, industry, and commerce—are clearly identified. In this figure, commerce includes the coffee-trade sector.

supply, and production of inputs for other sectors, agriculture should receive the lion’s share of financial resources. This is not the case, as figure 3.6 shows. Clearly, the allocation of credit does not reflect Burundi’s development priorities as articulated in the government’s medium- term objectives: (a) 5 percent annual rate growth of the agricultural sector, (b) reduce the rate of people with insufficient food or unbalanced diet from 84 percent of the population to 20 percent, and (c) reduce the rate of poverty from 67 percent to below 50 percent (République du Burundi 2007).20 In fact, the government of Burundi considers that the performance of the agricultural sector will not only determine the growth of other sectors, but also economic development in general. In spite of its importance for the national economy, agriculture not only attracts an insignificant amount of credit but the share has declined over time, from 2.5 percent of total credit in the period 1980– 1994 to 0.75 percent 20. The rate of poverty incidence used here is the $1 a day measure, which explains the difference with the statistic in table 3.1 that uses the $1.25 a day measure.

126

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

in the period 2003– 2008. Yet in 2005, agriculture contributed 42.4 percent of GDP and employed 84 percent of the active population (ISTEEBU 2008; République du Burundi 2008). Moreover, agriculture is the main source of economic growth. According to some conservative estimates, a 10 percent increase in agricultural production, excluding coffee, leads to a 3.5 percent increase in GDP (Lim and Rugwabiza 2009). In general, agriculture- led growth has been shown to have the highest impact on poverty reduction (World Bank 2008). Therefore, the resources allocated by the banking sector to agriculture do not reflect the economic importance of the sector, preventing Burundi from reaching its potential in terms of growth and poverty reduction. The neglect of the agricultural sector, despite the large number of people who depend on it, is yet another illustration of a narrow- based system of governance that was not interested in the well- being of the majority of the population. The decline in credit allocation has been even more dramatic in the industrial sector. The share of credit to this sector collapsed from 16 percent of total credit in 1980– 1994 to only 2 percent in 2003– 2008. In contrast, commerce, including coffee trade, is the most preferred sector, although it represents only 6.8 percent of GDP and accounts for 2.5 percent of total employment (République du Burundi 2008). The sector absorbed 67 percent of credit in 2003– 2008, up from 43 percent in 1980– 1994. These statistics call for a number of observations. First, the economic transformation needed for the country to achieve its development priorities, notably poverty reduction, requires massive investments in agriculture and industry. However, the current allocation of financial resources makes this objective hard, if not impossible, to achieve. Second, the excessively high concentration of credit on one sector, trade, increases bank vulnerability. Negative shocks to the trade sector, particularly its import- export segment, can severely undermine the stability of the banking sector. Indeed, when Burundi was placed under a total economic embargo from July 31, 1996 to January 23, 1999, this affected the portfolio of banks, as traditional bank clients, particularly those relying on import and export activities, saw their activities seriously curtailed (World Bank 1999). This may explain the drop in lending interest rates during this period (figure 3.1) and the decline in credit to the private sector in 1997 and 1998 (figure 3.5).21 Third, it could be argued that credit allocated to trade has an indirect positive effect on agriculture and the rural economy if it finances trade of agricultural inputs and outputs, and the provision of agriculture- related services, leading to job creation. In fact, a dynamic agricultural sector is often associated with high rural nonfarm activity (Nkurunziza 2007). In Burundi, due to the rudi21. The actual decline in credit to the private sector was more pronounced than shown by the GDP ratios in figure 3.6. The reason is that the rates of GDP growth in 1996 and 1997 were negative (– 8 percent and – 2 percent, respectively).

The Financial Sector in Burundi

127

mentary state of technology in the agricultural sector and the dominance of informal subsistence activities, agriculture is detached from other sectors of the economy. Hence, the small share of trade activities in GDP and employment creation, as shown earlier, suggests that the indirect effects of trade on agriculture and the rural economy are very limited. The picture depicted above suggests that there are sectoral allocative inefficiencies of credit. These inefficiencies are analyzed from ex-ante and ex-post perspectives. The ex-ante analysis compares the distribution of credit to a predetermined allocation rule. From this perspective, resources should be allocated to sectors where their marginal effect on development objectives is highest. In the case of Burundi, credit would produce more positive effects on development if it were mainly allocated to agriculture and industry. The ex-post analysis compares the actual distribution of credit to the distribution of an ex-post measure of risk with the assumption that more resources should normally be allocated to sectors with lower risk. To derive proxy measures of these inefficiencies, two indexes are computed on the basis of the sectoral allocation and term structure of risk. They measure the gap between the actual distribution of credit and the risk- adjusted distribution. If credit were allocated on the basis of the level of risk, the rate of default (amount in default relative to current credit) should be equal to the sector’s proportion in total credit. For example, the rate of default in the agriculture sector should be equal to the proportion of credit to agriculture in total credit. The index of allocative inefficiency is the ratio of the two proportions (multiplied by 100). If the index is equal to 100, the amount of credit allocated to the sector reflects the risk level in the sector. If the index is less than 100, the interpretation is that credit is too low considering the level of risk. Conversely, an index greater than 100 indicates that the allocation of credit to that sector is too high relative to the level of risk. Figure 3.7 confirms that it is riskier to lend to commercial and industrial activities than to agriculture. With a value of about 20, the allocative inefficiency index in agriculture means that too little resources are invested in the agricultural sector given the level of risk in the sector. In contrast, commerce has a value of 157, implying that the sector receives 57 percent more resources than warranted by the level of risk. Why do agriculture and industry in Burundi attract so little credit? The main reason is related to the political economy of agricultural development. Burundi’s agriculture is dominated by smallholder farmers who have little political voice to lobby politicians in order to defend their interests (Bates 1981; Nkurunziza and Ngaruko 2008). Moreover, 94.5 percent of all agricultural activities in Burundi are in the informal sector, which is typically cut off from modern financial services (ISTEEBU 2008). The lack of collateral and the high cost of loans to informal economic activities in the rural areas put agriculture at the fringes of the financial sector. Also, the fact that agriculture is mainly rain fed makes it vulnerable to weather shocks. This,

Fig. 3.7

Index of sectoral allocative inefficiencies of credit, 2008 (in percentage)

The Financial Sector in Burundi

129

combined with rudimentary production technologies, render agricultural production uncertain. Therefore, even if banks had the capacity to intervene, they might shy away from financing projects in the agricultural sector due to high uncertainty. Finally, over the last few years, banks have been so profitable that they have had no incentive to lend to activities perceived to be too risky.22 The low proportion of credit to industry may be explained by low profitability, in addition to uncertainty. The demand for credit to industrial activities is very low because the investment climate is poor (appendix B). Moreover, as discussed earlier, high transport costs and unreliability of input supply substantially increase production costs, reducing profitability and the risk- adjusted rates of return on investment. Furthermore, financing industrial activities requires medium- to long- term loans, but banks in Burundi have a strong preference for short- term lending. Hence, the mismatch between the needs of the industrial sector and banks’ lending capacity and preferences contributes to explaining the low level of credit to the industrial sector. We should acknowledge the possibility of endogeneity. On one hand, agriculture and industry may attract low credit because they are not developed; on the other hand, the two sectors are not developed maybe because they do not attract credit. A poor country like Burundi, which is heavily dependent on agriculture, ought to have an explicit policy to channel financing into the sector and, eventually, build a basis for industrial development. In Burundi, agricultural development is too important and too strategic to be left to market forces alone. 3.4.4

Term Structure of Bank Credit

In addition to the lopsided sectoral allocation of credit, its temporal allocation is also incompatible with the long- term needs of industry as well as the need to build a basic production infrastructure, as discussed earlier. The term structure of credit is characterized by the predominance of shortterm credit (figure 3.8), which hampers resource mobilization for long- term development projects. To illustrate, assuming that the $4.6 billion needed over the next twenty years to build Burundi’s infrastructure would be evenly spread over twenty years,23 $230 million would be needed every year. Even if all credit disbursed in 2008 were allocated to the implementation of these infrastructure projects, it would not be enough to cover the needs (see table 3.1). Adding to the basic infrastructure financing the investments needed to upgrade technologies in the agricultural and industrial sectors, as well as other needs such as consumer credit, it is clear that the financial sector in 22. The issue of bank profitability is discussed in some detail later. 23. This is a minimalist scenario because a large part of investments of this nature have to be frontloaded.

130

Fig. 3.8

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Term structure of credit in percentage of total credit

Source: Data from BRB, annual reports. Note: Short, medium, and long term refer to periods of less than one year, one to less than three years, and equal to or more than three years.

Burundi does not have the capacity to meet the country’s needs in development financing.24 The share of short- term credit in total domestic credit has steadily increased since the mid- 1990s from 67 percent to over 80 percent thereafter. The share of short- term credit reached its highest values during the war period (1993– 2003), with a peak of 83.5 percent in 2002. Most of this increase was in the form of working capital to firms faced with major cashflow difficulties. The increase in short- term credit was at the expense of long- term credit. The latter declined from about 17 percent of total credit in 1995 to less than 3 percent in 2007. Medium- term credit oscillated between 10 percent and 21 percent of total credit over the sample period, with an increasing trend since 2000 (from 12 percent to 21 percent of total credit). The investment needs underlying Burundi’s development objectives require scaling up of resources on medium- term and long- term credit. Analyzing the term structure of credit in light of the risk associated with different terms shows that there are also temporal allocative inefficiencies (figure 3.9). Despite banks’ concentration on short- term credit, these loans are riskier than long- term credit. Medium- term loans are associated with 24. Another way of illustrating the limited capacity of the financial sector in Burundi to raise the resources required for the country’s development is the fact that the volume of gross fixed capital formation is greater than total bank credit. According to IFS data, in 2008, it was 284.9 billion BIF or 101 percent of total bank credit.

The Financial Sector in Burundi

131

Fig. 3.9 Term structure of allocative inefficiencies of credit, average 2003–2008 (in percentage)

the lowest level of risk but account for one-fifth of total loans. Keeping the risk level constant, medium- term loans should be about five times higher than the current level. Why do banks concentrate on short- term loans despite the inefficiencies associated with short- term lending? According to our interviews with bank managers, this is justified by the lack of long- term resources in their portfolio. This is correct to some degree. Incentives for saving are weak, as illustrated by low savings interest rates relative to lending rates (figure 3.4). However, the lack of long- term resources does not fully explain limited long- term lending because the lending pattern does not reflect the term structure of available resources. According to data from the central bank, between 2003 and 2007 short- term bank loans were 110 percent of shortterm deposits per year, on average. In contrast, medium- and long- term loans represented only 53 percent of medium- term and long- term savings, each. This means that medium- term and long- term savings are used to finance short- term loans, implying a bias against medium- and long- term lending.25 25. For further investigation it would be interesting to compare the case of Burundi to other countries with regard to the “transformation ratio,” that is, MT & LT loans/MT & LT savings.

132

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

This bias may be explained as follows. First, most of the period under analysis was characterized by extreme political and economic instability, translating into high inflation, currency devaluations, and high interest rates (figure 3.1).26 The resulting political and economic uncertainty and risk induced financial institutions to be extremely cautious in their lending practices, privileging short- term loans. Second, the small proportion of long- term lending could be the result of limited demand due to high costs. As figure 3.4 shows, the liberalization of interest rates in the late 1980s resulted in their steady increase, making it difficult for firms to borrow and invest profitably. Third, the steady increase of money supply in the context of a shrinking economy over the 1990s and early twenty- first century (figure 3.1) contributed to increasing inflation, discouraging profitable investment and borrowing. The long- run semielasticity of inflation to real money in circulation trebled between the prewar to the war period (Nkurunziza 2005). Fourth, the industrial organization of banking where competition is limited allows financial institutions to extract maximum rents from the public. This enables them to generate high profits without the need to widen their market and take more risk. Fifth, banks are reluctant to take risk due to the absence of adequate information on prospective borrowers. The lack of credit bureaus, timely and reliable reporting of company financial statements, and modern systemic audits of firms explains the high risk aversion of banks. Despite these constraints, there are untapped opportunities for the mobilization of long- term domestic financial resources. If a fraction of the sizable profits of commercial banks (see section 3.5) and the major private and semipublic companies—which are currently held as cash—were pooled to constitute an investment fund, they would provide important long- term investment resources that are currently lacking (Nyamoya and Nkeshimana 2005). In 2004, for example, the combined profits of the eight commercial banks, the two development banks (BNDE and FPHU), the largest insurance company (SOCABU), and two semipublic companies (BRARUDI and SOSUMO)27 amounted to 18.8 billion BIF, which represented 15.7 percent of the country’s gross capital formation in that year. If half of these funds had been committed to investment, the country’s average gross capital formation would have increased from 15.3 percent of GDP to 23.1 percent of GDP. 26. Political instability led to frequent changes of political leaders, which created instability in the private sector as new political leaders established their own connections in the private sector in order to take advantage of the opportunity of using the state for personal gain as a strategic supplier and client. Between October 1, 1987 and January 11, 2000, Burundi had a new government every nine months; over the last three years, there has been a change in government every six months, on average. 27. The SOCABU = Société d’Assurance du Burundi (an insurance company); BRARUDI = Brasserie et Limonaderie du Burundi (a brewery); and SOSUMO = Société Sucrière du Moso (a sugar production and processing company).

The Financial Sector in Burundi

3.5

133

Performance of the Financial Sector and Access to Credit

The analysis of the performance of the financial sector in Burundi presents a contrasting picture. On one hand, the analysis in the previous section has shown that banks inefficiently allocate their resources. On the other hand, individual banks are highly profitable. In fact, commercial banks’ choice to concentrate resources on one segment of economic activity, namely commerce, is probably the reason why they are so profitable. The first part of this section discusses the profitability of banks. The second argues that the high profitability coexists with a high level of fragility of the banking system. The last part discusses the challenge of accessing credit in Burundi. 3.5.1

Banking Profitability

The banking sector in Burundi is highly profitable by standard measures of return to investment. The average return on equity ratio stands at 19.4 percent, with returns as high as 53 percent for BANCOBU and 39 percent for BCB (table 3.5). This high performance does not reflect the fundaTable 3.5 Bank BBCI BANCOBU BCB BGF FINBANK IBB SBF BNDE FPHU Total/average

Performance indicators of financial intermediaries in Burundi, 2008 Credit/deposits (1)

Equity (2)

Net profit (3)

Percent ROE (4)

No. accounts (5)

99 72 61 89 74 61 76 n/a n/a 67

4,690 6,413 8,816 3,433 3,757 12,404 783 6,900 5,150 52,346

796 4,171 3,401 151 1,173 3,265* –87 513 636 10,161

11.2 52.7 38.9 21.4* 24.2 26.3* –2.2 7.4* 12.3* 19.4

28,900 19,563 26,199 13,632 1,041 40,000 1,154 0 0 130,489

Source: Data from individual banks’ reports. Notes: BBCI = Banque Burundaise pour le Commerce et l’Investissement; BANCOBU = Banque Commerciale du Burundi; BCB = Banque de Crédit de Bujumbura; BGF = Banque de Gestion et de Financement; Finbank = Finalease Bank; IBB = Inter Bank Burundi; SBF = Société Burundaise de Financement; BNDE = Banque Nationale pour le Développement Économique; FPHU = Fonds de Promotion de l’Habitat Urbain. Column (1), the ratio of total credit to total deposits; column (2), the amount of equity capital in millions of Burundi francs; column (3), the amount of net profits in millions of Burundi francs; and column (4), the return on equity, which is the ratio of (3)/(2); column (5), the number of accounts opened in each bank. Note that BNDE and FPHU have neither branches nor accounts; they are not commercial banks so do not take deposits from clients. The figures on profitability are most likely underestimated as banks are not comfortable communicating the right figures would show the extent of rent extraction. This hypothesis is confirmed by an IMF-World Bank (2009) study that calculated, on the basis of more accurate data, an average return to equity of 33.15 percent in 2008. * All the numbers are for 2007.

134

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

mentals in the real economy as indicated by the low economic- growth rate (figure 3.1) and high poverty incidence. The high performance is even more surprising given the inadequate management of the financial sector and the often disruptive state intervention in the management of banks. It is more appropriate to say that the banking sector remains highly profitable despite serious institutional and structural constraints. The question is, therefore, what explains the high performance of Burundian financial intermediaries as business firms? High profitability of financial intermediaries in Burundi may be explained by several factors. The first is the oligopolistic nature of the banking sector, which facilitates rent extraction. As earlier noted, three commercial banks, namely BANCOBU, BCB, and IBB, control the banking sector in Burundi.28 They have implicitly divided up the market so they do not need to compete to attract clients. If there was competition, some banks would charge lower interest rates and fees and register lower but still comfortable profits. Banks extract rents from their clients through high lending interest rates and charges as well as low savings interest rates.29 Banks use deposits and savings, which are poorly remunerated, to lend at very high rates that do not reflect the cost of these funds. On average, the cost of funds to banks declined from 4.8 percent to 2.8 percent between 2005 and 2008.30 This low cost is mainly due to the fact that banks do not pay interest on short- term deposits, which represent more than 50 percent of total deposits from clients (IMF and World Bank 2009). In contrast lending rates are very high (see figure 3.4), even though they have declined over the last few years to reach 16 percent in 2010. The question is whether the large interest rate spreads reflect the actual level of risk faced by banks or whether, owing to the lack of competition, the high spreads are a manifestation of rent extraction. Detailed account- level data on the credit portfolio of one of the largest banks covering the period from January 2004 to August 2009 illustrates that the rents extracted from their clients are allocated to bank executives and employees through interest- free loans or very low interest rates (table 3.7). The median interest rate on short- term and medium- term loans to bank employees is only 4 percent, almost one- fifth of the average interest rate of 28. The case of Inter Bank Burundi (IBB) is particular. It is the youngest of the three largest banks (created in 1992, several decades after the oldest two banks were created), but has outperformed its older peers in terms of equity capital, savings, and credit. This is the result of its better management. The bank is fully privately owned by Burundi- based shareholders, unlike the other two banks, which until recently were largely state controlled, but with a sizable foreign shareholding. This, in a way shows that BANCOBU and BCB could have performed better had they been better managed. 29. Interviews with management of financial institutions revealed that the production cost of a checkbook is about BIF 400– 500. However, banks sell it for BIF 4,500, ten times the production cost. This is an important source of revenue for banks with large numbers of customers such as IBB, BBCI, BCB, and BANCOBU. 30. This cost is the ratio between total interests paid by the banking sector and total deposits by clients, as well as interbank deposits.

The Financial Sector in Burundi

135

19 percent used for external clients. Cheap credit encourages bank employees to borrow. Although they represent a small number relative to the client base (the largest bank employs only about 350 persons), out of 6,182 total loans accorded, employees accounted for 17.5 percent of this number. The lack of competition among financial institutions makes banks and their employees the main beneficiaries of the resources extracted from the public.31 High rates of return on equity are also a reflection of the undercapitalization of commercial banks. Until the late 1990s, the minimum capital was set at BIF 300 million, which at the time represented less than half a million dollars. In 2004, minimum capital requirement oscillated around USD 1 million. For new banks, only one- third of this amount had to be paid up before the bank could operate, thus allowing shareholders to start collecting deposits and savings from the public and engage in lending activities. In 2006, the central bank mandated trebling of the minimum capital over a two- year period. By December 31, 2008, all commercial banks were required to have a minimum capital of USD 2.8 million (Banque de la République du Burundi 2006). The most recent requirements are that commercial banks will need to have a minimum capital of BIF 5 billion (about $4 million) by December 31, 2009, and BIF 10 billion (about $8 million) by December 31, 2010. By December 31, 2008, all commercial banks had complied with the new capital requirements. It is highly likely that the increase in minimumrequired equity capital will reduce banks’ returns on equity because profits are not expected to increase in the same proportion as capital. The high level of profitability combined with the new equity- capital requirements have contributed to attracting foreign banks to Burundi. In order to respond to the central bank’s requirement for higher capital, a number of domestic ailing banks have been forced to seek external investors and partnerships with stronger African banks. In 2008, SBF was saved by ECOBANK, a West African multinational bank with operations in thirty African countries. Finalease Bank, another bank facing difficulties, was acquired by the Nigeria- based Access Bank Plc. Other investors were attracted by the potential of making high profits as well. In 2009, Diamond Trust Bank (DTB), an East African bank belonging to the Aga Khan Group, entered the Burundian banking sector. Bank of Africa, another successful West African bank, has acquired shares in the capital of BCB, one of Burundi’s most solid banks. Other foreign banks are considering opening branches in Burundi or participating in joint ventures with existing banks. They include Kenya Commercial Bank (KCB), Union 31. The Minister of Finance acknowledged before the country’s Senate that the cartelization of commercial banking in Burundi was one of the reasons why interest rates were so high. However, the minister did not offer any government plan to address the problem. She suggested, rather, that the promotion of microfinance, regional integration, and more political stability were likely to force banks to reduce their lending rates (see http://www.senat.bi/spip .php?article1122).

136

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Bank of Nigeria (UBA), FINABANK of Kenya, Barclays Bank of the United Kingdom, and Actis, a private- equity investor in emerging economies (Lienart 2010). Burundi is currently going through the most important transformation of its banking industry. Automated teller machines (ATMs) have been introduced for the first time and it is expected that opening the market to external banks will create opportunities for innovation, which in turn will improve efficiency and increase financial deepening. New payment instruments such as mobile banking and Internet banking have already been introduced, and higher use of information and communication technology (ICT) in banking operations is expected. This new era could also see the emergence of new savings/investment instruments, such as equity funds, which are now nonexistent. These new developments have also been influenced by the country’s recent entry into the East African Community (EAC). The Burundian authorities are aware that as the East African regional market opens up, the country’s banks will have to strengthen their capital base and improve efficiency if they are to survive competition from more solid financial institutions within the region. 3.5.2

Financial-Sector Fragility and Risk of Bank Failures

The positive developments highlighted in the previous section, particularly the high returns on equity, hide a serious problem underlying Burundi’s financial sector, namely its fragility due to three main factors: (a) undercapitalization of banks, (b) state involvement and mismanagement, and (c) concentration of bank credit portfolios. Undercapitalization of Financial Banks The increase in required minimum equity capital for banks has been a positive development for the stability of the banking sector. High equity capital makes banks more resilient when faced with short- term shocks. Properly capitalized banks are also more able to credibly engage in longterm relationships with their clients and partners; this is critical given the central importance of reputation for financial institutions. However, undercapitalized banks run the risk of insolvency, which has far- reaching effects on the credibility of the financial system as a whole. In Burundi, low capital requirements have enabled a small group of shareholders to extract rents from the public while limiting their involvement in development- oriented activities. Undercapitalization of the financial system in Burundi affected banking in two ways. First, it limited banks’ lending capacity, particularly credit to large clients. Indeed, central bank regulations require that credit to one client should not exceed 20 percent of a bank’s capital (République du Burundi 2003). If enforced, this prudential requirement penalizes poorly capitalized banks. Second, the low level of capital combined with bad lending prac-

The Financial Sector in Burundi

137

tices resulted in insolvency of several financial institutions, leading to their failure.32 In this light monetary authorities should carefully watch banks’ practices, particularly if the recent increases in profitability from a return on equity of 9.9 percent in December 2004 to 29.4 percent by November 2008 are due to increases in credit disbursements. If the new competition is pushing banks to issue more credit, commercial banks will need to consider increasing provisions for bad loans beyond the legal minimum just in case these loans become nonperforming (IMF and World Bank 2009). Political Pressure, Mismanagement, and Poor Central Bank Supervision Have Led to the Collapse of Several Financial Institutions Traditionally, the state and public- sector entities have been directly involved in the creation and management of financial institutions (Chrétien and Mukuri 2002). For example, the state and its affiliated public institutions had a controlling share in BANCOBU, BBCI, BCB, Banque Populaire, CAMOFI, SBF, and others. This gave state institutions the power to nominate managers as political appointees, who often had little managerial experience and were accountable to their political backers. Poor management caused several of these institutions to collapse. We briefly discuss five cases below. Caisse d’Épargne du Burundi (CADEBU) was created in 1964 as a fully state- owned financial institution whose main role was to mobilize financial resources and allocate them to the economy through low interest rate credit. The CADEBU had the monopoly over the collection of mandatory savings from public- sector workers. In turn, these funds were used to finance low- interest loans to businesses and the public, which made securing credit from CADEBU a privilege. This provided substantial power to CADEBU managers. As a result of their own abuse of authority and political pressure, CADEBU managers extended credit to less deserving applicants while denying it to more credit- worthy projects. With the liberalization of the financial sector in the late 1980s, CADEBU lost some of its traditional privileges. Competition and bad management led to its collapse in 1992. Caisse de Mobilisation et de Financement (CAMOFI) was created in 1977 as a fully state- controlled development bank providing funding for mediumand long- term projects. Its equity capital was BIF 200 million, but it was so poorly managed that it never made substantial profits. In 1997, for example, its losses before subsidies amounted to BIF 560 million (IMF 2000), almost three times its equity capital. Accumulated debt by CAMOFI resulted in its liquidation despite several attempts by the central bank to save it through 32. Since the 1990s, five financial institutions have collapsed. They are: Meridien Bank Burundi (MBB), Caisse d’Épargne du Burundi (CADEBU), Caisse de Mobilisation et de Financement (CAMOFI), Banque de Commerce et de Développement (BCD), and Banque Populaire du Burundi (BPB). The absolute number of bank failures may appear small, but in a shallow financial system like Burundi this has had a profound effect on banking in general.

138

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

injections of cash. The firm’s poor management by a prominent politician led to its collapse in November 1998 with debts amounting to five times its equity. The failure of CADEBU and CAMOFI was mainly the result of bad management. The ruling elites perceived them as sources of rents and managed them accordingly, as was the case with other state enterprises. It is likely that the failure of these two institutions benefited their managers and their friends, as well as a few politically connected people who had been given large loans.33 It is even possible that this group of influential people could have pushed these institutions over the cliff in order to ensure that the loans they had contracted would never be paid back. Almost twenty years after the collapse of CADEBU, its liquidation process is still ongoing. While very little success has been achieved in recovering loans from CADEBU debtors, its liquidators were quick to repossess other assets, particularly real estate, which was hastily sold in obscure circumstances. The political economy of financial- sector management in Burundi requires a deeper analysis beyond the scope of the current project. Meridien Bank Burundi (MBB) was created on August 1, 1988, as a limited liability company, but with some public shareholders. It was a subsidiary branch of Meridien-BIAO, a continental network of banks with headquarters in Zambia and spanning the African continent with branches in Burkina Faso, Burundi, Cameroon, Gabon, Gambia, Ghana, Kenya, Niger, Sierra Leone, Swaziland, Tanzania, and Togo. Meridien-BIAO’s initial capital was BIF 800 million, but this amount was reduced to BIF 506.31 million in 1994 when the bank faced a severe liquidity shortage. The MBB was finally put into receivership by the central bank on May 3, 1995. External auditors called in to probe the bank’s financial position discovered evidence of systemic mismanagement, both inside MBB and within the Meridien-BIAO network. Even though MBB’s management was flawed, the main cause of its collapse was a liquidity crisis following the failure of the parent company to pay back a large loan it had contracted from MBB. Meridien-BIAO had adopted a practice of financing its investments using large intragroup loans, with no clear repayment modalities. The pan-African bank eventually collapsed. Among the different unverified theories on the reasons of this failure is sabotage from Western banks, which had traditionally controlled the African market. Even if this had been the case, it is clear that the poor management of the network played an important role in precipitating the failure of all its subsidiaries (see Wright 1995). This case is also evidence of weak banking regulation and supervision, especially in the case of crossborder banking. 33. In her communication on October 10, 2003, the first deputy governor of the central bank acknowledged that CADEBU and CAMOFI collapsed as a result of gestion laxiste, or lax management, a diplomatic term meaning that they were plundered.

The Financial Sector in Burundi

139

Banque de Commerce et de Développement (BCD) was created on January 14, 1999, as a limited liability company with a capital of BIF 1.016 billion. Paradoxically, the bank’s CEO was the same politician who had led CAMOFI to failure. His appointment was in flagrant violation of Article 17 of the banking law (République du Burundi 2003), which stipulates that a person cannot be allowed to manage a bank if he (she) has played a key role in a company which, under his (her) leadership, was declared bankrupt. This, clearly, should have applied to CAMOFI’s former CEO given that the company under his watch had collapsed just two months earlier. This exceptional treatment was not unrelated to the fact that the individual was a highly influential political figure, thus defying central bank regulations. This example illustrated the lack of independence of the central bank and its weakness in upholding the law governing banking in Burundi. As should have been expected, BCD was very badly managed. The bank lasted only four years and its problems appeared well before it went bankrupt. An audit report established that by March 23, 2004, the date on which the central bank finally decided to put BCD under receivership after a long period of inaction despite warnings that the bank was collapsing, there were severe problems that would have been difficult to address in order to save the bank. For example, to continue its operations, the bank needed to raise BIF 7.5 billion through recapitalization, loan recovery, sale of assets, and so forth. It was impossible to raise this amount of money in a relatively short period because BCD was known to be poorly managed. The audit report also uncovered several cases of fraud that give a glimpse of the internal management of the bank. For example, there was reference to advances of BIF 3.185 billion made to purchase a plot to build a branch in Quartier Buyenzi, one of the poorest neighborhoods in Bujumbura. Not only was this amount exorbitant, but also the plot was never bought. The advances had not been recovered by the time the bank collapsed and it is unlikely that they will ever be. Banque Populaire du Burundi (BPB) was established in 1992 largely with public funds through several public institutions, including the national pension fund, and the state, which contributed 15 percent of equity capital. Just three years after its creation, there were rumors that the bank was in danger of collapsing due to mismanagement. The government responded by appointing a professional banker as its new head. The BPB was back on its feet, but it eventually collapsed in 2006 when the central bank judged that BPB had failed to recover 40 percent of its loans representing BIF four billion, leaving the institution in a state of extreme fragility. The central bank launched an inquiry to assess whether BPB’s failure was the result of mismanagement or corporate malpractices. The results of this inquiry were never made public. One constant factor linking all five cases reviewed above was the failure of the central bank to play its surveillance role and make prompt interven-

140

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

tions whenever problems were detected. With respect to BCD, for example, according to interviews with officials in the financial sector, the central bank had information that BCD was in a very bad financial situation and that it should have been ordered to stop its activities at least one year before the central bank eventually intervened. Apparently, there were instructions from the highest political authorities ordering the central bank not to intervene. These cases of bank failures raise serious questions about the banking supervision and regulation framework in Burundi. There are generally two main forms of supervision of the banking industry (Hubbard 2005): (a) direct supervision by the central bank, and (b) indirect supervision by financial markets, or financial market discipline. In the case of Burundi, financial market discipline is not applicable due to the absence of an equity market that could help in pricing risk. Moreover, the information flow on the financial situation of corporations and banks is so slow that the public has no basis to judge the riskiness of banks in real time. Regulation is therefore limited to direct intervention by the central bank. Clearly, the central bank of Burundi has faced severe constraints that have limited its ability to effectively regulate and supervise the banking industry. First, effective regulation requires independence of the central bank from political interference. In Burundi, as the case of BCD illustrates, the central bank failed to intervene due to political pressure. Another constraint to effective regulation is inadequate capacity, especially in the area of information technology. In particular, the fact that bank operations are not managed by a fully digitized system precludes speedy and timely examination. Lack of adequate training for the staff responsible for banking supervision and regulation is a critical barrier to effective regulation. This constraint is exacerbated by the fast- changing nature of the regulatory framework, especially in the wake of the recent financial crisis, which has placed a premium on modernization and harmonization of national banking regulations in line with global standards. In addition to the lack of independence, probably due to the lack of technical capacity, the central bank left loopholes in monetary policy that were exploited by commercial banks to increase their profits. For example, the central bank relies on liquidity management as the main tool of monetary policy used to control inflation. Under normal circumstances, it supplies liquidity to banks that are in need of extra funds and takes liquidity from those that have excess cash. In 2001, 2002, and 2003, commercial banks borrowed from the central bank at 14 percent, 15.5 percent, and 14.5 percent interest rates, respectively. They reinvested these funds into treasury bills issued by the same central bank and earned interest rates of 19 percent, 20 percent, and 16 percent, respectively. Hence, commercial banks used public resources to lend to the government, earning up to 5 percentage points of net interest. This was not illegal at the time because the central bank had not excluded commercial banks with central bank debt to participate in the

The Financial Sector in Burundi

141

treasury bills market. The anomaly was eventually corrected. Overall, these cases show that weak capacity and lack of independence from political influence severely hamper the efficiency of banking regulation and exacerbate the fragility of the financial system. Credit Concentration Account- level information from one of the major commercial banks illustrates how banks compete to capture the biggest clients, even when this exposes them to high risk. For example, one account had a loan balance of BIF one billion on October 27, 2008, then two billion on November 24, 2008, and three billion on May 25, 2009. This last amount represented half of the bank’s equity capital. Out of 6,180 loan contracts totaling BIF 69.8 billion over the period from January 2004 to August 2009, there were 132 loans of BIF 100 million or higher. These loans represented 60 percent of total bank credit and they were held by only sixty- two accounts.34 According to resilience tests of the banking sector based on 2008 (November) data, a decline in the quality of debts owed by the five largest debtors of the banking sector could reduce the solvency ratios of four out of seven commercial banks below the legal minimum (IMF and World Bank 2009). That five debtors could have such an important destabilizing effect on a country’s banking sector is a clear indication of its fragility. As summarized in the previous discussion of bank failures in Burundi, the high concentration of loans on very few clients, often without requiring proper collateral, has been at the root of bank fragility in Burundi. There is anecdotal evidence that the failure of the banks and other financial institutions discussed earlier was mostly the result of bad lending to “large” and politically connected clients. 3.5.3

The Challenge of Access to Credit

Despite some recent positive developments in the banking industry, such as competition and the introduction of new technologies, access to credit remains a major challenge for the majority of people and firms in Burundi. Limited Access to Credit and Banking Services Burundi has an extremely low rate of bank penetration. With less than two accounts per 100 persons,35 the banking sector in Burundi is narrowly focused on a small urban elite and business community, which together represent a tiny proportion of the population. The rural economy is not covered by financial institutions despite the fact that it represents the largest part of economic activity. By the end of 2008, only seven of the 73 bank branches or 9.5 percent were located in rural areas despite an urbanization rate of only 10 percent (table 3.6). Moreover, most of the branches were in Bujumbura, 34. Several accounts had more than one loan over the period. 35. The number of accounts is from table 3.6 and the data on population size from WDI.

142 Table 3.6

Country Burundi Botswana Madagascar Mauritius Mozambique Rwanda Uganda

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya Access to financial services and ICT in Burundi and selected SSA countries, 2009 Adults using Bank formal Deposit branches financial accounts in rural services per 100 area (%) adults (%) 17 47 21 54 12 23 20

2 48.1 3.4 210.9 11.2 20.2 15.4

9.5 40.8 6 52.9 5.4 26.1 56.8

Poverty Rural Mobile head population phone Internet count (% of subscribers subscribers (at $1.25/ total) (%) (%) day) 90 40.4 70.5 57.5 63.2 81.7 87

6 78 25.3 81.4 20.2 13.6 27

0.8 4.2 1.7 29.9 1.6 3.1 7.9

81.3 31.2 67.8 — 74.7 76.6 51.5

Source: World Bank, World Development Indicators; Financial Access Initiative (online data); UNDP, Human Development Report 2009. Note: The average usage of formal financial services by adults is 20 percent in SSA, 35 percent in Latin America, 41 percent in East Asia, 42 percent in South Asia, 51 percent in Central Asia and Eastern Europe, and 92 percent in high-income OECD countries.

with the remainder located at provincial capitals. Until recently, there were provinces without any bank presence. The strong concentration of banking services on a narrow segment of the population and business community could help explain why banking in Burundi has remained underdeveloped. Only 17 percent of the adult population in Burundi uses formal financial services, compared to an average of 20 percent for sub-Saharan Africa. Clearly the country faces the challenge of providing incentives for financial institutions to increase their presence in rural areas. Access to finance can be increased substantially by promoting nonbank institutions, especially microfinance institutions, which are more flexible and better equipped to serve the informal sector and the rural areas. Microfinance covers 431,000 people (table 3.4), which is more than three times the number covered by commercial banks. Despite high poverty in Burundi, savings mobilization is below potential. This is illustrated by the success of microfinance institutions in mobilizing rural savings which, in five years, increased by 176 percent (table 3.4). Credit Allocation Skewed in Favor of “Insiders” The cost of credit is highly differentiated according to the identity of applicants. The small group of credit recipients may be divided into two categories: “outsiders” who pay high interest rates and “insiders” who pay no or very low interest rates. The distribution of interest rates in figure 3.10 distinguishes these groups. The distribution of interest rates is bimodal, with the upper part of the distribution—with the highest mode—showing interest rates paid by “outsiders” who are the regular bank clients. This group pays interest rates

The Financial Sector in Burundi

Fig. 3.10

143

Distribution of lending interest rates

Source: Based on account-level data provided by a major commercial bank.

between 18 and 23.5 percent. The lower tail of the distribution corresponds with the low interest rates paid by “insiders,” who are bank executives and employees and a small group of special “clients” who have interest- free credit. They pay interest rates between zero and 7.5 percent, with a mode of 4 percent. Other indicators also confirm that the credit market excessively favors insiders at the expense of the majority of borrowers (table 3.7). Credit to “Outsiders” At a 19 percent interest rate, the majority of borrowers cannot invest in profitable projects, explaining why the range of projects that can be funded is so limited. Moreover, the high interest rate could be associated with high default rates as borrowers may fail to generate profits that are high enough to cover debt- service obligations. Adding to this the weak purchasing power, few Burundians can rely on the financial sector to pursue relatively large projects. The earlier example of the inability of a graduate couple to service a fifteen- year mortgage for a relatively small house illustrates this point. Therefore, high interest rates contribute not only to the creation of a difficult business climate, but also perpetuate low living standards. Related to the declining purchasing power is the fact that a sizable portion of the population is in an income bracket that is not serviceable by either the banking sector or microfinance. Whereas the low extreme of the income

144 Table 3.7

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya Characteristics of loans to different groups (amounts in Burundi francs)

Maximum amount Minimum amount Mean amount Median amount Median interest rate (%) Median monthly payment Loan duration (months) Median grace period (months) Average default rate (%) Number of observations

ST outsider

Employees

MT outsider

Others

778,000,000 68,867 4,253,439 1,500,000 19 138,235 12 1

120,000,000 6,390 4,604,983 1,500,000 4 37,383 48 1

3,000,000,000 508,146 104,000,000 30,100,000 19 1,107,011 36 1

700,000,000 200,000 55,900,000 14,800,000 19 655,312 24 1

14.5 4,598

0.3 1,078

17.6 359

17.9 145

Source: Computed by the authors based on data provided by one major commercial bank. Note: The ST and MT outsider refer to short-term and medium-term credit to external clients, respectively.

distribution can rely on the informal and semiformal credit institutions, and the highest end of the distribution can access credit in the formal- banking sector, the needs of those in the income bracket in the lower middle are too high for the informal sector and too low to be of interest to the formal sector. Taking the example of mortgage payment again, the income level of the couple does not allow it to borrow and service the median loan of BIF 1,500,000 in one year at a 19 percent interest rate. Likewise, with an average of BIF 244,238 per loan, microfinance institutions are of little help to middle- income households. Using these two indicators, we may conjecture that borrowers seeking loans between BIF 250,000 and BIF 1,500,000 face a particular challenge in the financial market. Hence the “stranded middle,” representing middle- income households and medium- size firms, remains underserved due to the “missing middle” in the credit market. Credit to “Insiders” Using the account- level data, we identified a group of “special” clients represented by 32 interest- free loans given to “outsiders.” Bank executives explained the presence of these accounts as cases of restructured loans where the beneficiaries renegotiated repayment terms that excluded the payment of interest. However, the amounts involved were so large—the median interest- free loan was nine million francs, which was six times the median loan given to external clients and employees—that the beneficiaries do not seem to be just any normal external clients. Such loans were more prevalent in 2009, and in 2008 to a lesser extent. Indeed, there were only two interest- free loans in 2004, one in 2005, and none in 2006 and 2007. The number increased to seven in 2008 and 22 in 2009. As it is impossible to know the exact identity of these privileged credit

The Financial Sector in Burundi

145

recipients we may only speculate as to who they could be on the basis of some discussions held with people knowledgeable about recent developments at the bank under review. In June 2008, the appointment of a new management team by the government was contested by several shareholders, particularly the group of private shareholders, who argued that the new candidates did not represent their interests. Apparently, the government imposed its candidates even though public institutions controlled only 49 percent of the bank’s capital. Our guess is that once in place, the new management might have decided to buy off those who opposed their appointment by offering them interest- free loans. Otherwise, as they sat on the executive board of the bank, representatives of private shareholders would continue to pose problems to the new management of the bank.36 Assuming that these were the conditions under which the loans were granted, it would not be surprising if they were not reimbursed.37 Bank employees hold 17 percent of all credit contracts (short-, medium-, and long- term loans). Even though, on average, they borrow the same amount as external clients, they have a privileged access to credit as illustrated by information in table 3.7. Most importantly, they pay very low interest rates as their median rate of interest is only 4 percent, which is almost one- fifth of the normal interest rate paid by external clients. Secondly, by reimbursing their loans over longer periods of time (forty- eight months instead of twelve months for external clients), their monthly payments represent almost a quarter of the amount required from external clients. These privileges illustrate two levels of discrimination in the credit market. The first is the fact that only a very small proportion of the population has access to bank credit. The second is that even among those accessing credit, bank employees enjoy by far the best conditions in terms of credit repayment. This has important implications for income inequality and economic equity. Indeed, access to cheap loans enables bank employees to profitably invest in projects that turn out to be unprofitable if undertaken by external clients. The comparison between the 4 percent interest rate charged to employees’ loans and the 19 percent paid by outsiders suggests that banks either recoup their cost with a 4 percent interest rate or they take losses by lending to their employees. If the latter case holds, banks then set high interest rates on loans to outsiders to cover the losses incurred on loans to employees and the few interest- free loans. In both cases, outsiders who constitute the most important client base are unduly penalized. By imposing high interest rates to extract maximum rents from their clients, banks may be responsible for the high rate of default and the narrow market for credit. What about the argument that the low interest rates paid by bank employ36. See, for example, http://www.omac- afrique.org/article.php3?id_article=1029. 37. The largest interest- free loan (300 million francs) is already in default.

146

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

ees are a reflection of their low level of default risk?38 Information in table 3.7 shows that, indeed, bank employees have a very low default risk. Three factors may help explain why. First, the close proximity between banks (lenders) and their employees reduces information asymmetry between the two actors, which in turn reduces employees’ default. Second, by being both employers and lenders, banks have a strong control over their employee borrowers. Defaulting could be too costly as the concerned employees could also lose their job. Third and most important, the way loans are structured encourages default by external borrowers while reducing employees’ default. Bank employees have some of the highest wages in the country but pay only a fraction of their monthly income to reimburse their loans (given that they pay on longer periods). In contrast, external clients have much lower incomes but, at the same debt level, they are required to make monthly payments four times larger than that of bank employees. The example of the couple of university graduates who cannot afford to service their mortgage contrasts with the fact that a young couple of university graduates working in a bank at entry wages would be required to pay about 4 percent of their monthly income to service the same mortgage.39 This is one reason why access to jobs in the banking sector has traditionally been and still is considered the domain of members of the political elite. Normally, it should be the responsibility of the central bank to prevent these lending anomalies through its supervisory and regulatory role. Using Innovative Products to Mobilize Financial Services for Poverty Reduction The discussion in the preceding section raises an important question: What should be the role of the financial sector in helping to reduce poverty in a poor country like Burundi? In addition to the inefficiencies in the financial sector documented in this chapter, and partly because of these inefficiencies, the financial system has had a limited impact on poverty reduction. Could innovations in the financial sector help the country to mobilize more resources in order to make progress in the fight against poverty? By all measures, access to finance in Burundi is lower than in other countries with comparable levels of income. This implies that there is untapped potential that can be harnessed to increase access. As in other African countries, more savings could be mobilized and more financial services supplied by using innovative products that have proved to be better adapted to the needs of the rural population. Mobile banking is a case in point. Although mobile 38. In a regression where default, the identity of the borrower, and year dummies are used as the determinants of the log of nominal interest rates, bank employees still pay lower interest rates, suggesting that low default alone does not explain why bank employees pay such low interest rates (results can be provided upon request). 39. We assumed a salary of 500,000 Burundi francs for a young university graduate who joins a commercial bank.

The Financial Sector in Burundi

147

telephony coverage is still low in Burundi (table 3.6), it has grown fast over the past years. Regional integration will accelerate the rate of penetration of mobile telephony even further with the involvement of the dominant mobile networks in the region such as Safaricom. Mobile banking, which has just been introduced in Burundi, has demonstrated its remarkable power to expand access to finance in African countries as seen in South Africa and Kenya, two countries that are leading the digital revolution in the financial system in Africa. The mobile payment service known as M-Pesa (M for mobile and Pesa for money in Swahili) introduced by Safaricom in Kenya is transforming a cellular phone into a powerful payment tool, bringing the bank to the unbanked. This new banking technology allows the cellular handset to perform four important financial services: (a) a virtual bank card that stores basic information on the customer and the financial institution, thereby saving on costs for bank card distribution; (b) a point of sale terminal, allowing execution of transactions with the bank, the buyer, and the seller, thus facilitating payment for goods and services; (c) an ATM serving as a cash collection and distribution point; and (d) an Internet banking terminal providing instant access to the account and the ability to make remote payments and transfers. In Burundi, where physical infrastructure is underdeveloped in the rural area, with a high population density (306 inhabitants per square km in 2010), and where bank penetration is low, mobile banking has the potential to help bridge the financial service gap in the rural area. In addition to providing financial services, mobile banking is also a source of employment creation and income generation. The experience in Kenya suggests that mobile banking retail operators can earn up to $1,000 per month in revenue (Eijkman, Kendall, and Mas 2010). With increased mobile telephony penetration, this experience can be replicated in Burundi with substantial gains in employment creation and income generation. More generally, poverty reduction in Burundi will require an increased contribution from the private sector through employment creation and income generation. In this context, a key component of the strategy is to increase access to finance for small and medium enterprises (SMEs). Given that traditional commercial banks are not particularly interested in lending to this segment of the private sector, the role of microfinance institutions and innovative payment and credit services is critical to building a vibrant SME sector. Bringing financial services to the traditionally “unbanked” and meeting the needs of SMEs can enhance the contribution of the financial sector to growth and poverty reduction. 3.6

Conclusion

The objective of this chapter was to study the financial system in Burundi and examine its efficiency in mobilizing and allocating financial resources

148

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

within the economy. Analyzing the relationship between the financial sector and the real economy in Burundi helps to shed light on some of the bottlenecks preventing the country from reaching higher levels of economic growth that will substantially reduce poverty. Access to financial services encourages investment and enterprise development, which have a direct effect on poverty reduction, notably through employment creation, and an indirect effect through economic growth. Hence, in Burundi, where most economic agents—particularly firms and households—do not have access to financial services, their contribution to economic growth and poverty reduction remains below their potential. The study showed that the financial system in Burundi is very shallow and highly concentrated. Three traditional commercial banks, namely BANCOBU, BCB, and IBB, together represent three- quarters of total bank assets, credit, and deposits and savings. The lack of competition in the banking industry has prevented the sector from modernizing and offering products that reflect the needs of the market. For example, lending interest rates are so high that investors find it difficult to borrow and invest profitably. But even if investors were able to pay the current rates, other dimensions of the investment climate deter long- term investment. Indeed, within the East African community, Burundi not only has the highest proportion of firms identifying access to credit as a major constraint, but also the highest level of policy uncertainty. Demand for credit is also limited by the low and deteriorating purchasing power of the population. Therefore, the financial market in Burundi is constrained by both the demand for and supply of credit. This explains why credit rationing coexists with excess liquidity in the banking sector. The concentration of bank lending on short- term activities, particularly trade, is a major source of inefficiency in the financial system. Agriculture and industry, the two sectors with the highest growth and employment creation potential, require medium- to long- term credit, so they are out of the credit market. The study finds that credit allocation is subject to three forms of inefficiencies. First, relative to other African countries a large proportion of credit is allocated to the government despite its inherent inefficiencies in Burundi. Second, less than 1 percent of total credit goes to agriculture, the backbone of Burundi’s economy that has been identified as a development priority. This is problematic in a country where agriculture is the source of livelihood for the majority of the population. The lack of credit to agriculture limits the contribution of the financial system to growth, employment creation, and poverty reduction. Third, contrary to the basic principle of modern portfolio theory, credit is not allocated according to the distribution of risk. Commerce has among the highest levels of default risk; yet the sector has the largest share of credit. Similarly, credit defaults are highest with short- term credit but the bulk of lending is short term. One of the

The Financial Sector in Burundi

149

reasons why banks do not diversify their credit portfolios is that they have so far been very profitable, so they have no incentive to change their lending strategies. Among the root causes of the inefficiencies observed in Burundi’s financial sector are important political economy dimensions. The country has been led by groups seeking to advance and preserve their own interests. So they used the financial sector, and the economy in general, to extract rents. Banks and other financial institutions were put in the hands of incompetent managers who used them to offer employment to their relatives and political associates, eventually ruining them to appropriate their assets. Moreover, the central bank failed to regulate and ensure compliance in the financial sector mostly due to political pressure, the limited instruments for intervention, and the limited technical capacity to deal with the challenges of financialsector supervision in a changing international environment. In addition, persistent political and economic instability created high uncertainty that discouraged long- term credit and investment needed to address the country’s structural challenges. One of the consequences has been a succession of failures of financial institutions over the years. Even the current banks are fragile, in spite of their apparent high profitability. Credit is so concentrated on a few clients and very few activities that in the event of economic or political shocks to the economy, the whole banking sector could be seriously affected.40 Despite the serious challenges facing the financial system in Burundi, there are some positive developments that will contribute to defining the future of the sector. First, the central bank has realized that Burundian banks were substantially undercapitalized and it decided to gradually raise the required equity capital. Since 2004, the minimum- required capital has been raised eightfold. The second positive development has been the opening of the financial sector to competition from banks in the region. The traditional banks in Burundi never faced real competition until 2007 when the country joined the East African Community. Burundian banks have to adapt to this new reality if they are to compete with more established financial institutions, particularly from Kenya. In this regard, the entry of new banks such as Diamond Trust Bank and Bank of Africa, as well as the large number of other banks considering entry into the Burundian market, are going to transform the financial sector in Burundi. Already, the introduction of ATMs in 2009 and the decline of bank interest rates from 22 percent in 2005 to 17 percent in 2008 could be considered a result of increasing com40. Rwanda, a neighboring country, is just recovering from such a shock. According to interviews with Rwandan banking officials, the Rwandan banking system was facing serious difficulties when a few clients withdrew their deposits and savings from the system to buy Safaricom stocks on the Kenyan stock exchange. As a result, liquidity was so low that banks decided to ration the amount that could be withdrawn.

150

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

petition and diversification. As part of this diversification process, telephone banking has already been introduced as a cheaper and more efficient form of banking. It is expected that new savings and investment instruments such as equity funds will be offered as competition increases. The third positive development is the expansion of microfinance. In terms of client base, microfinance has outperformed the traditional banking sector by a factor of three to one. By increasing savings collection by 176 percent in just five years, microfinance has also demonstrated that there were substantial savings that were out of reach of the traditional banking sector. This is remarkable considering that microfinance is a relatively new phenomenon in Burundi. Even though microfinance issues small loans relative to the banking sector, it is in a better position to address the needs of a large market segment that is not covered by commercial banks. This has opened up opportunities for entrepreneurship, particularly in rural areas. As a result, given that poverty in Burundi is by and large a rural phenomenon, microfinance is more responsive toward the objective of fighting poverty in the country. Over time, some microfinance institutions could grow and compete with banks, which would benefit borrowers and savers. Alternatively, some microfinance institutions could become so important that traditional banks will seek partnerships with them to cover a larger and more diversified market, including the currently “stranded middle” whose financial needs are beyond the capacity of microfinance, but at the same time too small to be of interest to traditional banks. Either way, the financial sector in Burundi will contribute more to the economy if it is more diversified and more integrated.

The Financial Sector in Burundi

Appendix A Highlights on the Financial Liberalization in Burundi Box 3A.1 Key Control Measures in the Financial Sector before Liberalization 1. Administrative determination of the minimum deposit interest rate by the central bank. This discouraged savings mobilization, as banks often judged real deposit rates as too high. Indeed, in 1986, banks nearly suspended acceptance of new deposits. 2. Administrative determination of the maximum lending interest rate by the central bank. This and the minimum deposit rates limited the banks’ flexibility vis-à-vis their profit margin. This measure discouraged lending to activities judged as risky, which typically happened to be among the important drivers of growth, such as agriculture and industry (see section 3.5). 3. Direct control of credit by fixing preferential lending interest rates and refinancing rates for priority sectors, thus establishing multiple refinancing interest rates. In practice this policy was ineffective due to the difficulty of tracking and enforcing the final destination of loans. 4. Preapproval by the central bank of loans above a given threshold (ten million BIF in 1987). This policy was also ineffective at regulating aggregate domestic credit given the large number of small loans that did not fall under this regulation, and most importantly, due to delays in the release of information on banks’ loan portfolios. 5. The granting of monopoly to the state-owned Caisse de Mobilisation et de Financement (CAMOFI) for the purchase of treasury bonds. 6. Monopoly privileges accorded to selected institutions suffocated competition in the system. For example, only CAMOFI and the central bank handled deposits by state-owned enterprises. Moreover, the central bank established arbitrary quotas for banks in the financing of the coffee campaign, thereby undermining competition and efficiency in the system.

151

Box 3A.2 Chronology of Financial Sector Liberalization in Burundi The liberalization of the financial sector was implemented in four phases: Phase 1: Starting in April 1988 1. Removal of credit preapproval by the central bank. 2. State-owned enterprises are allowed to deposit their liquidity in any institution of their choice. 3. Treasury bonds are open to all financial institutions. 4. Deregulation of lending and deposit interest rates. Phase 2: Starting in September 1988 5. Introduction of Treasury bonds with variable interest rates. 6. Reduction of the number of refinancing rates from 7 to 3. 7. Reduction of the number of maximum lending rates from 8 to 3. 8. Removal of all minimum deposit interest rates except the rates on some special savings accounts (e.g., comptes sur carnets). 9. Revision of the use of the medium-term liquidity coefficient. Phase 3: Starting in December 1988 10. Reduction of the number of refinancing rate to only two (normal rate and preferential rate). 11. Removal of all ceilings on lending rates and floors on deposit rates. Phase 4: Starting in October 1989 12. Complete liberalization of lending and deposit interest rates. 13. Reform of the national guarantee fund (Fonds National de Garantie). 14. Revision of the land law giving preferential treatment to the state in terms of repossession of guarantees in cases of default on land-backed credit. Sources: Compiled by authors from various reports of the Central Bank and the Ministry of Planning.

153

The Financial Sector in Burundi

Appendix B Table 3B.1

Indicators of the investment climate, Burundi and EAC (average 2000–08)

Electricity (% of firms identifying this as a major constraint) Access to finance (% of firms identifying this as a major constraint) Tax rates (% of firms identifying this as major constraint) Transportation (% of firms identifying this as a major constraint) Corruption (% of firms identifying this as a major constraint) Policy uncertainty (% of managers surveyed ranking this as a major constraint) Labor skill level (% of firms identifying this as a major constraint) Labor regulations (% of firms identifying this as a major constraint)

Burundi

Rwanda

Kenya

Tanzania

Uganda

EAC

72.3

55.0

48.2

73.7

64.3

62.7

50.9

36.0

44.1

44.5

46.4

44.4

36.1

44.7

68.3

55.1

55.5

51.9

21.1

27.4

37.4

18.5

22.6

25.4

19.7

4.4

73.8

35.4

30.9

32.8

14.5

0.9

0.5

0.3

4.0

11.8

11.7

27.6

22.3

20.5

18.8

3.9

2.8

22.6

8.5

6.0

8.7

Source: World Bank, Doing Business database.

References African Development Bank (AfDB). 2009. An Infrastructure Action Plan for Burundi: Accelerating Regional Integration. Tunis: African Development Bank. Banque de la République du Burundi. 2006. “Circulaire No. 01/06 Relative au Capital Minimum des Banques et des Établissements Financiers Édictée en Vertu du Décret-Loi No. 1/017 du 23 Octobre 2003 Portant Règlementation des Banques et des Établissements Financiers.” Bujumbura. Barro, R. J. 1991. “Economic Growth in a Cross-Section of Countries.” Quarterly Journal of Economics 106 (2): 407– 43. Bartelsman, E., J. Haltiwanger, and S. Scarpetta. 2009. “Cross-Country Differences in Productivity: The Role of Allocation and Selection.” IZA Discussion Paper Series no. DP 4578, Institute for the Study of Labor. Bates, R. 1981. Markets and States in Tropical Africa: The Political Basis of Agricultural Policies. Berkeley: University of California Press. Bertelsmann Stiftung. 2009. BTI 2010: Burundi Country Report. Gütersloh, Germany: Bertelsmann Stiftung. http:// www.bti- project .org/ fileadmin/ Inhalte /reports/2010/pdf/BTI%202010%20Burundi.pdf. Bigsten, A, P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J. Gunning, M. Soderbom, A. Oduro, R. Oostendorp, C. Pattillo, F. Teal, and A. Zeufack. 2003. “Credit Constraints in Manufacturing Enterprises in Africa.” Journal of African Economies 12:104– 25.

154

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Chrétien, J. P., and M. Mukuri, eds. 2002. Burundi, La Fracture Identitaire: Logiques de Violence et Certitudes “Ethniques.” Paris: Karthala. Eijkman, F., J. Kendall, and I. Mas. 2010. “Bridges to Cash: The Retail End of M-Pesa, The Challenge of Maintaining Liquidity for M-Pesa Agent Networks.” Savings and Development 34 (2): 219– 52. Faccio, M. 2006. “Politically Connected Firms.” American Economic Review 96 (1): 369– 86. Financial Access Initiative. Online data and reports. www.financialaccess.org. Fisman, R. 2001. “Estimating the Value of Political Connections.” American Economic Review 91 (4): 1095– 102. Hsieh, C.-T., and P. Klenow. 2009. “Misallocation and Manufacturing TFP in China and India.” Quarterly Journal of Economics 124 (4): 1403– 48. Hubbard, G. 2005. Money, the Financial System, and the Economy. Reading, MA: Addison-Wesley. Institut de Statistiques et d’Études Économiques du Burundi (ISTEEBU). 2008. Comptes de la Nation 2005. Bujumbura, Décembre. International Monetary Fund (IMF). 2000. “Burundi: Statistical Annex.” IMF Staff Country Report no. 00/58, April. http://www.imf.org/external/pubs/cat/longres .aspx?sk=3532.0. ———. 2005. “Burundi. Deuxième Revue de L’accord Triennal au Titre de la Facilité pour la Réduction de la Pauvreté et pour la Croissance.” July 8. http://www.imf .org/external/pubs/ft/scr/2005/fra/cr05322f.pdf. International Monetary Fund (IMF) and World Bank. 2009. “Programme d’Évaluation du Secteur Financier.” http://www.imf.org/external/np/exr/facts /fre/fsapf.htm. Khwaja, A., and A. Mian. 2005. “Do Lenders Favor Politically Connected Firms? Rent Provision in an Emerging Financial Market.” Quarterly Journal of Economics 120 (4): 1371– 411. Lemarchand, R. 1994. Burundi: Ethnocide as Discourse and Practice. Cambridge: Cambridge University Press. Leuz, C., and F. Oberholzer-Gee. 2003. “Political Relationships, Global Financing and Corporate Transparency.” Working paper no. 19– 04, The Rodney L. White Center for Financial Research, The Wharton School, University of Pennsylvania. Lienart, T. 2010. PowerPoint presentation made at the Investment Policy Review for Burundi, Geneva. http://unctad.org/en/pages/PublicationArchive.aspx ?publicationid=488. Lim, C., and L. Rugwabiza. 2009. “A Shock Analysis of Burundi’s Economy. The Financial Crisis and Other Shocks.” Unpublished Manuscript, African Development Bank, Regional Department East Africa A (OREA). http://www.afdb.org /fileadmin/uploads/afdb/Documents/Project- and-Operations/A%20shock%20 analysis%20of%20Burundi’s%20economy%20v1.3.pdf. Ndikumana, L. 1998. “Institutional Failure and Ethnic Conflicts in Burundi.” African Studies Review 41 (1): 29– 47. ———. 2005. “Distributional Conflict, the State and Peace Building in Burundi.” The Round Table 94 (381): 413– 27. Ngaruko, F., and J. Nkurunziza. 2006. “Political Economy of Reform in War-Prone and Polarized Societies.” In Understanding Economic Reforms in Africa: A Tale of Seven Nations, edited by J. Mensah. New York: Palgrave Macmillan. Nkurunziza, J. D. 2005. “Political Instability, Inflation Tax and Asset Substitution in Burundi.” Journal of African Development 7 (1): 42– 72.

The Financial Sector in Burundi

155

———. 2007. “Generating Rural Employment in Africa to Fight Poverty.” In Towards Full and Decent Employment, edited by J. A. Ocampo and K. S. Jomo, 158– 90. London: Zed Books Ltd. ———. 2010a. “Inequality and Fiscal Policies in Burundi.” In Post-Conflict Policy and Horizontal Inequalities, edited by F. Stewart. Oxford: Oxford University Press. ———. 2010b. “Why Is the Financial Sector in Burundi Not DevelopmentOriented?” EUI-RESCAS Working Paper no. 29, Robert Schuman Centre for Advanced Studies, European University Institute. Nkurunziza, J. D., and F. Ngaruko. 2000. “An Economic Interpretation of Conflict in Burundi.” Journal of African Economies 9 (3): 370– 409. ———. 2005. “War and its Duration in Burundi.” In Understanding Civil War: Evidence and Analysis, vol. 1, edited by N. Sambanis and P. Collier. Washington, DC: World Bank Publications. ———. 2008. “Why Has Burundi Grown so Slowly? The Political Economy of Redistribution.” In The Political Economy of Economic Growth in Africa, 1960– 2000: Country Case Studies, vol. II, edited by B. J. Ndulu, S. A. O’Connell, J. P. Azam, R. H. Bates, A. K. Fosu, J. W. Gunning, and D. Njinkeu, 51– 85. Cambridge: Cambridge University Press. Nyamoya, P. 2004. “Analyse de la Politique Monétaire du Burundi 1980– 2002.” Cahiers du CURDES, December. Nyamoya, P., and L. Nkeshimana. 2005. “Project de Création d’un Fonds D’investissement.” Document de Travail no. 01 Préparé pour l’ABEF, April. République du Burundi. 2003. La Loi Bancaire. Réglementation des Banques et des Établissements Financiers. Loi n°1/017 du 23 Octobre 2003. Bujumbura. ———. 2007. Programme D’actions Prioritaires de Mise en œuvre du Cadre Stratégique de Lutte Contre la Pauvreté. Bujumbura, May 24– 25. ———. 2008. Economie Burundaise 2007. Vice-Ministère de la Planification, Service de la Planification Macroéconomique, Bujumbura. Robinson, M. S. 2001. The Microfinance Revolution: Sustainable Finance for the Poor. Washington, DC: World Bank. Sensarma, R. 2006. “Are Foreign Banks Always the Best? Comparison of StateOwned, Private and Foreign Banks in India.” Economic Modelling 23:717– 35. Shleifer, A., and R. W. Vishny. 1994. “Politicians and Firms.” Quarterly Journal of Economics 109 (4): 995– 1025. Ntibazonkiza, Raphael. 1993. Au Royaume des Seigneurs de la Lance: Une Approche Historique de la Question Ethnique au Burundi, tome 2. Bruxelles: Bruxelles-Droits de L’Homme. United Nations Development Program (UNDP). 2010. Human Development Report 2010. New York: UNDP. http://hdr.undp.org/en/reports/global/hdr2010. World Bank. 1999. “Burundi: Poverty Note. Prospects for Social Protection in a Crisis Economy.” Report no. 17909-BU, Washington, DC. ———. 2005. World Development Indicators. Washington, DC: World Bank. http:// documents.worldbank .org/ curated/ en/ 2005/ 03/ 6428863/ world- development - indicators- 2005. ———. 2008. World Development Report. Washington, DC: World Bank. ———. 2010. World Development Indicators. Washington, DC: World Bank. http:// data.worldbank.org/data- catalog/world- development- indicators/wdi- 2010. Wright, D. J. 1995. “A Human-Scale Development Organization Crosses Paths with a Vast but Shoddy African Banking Network and an Even Vaster International System.” Compass 14 (4): Part 2. http://www.gvanv.com/compass/arch/v1403 /wright2.html.

156

Janvier D. Nkurunziza, Léonce Ndikumana, and Prime Nyamoya

Wurgler, J. 2000. “Financial Markets and the Allocation of Capital.” Journal of Financial Economics 58 (2): 187– 214. Yunus, M. 2003. Banker to the Poor: Micro-Lending and the Battle against World Poverty. Dhaka, Bangladesh: The University Press. Zeller, M., and R. Meyer, eds. 2002. The Triangle of Microfinance: Financial Sustainability, Outreach and Impact. Baltimore: Johns Hopkins University Press.

4

Were the Nigerian Banking Reforms of 2005 a Success . . . and for the Poor? Lisa D. Cook

The recent global financial crisis provides the latest evidence that resolving banking crises can be costly in any country. Losses typically represent a larger share of income in developing countries than in industrialized countries. For example, between 1987 and 1993, Norway, Sweden, and Finland had bank insolvency crises whose resolution cost 4.0, 6.4, and 8.0 percent of gross domestic product (GDP). At approximately the same time similar crises in Mauritania, Senegal, and Cote d’Ivoire cost 15.0, 17.0, and 25.0 percent of GDP.1 Such potential losses provide a compelling reason for economists to identify, and for policymakers to consider, policies to prevent such episodes in the poorest countries. A recent literature has focused on analyzing which policies promote the development, efficiency, corporate governance, and accessibility of banks (e.g., Barth, Caprio, and Levine 2001, 2008; Abiad and Mody 2005; Beck, Demirgüç-Kunt, and Levine 2009). Another literature makes the further link between finance and growth (e.g., King and Levine 1993a, 1993b; Levine and Zervos 1998; Rajan and Zingales 1998). Given that African countries have lagged other countries in adopting bank reforms, Lisa D. Cook is associate professor of economics and international relations at Michigan State University. I am grateful to Magnus Adiele, Bode Agusto, Edna Ishaya, Eric Naivasha, Doyin Salami, and bankers and banking officials in Nigeria for their helpful discussions and Yaw Ansu and Benno Ndulu for their helpful comments. I also benefited from comments from seminar participants at the Central Bank of Nigeria, Michigan State University, and the National Bureau of Economic Research and able research assistance from Chaleampong Kongcharoen, Suhyeon Nam, and Íñigo Verduzco. Funding from the NBER Africa Project is gratefully acknowledged. A substantial part of the chapter was written while visiting the National Poverty Center at the University of Michigan, and its support is also acknowledged. For acknowledgments, sources of research support, and disclosure of the author’s or authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13361.ack. 1. Caprio and Klingebiel (1996, 13–19).

157

158

Lisa D. Cook

knowing whether the findings from cross-country evidence are relevant is difficult. Nigeria’s banking reform of 2005 provides a natural experiment in which to test the efficacy of best practices in Africa. Did the banking regulation and supervision reforms of 2005 make the financial system more sound? Did they contribute to development, efficiency, and accessibility of the banking system? If so, which mechanisms were most important? Data from Nigeria’s experiment are combined with new survey data to address these questions. I find that in the immediate aftermath of the policy changes, the financial system was more stable than it was previously. While it is found that development and efficiency have increased, the successes in introducing more small savers and borrowers to the formal-banking sector have been more limited. Seventy-four percent of Nigerian residents remain unbanked, including 70 percent or more of business owners and traders. This has large implications with respect to changing the incentives and constraints facing most economic agents in Nigeria. 4.1

The Nigerian Bank Reform of 2005

The 2005 Nigerian banking reform was a watershed event. To put its significant changes in historical perspective, I will review the principal institutional features of Nigerian banking preceding reform. 4.1.1

Recent Reform Efforts

A nationalization effort in the 1970s and 1980s left Nigeria’s banks subject to extensive state intervention and control. Before Nigeria initiated its Structural Adjustment Program (SAP) in 1986, the banking sector was plagued by problems characteristic of many African and poor countries at the time. Direct intervention by the state was accomplished using a number of instruments, including credit and interest-rate controls and restrictions on entry. There were few banks—twenty-nine commercial banks and twelve merchant banks—for a population of 84 million. There was little activity outside the government sector, as it accounted for 80 percent of commercial banks’ and 45 percent of merchant banks’ assets. There was little competition, entry, and exit. The financial liberalization program accompanying SAP was designed to address these issues and to extend lending and other banking services. Specifically, its measures included reducing barriers to entry, liberalization of lending and savings rates, introducing an interbank foreign exchange market, deregulating interbank lending, and privatizing a number of banks and financial institutions. The success of this reform was mixed. The number of market participants increased. Eight times the number of banks entered annually from 1987 to 1990 than had in the previous decade. Figure 4.1 displays this exponential growth in each sector of the banking industry. Yet much of the resulting banking activity was not concentrated on lending to the private sector and households, but on exploiting new arbitrage

Were the Nigerian Banking Reforms of 2005 a Success?

159

Fig. 4.1 Community/microfinance banks, commercial banks, and merchant banks, 1970–2008 Source: CBN Statistical Bulletin 2007. Notes: After 2000, universal banking was introduced in Nigeria, which eliminated the classification into merchant and commercial banks.

opportunities in foreign-exchange operations and money market interestrate spreads.2 If consistent with cross-country evidence, this outcome of reform would limit the breadth and depth of the banking sector.3 The number of banks peaked at 120 in 1991. Simultaneously, banks began accumulating nonperforming loans at an increasing rate, and the share of distressed banks doubled from 26 percent in 1991 to 52 percent in 1995. There was a bank run in 1993, and the banking sector entered a period of sustained crisis.4 To address bank insolvency, minimum capital requirements were increased in 1988 and 1989. They were increased again in 1991, and other measures were implemented to enhance the regulatory and supervisory powers of the Central Bank of Nigeria (CBN) in the Banks and Other Financial Institutions Law of 1991. The licenses of twenty-six banks were revoked in 1998, and, by 2002, only 16 percent of banks were insolvent. 2. See Lewis and Stein (1997) for a detailed account of this reform episode. 3. See, for example, Beck and Fuchs (2004). 4. While most researchers agree that expansionary fiscal and monetary policy contributed to financial collapse in the 1990s, alternative views of the principal causes of the banking crisis have emerged in the literature. Beck, Cull, and Jerome (2005) attribute the crisis to flaws in the liberalization agenda that resulted in rent-seeking behavior by banks and a concomitant shift in resources from the real economy to banks. Lewis and Stein (1997) emphasize the bad political and institutional setting in which reforms were undertaken in explaining the crisis.

160

Lisa D. Cook

Nonetheless, this was still high by international standards, as was the ratio of nonperforming loans to the total, 25 percent in 2002. New policies and institutional changes in 2001 were aimed at increasing the stability of the banking system. In addition, to increase the flow of credit to and speed up the development of the private sector, universal banking was adopted in 2001, and the distinction between commercial and merchant banking disappeared. 4.1.2

The Need for Reform

While relatively more credit flowed to the private sector, the fundamental issue of insolvency had not been addressed by the 2001 law nor by incremental increases in capital requirements and still threatened the system. By 2004, many banks were undercapitalized, despite having met minimum capital requirements of roughly $7.5 million for existing banks and $15 million for new banks. Shareholders’ funds had been reduced by operating losses, further contributing to insolvency. More than one-third of all banks were “marginal” or “unsound” according to CBN criteria.5 Twenty-eight percent of bank loans were nonperforming.6 While bank concentration was falling, it was still high, with ten banks accounting for half the deposits and assets of the banking system (see figure 4.2). Other conditions prevailing in 2004 that threatened bank development, bank efficiency, corporate governance, and accessibility were overreliance on public-sector deposits, weak corporate governance, and substantial insider lending that resulted in large portfolios of nonperforming loans, and neglect of small and medium-sized savers.7 Many of the problems in the larger banking sector were reflected in the microfinance sector among community banks, the latest institution designed to address the lack of access to finance among firms, households, and the rural poor.8 An additional threat to the market for microfinance was regulatory arbitrage, because operators could select to which body they would report and, by extension, by whom they would be regulated—the central bank or the National Board of Community Banks, which was appointed by the Ministry of Finance.9 In July of 2004, the governor of CBN, Charles Soludo, announced an ambitious thirteen-point reform agenda to comprehensively reform the 5. CBN (2006). 6. Ibid. These data for the year ending on September 30, 2005. 7. See Soludo (2004) for a complete description of prevailing conditions. 8. A number of experiments had been implemented in the past three decades to address rural and poor savers and borrowers, including the People’s Bank of Nigeria, which offered soft loans, and mandated rural bank branching for commercial banks. Cook (2004) reviews a number of these schemes. See FAO (2004) and Isern et al. (2009) for additional assessments of the performance of community banks and other microfinance institutions. 9. Cook (2004) cites regulatory arbitrage as a major threat to the community-banking sector. The problem of uneven reporting, late reporting, nonreporting, and the resulting lack of transparency were common to the entire financial system.

Were the Nigerian Banking Reforms of 2005 a Success?

161

banking system. The centerpiece of the proposed changes was a more than tenfold increase in the minimum capital requirement for banks from NGN 2 billion to NGN 25 billion (roughly $190 million). Meeting the new capital standard could only be accomplished by mergers, acquisitions, or injections of new capital. This type of bank consolidation was a novel feature of reform, because there were no such restrictions in earlier rounds of raising capital requirements and because there was little history of mergers and acquisitions in the Nigerian banking sector.10 Other major elements of the reform program were a phased withdrawal of public-sector funds from Nigerian banks, adoption of a rule-based regulatory framework that was more risk focused, and restructuring of the information-gathering and reporting mechanism to ensure greater compliance and transparency. Importantly, while insider lending had been identified as a major problem, corporate governance was on the list of reforms, and a Code of Corporate Governance was enacted by the CBN for banks in March 2006, corporate governance was given less attention relative to bank consolidation and higher capital requirements.11 The central bank also anticipated higher capital requirements (NGN 20 million) and greater supervision and separation of the microfinance sector, since community banks were found to have inadequate capital relative to lending risk and weak institutional capacity. In the remainder of the chapter, I will assess the effects of these regulatory changes in the banking sector. 4.1.3

Outcomes

The bank-consolidation process was largely complete as mandated by the end of 2005. All but four banks participated in mergers and acquisitions. Fourteen banks that failed to sufficiently increase their capital base lost their licenses, and twenty-five banks remained, as is reported in figure 4.1. Community banks were also asked to increase their capital base and convert to microfinance banks (MFBs) starting in 2006. By the end of 2008, 603 of the 757 community banks had converted, applications for new banks were received, and the number of MFBs totaled 840.12 Some larger banks also acquired community banks. A number of auxiliary institutions were created or invited to participate in the microfinance support network, including the MFB Development Fund, the National Microfinance Consultative Committee, the Association for Microfinance Institutions, a credit reference bureau, credit-rating agencies, and programs for deposit insurance. Before 10. Failed banks during the 1998 reform were liquidated by the Nigerian Deposit Insurance Corporation, much in the way the S&L crisis was resolved in the United States in the late 1980s and 1990s. See Ezeoha (2007) for an historical account. 11. See Ogunleye (2003) for an analysis of the relation between bank failures and corporate governance in the 1990s. 12. CBN (2008). Not all community banks converted to MFBs due to insolvency and subsequent license revocation.

162

Lisa D. Cook

embarking on graphical and empirical analysis, I describe the data collected and their sources. 4.2

Data

Bank- and system-level data sets are constructed for analysis. Bank-level data are collected from the financial statements of individual banks and from Statistical Bulletins, Banking Supervision Annual Reports, and Annual Reports of CBN for various years. Banks are required to report balance sheet and profit data to CBN, and a subset of these data are reported in these publications. Given the small number of banks, each can be tracked over time, and a panel data set is constructed for the years 2001 to 2008. System-wide data are gleaned from several sources for the years 1990 to 2008. In addition to the aforementioned CBN sources, aggregate data have been collected from Beck, Demirgüç-Kunt, and Levine (2009), Beck et al. (2009), International Financial Statistics 2010 (IMF 2010), and the Economic Intelligence Unit (EIU). Data on consumer finance are taken from the 2008 national survey of 25,000 households conducted by Enhancing Financial Innovation & Access (EFinA). Data on the Nigerian banking system are also extracted from three rounds of surveys of bank regulatory and supervisory authorities to identify features of bank regulation, supervision, and structure found in Beck, Demirgüç-Kunt, and Levine (2000). 4.3

Graphical Evidence

Before moving on to the formal empirical tests, in some cases mere inspection of the data will be sufficient to convey general patterns in the data. Graphical evidence is presented below and is followed by regression analysis. 4.3.1

Major Stated Objectives

Figure 4.2 shows that the ratio of distressed banks to total dropped from 14 percent in 2005 to 4 percent in 2006 to 0 percent in 2007, and that the share of nonperforming loans relative to total loans and advances fell from 28 percent in 2004 to 8 percent in 2008. As anticipated, there is less government intervention in the banking sector, whether measured by deposits or ownership of government securities (figure 4.3), and the level of bank concentration, typically a measure of competitiveness in a banking system, had fallen by the end of 2008 (figure 4.4). Credit to the public sector fell as credit to the private sector rose (figure 4.5). However, figure 4.6 shows that the ratio of bank credit to deposits has increased markedly since 2004, which means that banks must rely on other sources of funding, for example, capital markets, to support significantly higher lending activity. I return to this point below.

Were the Nigerian Banking Reforms of 2005 a Success?

Fig. 4.2

163

Distressed banks and nonperforming loans, 1990–2008 (percent)

Source: For distressed banks: Author’s calculation, CBN Statistical Bulletin 2007, and Onaolapo (2008); nonperforming loans, 2001–2008: Author’s calculation, and CBN Banking Supervision Annual Reports 2007, 2005, 2008; nonperforming loans, 1990–2000: Alashi S. O. (2002) cited in Adegbite E O (2005).

Bank supervisors and banks were charged with taking greater account of risk. Data on capital adequacy, liquidity, and asset quality in table 4.1 demonstrate the extent to which this happened. Although the minimum capital adequacy ratio is 10 percent, most banks have significantly exceeded the prescribed ratio since 2005. The liquidity ratio increased by more than 50 percent before settling slightly above the 2005 ratio. In tandem with the decline in the share of nonperforming loans was a reduction in bad-debt provisioning from 27 percent of total loans and advances to 6 percent, which freed up resources for other uses. From the EIU index of banking risk, which reflects an evaluation of risk management and potential for crisis, risk has fallen sharply since the 2005 reforms, as can be seen in figure 4.4. Banks’ reporting has increased, and relatively more data are available. Among MFBs, the share of nonreporting banks has declined from 82 percent in 2001 to 0 percent in 2008.13 Using data from the Beck et al. survey, capital regulatory and official supervisory power indices were constructed as in Barth, Caprio, and Levine (2001). Higher levels of the index imply better positioning of the financial system with respect to initial and overall 13. CBN Statistical Bulletin (2009).

Composition of banking system assets, 2001 and 2008

Source: CBN Annual Reports.

Fig. 4.3

Fig. 4.4

Bank concentration and banking-sector risk, 1992–2008

Source: For bank concentration, Beck, Demirgüç-Kunt, and Levine (2009); banking-sector risk, EIU.

Fig. 4.5 Lending to public sector/total, lending to private sector/total, lending per income earner, lending per household, total lending/GDP, 1990–2008 Source: EIU.

Fig. 4.6 Bank Credit/deposits, liquid liabilities/GDP, bank deposits/GDP, 1990–2008 Source: CBN Annual Reports.

Table 4.1

Financial soundness indicators, Nigerian banking system (2004–2008) 2004–2008

Indicator Capital adequacy Capital adequacy ratio Liquidity Liquidity ratio Aggregate credit to deposit ratio Asset quality Nonperforming credits to total loans and advances Bad-debt provision total loans and advances

2004

2005

2006

2007

2008

Average

S. D.

14.2

17.8

22.6

20.9

21.9

19.5

3.5

52.0 74.0

41.3 66.6

65.1 65.1

60.9 70.9

46.5 70.9

53.1 69.5

9.9 3.6

27.9

24.2

10.7

10.2

7.5

16.1

9.2

22.6

19.1

6.3

8.1

6.1

12.5

7.8

Source: CBN Annual Reports 2004–2008. Note: Capital adequacy ratio = (qualifying capital/risk-weighted assets) × 100. At least 50 percent of qualifying capital shall comprise paid-up capital and reserves. From January 2004, minimum capital adequacy ratio is statutorily set at 10 percent. Liquidity ratio = (total specified liquid assets/total current liabilities) × 100. Statutory minimum liquidity ratio was changed from 40 percent to 30 percent in 2008. Maximum recommended aggregate credit to deposit ratio is 80 percent.

Were the Nigerian Banking Reforms of 2005 a Success?

167

capital stringency and official supervisory power. While the capital regulatory index was unchanged between 2000 and 2007, the index of supervisory power improved more than 10 percent between 2000 and 2007. This would be consistent with increases in the scope and depth of CBN’s supervisory role stemming from the 2005 reforms. 4.3.2

Other Financial Indicators

With respect to bank development, financial deepening increased between 2004 and 2008, using a variety of measures. As in King and Levine (1993b), a broad measure of financial depth is the ratio of liquid liabilities (currency plus demand and interest-bearing liabilities of banks and other financial intermediaries) to GDP. This ratio fell initially and then rose from 2006. The M2/GDP nearly doubled from 19.8 percent to 38 percent, and credit to the private sector more than doubled from 13.2 percent to 33.5 percent.14 Financial development may also be measured by the relative importance of deposit money banks assets to central bank assets. The ratio of deposit money bank assets to deposit money bank assets and central bank assets has increased from 0.83 in 2004 to 1.15 in 2008.15 How much better were Nigerian banks at intermediating society’s savings into private-sector credits than before? Cross-country evidence (e.g., Beck and Fuchs 2004) shows that higher interest spreads are inversely related to credit to the private sector as a fraction of GDP. Figure 4.6 demonstrates that Nigerian banks are more efficient at intermediation than prior to 2005, as measured by bank credit to bank deposits. Figure 4.7 shows that they are more efficient by other measures, that is, by declining overhead costs as a fraction of total assets. There is mixed evidence on efficiency using data on net interest margin from the Beck and Demirgüç-Kunt (2004) series, which suggests that it has fallen, and the EIU series, which suggests that it is unchanged in figure 4.8.16 Are banks more profitable in the postreform years? Figure 4.7 indicates they were slightly more profitable in terms of return on assets and return on equity. More generally, these findings conform to cross-country evidence as well (e.g., Beck and Demirgüç-Kunt 2009). Returns on assets have decreased and then increased across countries. While returns on equity have converged to 15 percent in high- and middle-income countries, as in Nigeria, their variance is significantly higher in lower income countries, also as in Nigeria. 14. CBN Annual Reports 2000–2008. 15. While the explicit relation between financial development and growth is beyond the scope of this chapter, King and Levine (1993a, 1993b) and Levine, Loayza, and Beck (2000) show that financial development and growth are positively related. 16. Although not pictured here, data were collected on the bank cost-income ratio, overhead costs relative to gross revenues, between 1992 and 2008. These data also indicate that banks are more efficient, as the lowest levels in this range are 0.547 and 0.510 in 2007 and 2008, which reflected a downward trend starting most recently in 2003.

Fig. 4.7

Bank ROA, ROE, bank overhead costs/total assets, 1990–2008

Source: CBN Annual Reports.

Fig. 4.8

Deposit money bank assets/GDP, net interest margin, 1990–2008

Source: Beck (2004), CBN, and EIU.

Were the Nigerian Banking Reforms of 2005 a Success? Table 4.2

169

Characteristics of the banked and unbanked in Nigeria, 2008

Gender Male Female Area Urban Rural Socioeconomic status AB C1 C2 DE Regular sources of income Salaries Subsistence farming Commercial farming Trading of goods and services (nonfarming) Money from family/friends Own business (nonfarming) Do not receive income

Never banked

Previously banked

Currently banked

67 82

7 3

27 15

54 82

7 4

39 14

23 7 25 81

0 3 4 5

77 91 70 13

23 88 85 69 74 71 91

5 4 6 5 3 6 4

71 8 9 25 23 23 5

Source: FinScope/EFInA (2008). Note: Calculation of socioeconomic status includes education, type of house, occupation, and ownership of household durables. Categories are roughly equivalent as follows: AB ≈ high income (0.4 percent of adult population); C1 ≈ upper-middle income (1.2 percent of adult population); C2 ≈ lower-middle income (11.2 percent of adult population); and DE ≈ low income (87.3 percent of adult population).

4.3.3

Microfinance Institutions

In tandem with activity in the larger banking sector, the MFBs’ capitalto-asset ratio increased from 24 percent in 2003 to 30 percent in 2008 (see table 4.8). With respect to accessibility, MFBs have increased their lending relative to deposits from 59 percent in 1998 to 69 percent in 2008. However, according to a large national household survey conducted in 2008, large swathes of the population, including farmers, traders, owners of firms, and the poorest, remain without access to banking services. Results from analysis of these data appear in tables 4.2, 4.3, 4.4, 4.5, 4.6, and 4.7. More than threequarters of respondents have had neither a savings nor a checking account. Two percent of respondents have ever had a loan from a community bank. In terms of loans to the rural population, 1.1 percent of the adult population has or once had a loan from MFBs compared to 0.7 percent from commercial banks.17 Ninety-three percent of those in the sample prefer to receive cash as payment, and the share is 0.99 for business owners. Since these are 17. Author’s calculation from EFInA (2008).

Table 4.3

Means of receiving income

Cash Cheque Into bank account Other

All

DE status

DE + C2 status

Own business

Own business and DE status

Own business and DE + C2 status

92.7 1.0 6.3 0.1

96.1 0.7 3.2 0.1

93.3 0.9 5.7 0.1

98.6 0.7 0.6 0.1

99.2 0.5 0.3 0.0

98.6 0.7 0.6 0.1

Source: Author’s calculation from FinScope/EFInA (2008). Note: Calculation of socioeconomic status includes education, type of house, occupation, and ownership of household durables. Categories are roughly equivalent as follows: C2 ≈ lowermiddle income (11.2 percent of adult population) and DE ≈ low income (87.3 percent of adult population). Columns (2) to (4) are calculated from all sources of income. Columns (5) to (7) are calculated for people whose regular source of income is own business.

Table 4.4

Bank accounts and loan activity

Never had savings account Never had current account Never had credit card Never had loan from microfinance/ community bank Currently have loan Loan to start business (of those with loans currently) Source of loan from commercial bank Source of loan from microfinance/ community bank Loan to expand business (of those with loans currently) Source of loan from commercial bank Source of loan from microfinance/ community bank

Own business and DE + C2 status

All

DE status

DE + C2 status

Own business

Own business and DE status

76.1 89.3 96.6

83.3 93.6 98.0

77.1 90.2 96.9

73.5 89.4 96.1

80.5 93.6 97.4

74.3 90.3 96.4

98.1 6.8

98.4 6.5

98.1 6.6

97.8 9.3

98.1 9.2

97.9 9.2

22.7 10.0

23.6 5.8

23.2 9.9

29.4 3.9

30.7 0.6

29.7 3.5

4.3

3.9

4.4

1.4

0.5

1.4

30.2 7.7

31.2 3.6

30.5 6.0

49.2 7.5

50.6 3.7

48.76 5.5

6.0

4.2

2.3

1.9

1.8

5.48

Source: Author’s calculation from FinScope/EFInA (2008). Note: Calculation of socioeconomic status includes education, type of house, occupation, and ownership of household durables. Categories are roughly equivalent as follows: C2 ≈ lower-middle income (11.2 percent of adult population) and DE ≈ low income (87.3 percent of adult population).

Table 4.5

Summary statistics

Variables Time series (1992–2008) Private credit/GDP Output gap Real GDP per capita (N, 2000 prices) Nonperforming loans/total assets Real interest rate (%) Overhead cost/total assets ROE ROA Interest spread (%) Bank assets/GDP Bank concentration Liquidity assets/deposit Equity (N million) Loans/liabilities Real treasury bill rate (%) Inflation (%) Change in industrial production index Net interest margin Panel data (1992–2008) Provision of bad debt/loans and advances Dummy for new or merged bank Dummy for foreign-owned bank (1993–2008) Provision of bad debt/loans and advances Dummy for new or merged bank Equity/lag of total assets Dummy for foreign-owned bank

N

Mean

Std. dev.

Min.

Max.

17 17 17 17 17 17 17 17 17 17 17 17 17 17 17 17 16 17

0.13 –0.90 47,306 0.13 –2.76 0.08 0.25 0.03 12.21 0.17 0.51 0.25 434,396 0.37 –9.46 23.34 0.01 0.08

0.04 5.13 9,747 0.05 19.53 0.02 0.08 0.01 4.39 0.05 0.19 0.07 752,941 0.08 20.18 21.16 0.04 0.02

0.08 –11.13 38,691 0.03 –52.60 0.04 0.09 0.01 1.66 0.10 0.39 0.11 5,450 0.19 –60.34 5.38 –0.06 0.05

0.25 4.81 64,773 0.19 14.34 0.10 0.44 0.05 20.70 0.33 0.95 0.36 2,788,537 0.49 11.20 72.84 0.07 0.11

174 176 176

0.11 0.74 0.20

0.09 0.44 0.40

0.01 0.00 0.00

0.47 1.00 1.00

150 150 150 150

0.11 0.75 0.28 0.19

0.10 0.44 0.23 0.40

0.01 0.00 0.06 0.00

0.47 1.00 2.09 1.00

Source: For output gap: Author’s calculation by HP-filter real GDP, IMF (2009) World Economic Outlook; real interest rate: Author’s calculation from lending rate and CPI inflation, IMF (2010) IFS; inflation: Author’s calculation, IMF (2010) IFS; private credit/GDP, overhead cost/total asset, ROE, ROA, interest spread, bank assets/GDP, bank concentration, net interest margin: Beck, Demirgüç-Kunt, and Levine (2009); nonperforming loan/total asset, provision of bad debt/loans and advances, new or merged bank, foreign-owned bank: CBN; liquidity assets/deposit, equity, loans/liability: Author’s calculation, CBN, Statistical Bulletin. vol. 18 (Dec 2007) and Golden Jubilee Edition (Dec 2008); real GDP per capita: IMF (2009) World Economic Outlook; real Treasury bill rate, industrial production index: IMF (2010) IFS.

Table 4.6

Types of banks and bank account activity

Status Ever used large bank Currently use large bank Use large bank as main bank

Banked

Own business + banked

94.2 96.3 96.9

93.1 95.7 96.2

Source: Author’s calculation from EFInA (2008). Note: “Banked” means that one has had any type of account at a bank.

172

Lisa D. Cook

Table 4.7

Sources of loans

Source

Borrower

Own business + borrower

8.4 4.9 86.8

6.0 3.7 90.3

Loan from large bank Loan from small bank Loan from other

Source: Author’s calculation from EFInA (2008). Note: Loans from “other” include pool/savings, employer, family, and friends. Table 4.8

Financial indicators, community and microfinance banks

Indicators

2003

2004

2005

2006

2007

2008

Loans/assets Capital/assets Loans/deposits

34.8 24.4 55.1

33.4 24.0 53.3

31.7 22.3 50.9

29.9 23.0 48.5

30.3 28.9 55.6

34.8 30.2 69.9

Source: CBN Annual Report, 2003; CBN BSD, 2004, 2006, 2007, 2008. Note: Data between 2003 and 2005 are for community banks. Data between 2006 and 2008 are for microfinance banks.

cross-section data, it is unclear whether over time small savers and earners have benefited from the recent reforms, but it is clear that most in Nigeria remain unconnected to formal means of saving and borrowing. Some time-series data exist from other sources. The CBN reports that lending by MFBs to manufacturing, transportation, and communications has fallen dramatically as a share of the total, while the share of lending for trading activities has increased from 36 percent to 44 percent between 2001 and 2008. There are no data on lending to households, and it is unclear among sectors of economic activity which may be better for microenterprises and, by extension, poverty alleviation.18 Nonetheless, given innovation and excess demand in transportation (e.g., “okada” motorbikes as a form of transportation) and communications (e.g., rapid increases in penetration of mobile phones), there may exist some missed opportunities among microenterprises. Another indicator of access is the number of banks reporting significant investment in microfinance activities. In 2008, six large banks report having MFBs as subsidiaries or microfinance units in their banks, and four report investing in two MFBs, including Accion MFB Ltd.19 In this instance, too, it is difficult to evaluate changes that may benefit the poor. 18. While manufacturing may be an obvious mechanism for raising employment, it is not obvious that microenterprises are the best candidates among other potential borrowers for credit extension, since many manufacturing activities depend on exploiting economies of scale. 19. Another MFB, AB MFB (Nig.) Ltd., is wholly owned by foreign interests, including two German non-governmental organizations (NGOs), the African Development Bank, and the International Finance Corporation (IFC).

Were the Nigerian Banking Reforms of 2005 a Success?

173

More formal assessments of the reforms have been mixed. Ezeoha (2007) finds evidence of a fundamental change in the financial structure, but suggests that the sustainability of reforms will depend on continuously improving macroeconomic conditions and on public confidence in the government’s commitment to reform. Hesse (2007) analyzes prereform interest-rate spreads and finds that they warranted intervention. However, there is no test of the effect of intervention on postreform spreads. World Bank (2006) uses banklevel data from 2000 to 2005 to test the effect of overhead costs and other covariates on two different measures of interest spreads and finds support for increased efficiency of intermediation among Nigerian banks. Somoye (2008) examines key financial variables and simultaneously rejects the null of no change (e.g., in total assets and bank capitalization) and fails to reject the null of no change (e.g., in efficiency and bank lending) to the private and nonbanking sectors. Onaolapo (2008) evaluates the relation between bank capitalization and financial soundness using data on the Nigerian banking sector from 1990 to 2006. He finds evidence of a positive relation between bank capitalization, distress management, and asset quality. These analyses were able to exploit, at most, one year of postreform data. The passage of time allows a more comprehensive analysis to be undertaken, which is the contribution of this chapter. 4.4

Reform and Changes in Financial Indicators

The empirical strategy in the chapter rests on using different measures of bank efficiency and risk management to test the null hypothesis of no effect on banks and the banking system. It is anticipated that the set of reforms will allow banks to become more efficient at intermediation. 4.4.1 Efficiency A time-series data set is constructed from all banks and financial intermediaries operating in the period 1992 to 2008. Two measures of bank efficiency are used in the empirical analysis. Following Stulz (1999), Demirgüç-Kunt and Levine (1999), and Demirgüç-Kunt and Huizinga (2010), I conjecture that financial development and structure affect firm performance and, more particularly, bank performance. Bank efficiency also depends on overhead cost. One model uses bank efficiency measured by the ex ante interest margin, that is, interest-rate spreads, or the difference in saving and lending rates. Another model uses bank efficiency measured by the ex post interest margin, that is, net interest margins, or the ratio of net interest income to total assets, which accounts for the possibility that banks that charge high interest rates may experience high default rates. To distinguish the effect of cost, development, structure, and bank reform from general economic conditions, macroeconomic variables are included in estimation. Specifically, I estimate the basic regression

174

Lisa D. Cook

(1) EFt = α + βOV/tat + γBDSt + δXt + ζ2005t + η2005 * OV/tat + εt , where EFt is the interest spread or net interest margin at time t; OV/tat is overhead cost scaled by total assets at t; BDSt are financial-development and -structure variables—bank assets/GDP, bank concentration as measured by the ratio of the three largest banks’ assets to total assets, and loans/ liabilities (measures extent of intermediation by the banking system); X t contains measures of opportunity costs and macroeconomic variables— liquid assets/deposits, equity, the real treasury bill rate, inflation rate, and log of industrial production at t; and εt is a random error term.20 It would be important to control for opportunities and opportunity costs associated with liquidity and capital, as Nigerian banks had high levels of both following bank consolidation. There is reason to believe that a mechanism most affected by reforms was cost, since the cost of allocating resources should fall with falling costs associated with economies of scales achieved following mergers and higher levels of capitalization, hence the interaction term.21 The ordinary least squares (OLS) models are executed in the aggregate sample. Due to persistence in the series, models are estimated in first differences with trend. Tests for structural breaks identify 2005 as the break year in the series, and this dummy is included. Standard errors robust to autocorrelation are calculated and reported. Table 4.5 contains mean values of industry- and bank-level variables used in estimation. Table 4.9 reports the results of regressing interest-rate spreads on bank costs, development, and structure. The estimated coefficient on the 2005 year dummy is negative and significant, as anticipated. However, it is also negative and significant for overhead cost/total assets. This result likely reflects temporarily elevated costs associated with restructuring, for example, adoption of the e-FASS system, that nonetheless increased efficiency. This finding is consistent with systematic evidence from an industry study by PwC (2009), which shows that the cost-income ratio fell from 0.65 on average to 0.56 among the largest banks. There is also evidence that banks invested in R&D to make them international with respect to customer access (e.g., ATMs and Internet banking).22 While the sign on the interaction term is negative, it is not significant. 20. The inclusion of a measure of liquidity and of equity follow Martinez Peria and Mody (2004) in their analysis of Latin America. They reflect the fact that high levels of liquidity will increase costs that are passed on to borrowers and increase spreads, while high levels of capital will impose opportunity costs with respect to equity. 21. Among the reforms implemented was the introduction of an electronic financial analysis and surveillance system (e-FASS), which permitted automated submission of returns by financial institutions. Adoption of such technology and related technological innovations should aid in mitigating operating costs in information-poor environments. 22. See Kasekende et al. (2009) for a review of banking competitiveness in Africa’s four largest economies.

Were the Nigerian Banking Reforms of 2005 a Success? Table 4.9

Bank efficiency estimation. Dependent variable: Interest spread (percentage points)

Explanatory variables Year ≥ 2004 Overhead cost/total assets

(1)

(2)

(3)

(4)

(5)

–6.100* (3.342) –273.235*** (88.680)

–5.729* (3.225) –268.070*** (76.574) –22.968 (23.232)

–6.980** (3.177) –284.957*** (69.467) –15.324 (19.820) –26.619 (18.843)

–5.159 (3.909) –270.452*** (78.566) –28.406 (23.867)

–6.523* (3.914) –285.726*** (71.941) –19.462 (21.532) –25.246 (20.804) –19.662 (21.551)

Bank assets/GDP Bank concentration Overhead cost/total assets × Year ≥ 2004 Time trend F N

175

–28.558 (29.596) Yes 5.828 16

Yes 3.999 16

Yes 4.903 16

Yes 3.979 16

Yes 3.987 16

Source: CBN; Beck, Demirgüç-Kunt, and Levine (2009). Note: Results are reported for OLS models. Data are for 1992 to 2008. Newey-West robust standard errors for autocorrelation are in parentheses. Interest spread, overhead cost/total assets, bank assets/GDP, and bank concentration are first-differenced. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Table 4.10 reports the results of regressing interest-rate spreads on an alternative measure of bank development, controls for opportunity costs related to excess liquidity and capital, and macroeconomic variables. The estimated coefficients from this specification are similar to those reported in table 4.9. The macroeconomic variables have the anticipated effect on interest spreads, although the estimated coefficient on the inflation rate is significant. Table 4.11 reports the results of regressing net interest margin on bank costs, development, and structure. Again, the estimated coefficient on overhead cost/total assets is positive and significant. However, from Beck, Demirgüç-Kunt, and Levine (2009) this finding is consistent with those from low-income countries where bank overhead costs/total assets and net interest margin are correlated and both high relative to middle- and high-income countries. The results on overhead cost are also very close in size, direction, and significance to those of World Bank (2006), which uses Nigerian bank data for 2001 to 2005 and includes additional controls for excess liquidity and capital (see above) and macroeconomic indicators. Demirgüç-Kunt and Huizinga (1999), find a positive relation between cost and net interest margin using data from eighty countries. From the positive and significant coefficient on the interaction term, this relation was stronger after 2004.

Yes 4.928 16

–25.166* (14.972) 0.000003 (0.000)

–8.086** (3.406) –283.811*** (87.277)

Yes 6.130 16

–29.035 (18.275) 0.000003 (0.000) –1.575 (20.268) –0.042 (0.033)

–8.594** (3.662) –302.773*** (86.989)

(2)

Yes 6.467 16

0.058** (0.025)

–31.574* (17.560) 0.000003 (0.000) –0.498 (20.030)

–8.770** (3.604) –320.082*** (85.901)

(3)

Yes 22.033 16

–18.344 (22.527)

–26.194* (15.445) 0.000001 (0.000) 5.589 (23.057)

–7.879** (3.745) –268.028** (110.350)

(4)

Yes 8.129 16

–7.819** (3.778) –282.775*** (92.193) –10.124 (11.568) –26.305 (16.948) 0.000 (0.000)

(5)

Yes 9.938 16

–8.290** (4.119) –301.656*** (93.412) –11.699 (14.775) –30.438 (21.000) 0.000 (0.000) –1.448 (21.755) –0.042 (0.035)

(6)

Yes 9.694 16

0.061** (0.027)

–8.385** (4.096) –320.213*** (92.335) –15.875 (14.921) –33.718* (20.348) 0.000 (0.000) –0.416 (21.439)

(7)

Yes 36.281 16

–19.030 (24.233)

–7.498* (4.218) –265.844** (118.484) –14.066 (12.502) –27.841 (17.434) 0.000 (0.000) 5.919 (24.695)

(8)

Source: For overhead cost/total assets, spread: Beck, Demirgüç-Kunt, and Levine (2009); liquidity assets/deposit, equity, loans/liabilities: Author’s calculation, CBN, Statistical Bulletin (2007, 2008); real Treasury bill rate, inflation, industrial production index: IMF (2010) IFS. Note: Results are reported for OLS models. Newey-West robust standard errors for autocorrelation are in parentheses. Interest spread, overhead cost/total assets, bank assets/GDP, and bank concentration are first-differenced. Data are for 1992 to 2008. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Time trend F N

Log of industrial production index

Inflation

Real Treasury bill rate

Loans/liabilities

Equity

Overhead cost/total assets × Year ≥ 2004 Liquid assets/deposits

Overhead cost/total assets

Year ≥ 2004

(1)

Bank efficiency estimation. Dependent variable: Interest spread (percentage points)

Explanatory variables

Table 4.10

Were the Nigerian Banking Reforms of 2005 a Success? Table 4.11

Bank efficiency estimation. Dependent variable: Net interest margin

Explanatory variables Year > = 2004 Overhead cost/total assets Bank assets/GDP Bank concentration Overhead cost/total assets × Year ≥ 2004 Constant Time trend F N

177

(1)

(2)

(3)

(4)

(5)

(6)

(7)

–0.004 0.004 0.004 0.005 0.003 (0.012) (0.009) (0.010) (0.010) (0.009) 1.080*** 1.071*** 1.091*** 1.088*** (0.227) (0.250) (0.247) (0.237) 0.038 0.029 (0.058) (0.055) 0.031 (0.056) 0.066* (0.035)

0.002 0.003 (0.010) (0.010) 1.080*** 1.095*** (0.265) (0.262) 0.057 0.048 (0.062) (0.059) 0.025 (0.059) 0.100** 0.091** (0.045) (0.045)

0.041 0.004 (0.050) (0.047)

0.002 (0.053)

Yes 2.602 16

Yes 47.363 16

0.006 (0.051) Yes 27.427 16

0.020 (0.069) Yes 20.585 16

0.000 (0.050) Yes 46.045 16

Yes 22.118 16

0.013 (0.073) Yes 15.438 16

Source: CBN; Beck, Demirgüç-Kunt, and Levine (2009). Note: Results are reported for OLS models. Data are for 1992 to 2008. Net interest margin, overhead cost/total assets, bank assets/GDP, and bank concentration are first-differenced. Newey-West robust standard errors for autocorrelation are in parentheses. Data are for 1992 to 2008. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

4.4.2

Risk Management and Bank Performance

If there is less distress in the system, for example, a lower share of nonperforming loans to the total, it is anticipated that banks would require fewer resources for loan-loss provisioning. Alternatively, the ratio of loanloss provisions to total loans may indicate portfolio quality. The bank-level sample is constructed from banks existing in 2006.23 The sample period is 2001 to 2008. In these regressions the dependent variable is the provision of bad debt to total loans and advances. Specifically, the model estimated is (2)

PBDit = α + βBANKit + γBDSt + ζ2005t + ηMACROt + εit ,

where PBDit is provision for bad debt/total loans and advances, and the indicators of bank development and bank structure are bank assets to GDP and bank concentration; BANKit comprises controls for bank-specific characteristics: the ratio of equity to lagged total assets, dummies for new (or merged) bank and foreign ownership, and bank and year dummies. The 23. A sample is constructed from variables available from uniform reports on individual large banks available from 2001.

178

Lisa D. Cook

macroeconomic indicators included in MACROt are GDP growth rate and inflation rate. To account for a structural break in the dependent variable, Driscoll-Kraay standard errors are calculated and reported with estimated coefficients in table 4.12. I find that the estimated coefficient on the year 2005 is not significant when accounting for bank and macroeconomic characteristics. Consistent with the findings of Demirgüç-Kunt and Huizinga (1998) and Bikker and Hu (2002), this evidence is suggestive that asset quality will be positively affected by higher growth and lower inflation. Further, it is consistent with Beny and Cook (2009), which shows that better economic

Table 4.12 Explanatory variables

Performance, risk management estimation. Dependent variable: Provision of bad debt/loans and advances (1)

(2)

Year ≥ 2004

–0.185** (0.072)

–0.185** (0.072)

New bank

0.182*** (0.031) 0.055* (0.032)

Equity/lagged total assets Foreign ownership

(3) –0.019* (0.011)

(4) –0.019* (0.011)

Bank characteristics –0.014 0.184*** –0.018 (0.042) (0.033) (0.043) 0.055* 0.050 0.050 (0.032) (0.035) (0.035) –0.257*** 0.040 (0.027) (0.028)

(5)

(6)

(7)

–0.001 (0.004)

0.006 (0.006)

–0.007 (0.006)

–0.012 (0.043) 0.065** (0.029) 0.022 (0.028)

–0.010 (0.043) 0.067** (0.029) –0.257*** (0.029)

–0.016 (0.042) 0.058** (0.028) –0.258*** (0.027)

–0.390*** (0.047) –0.404*** (0.145)

–0.372*** (0.046) –0.322** (0.156)

–0.427*** (0.017) –0.159 (0.139)

Financial structure Bank assets/GDP Bank concentration Macroeconomic indicators Growth rate, percent Inflation rate, percent Constant Bank dummies Year dummies R2 N

0.001** (0.001)

0.116*** (0.035)

0.312*** (0.047)

0.131*** (0.032)

0.034*** (0.007)

0.258*** (0.055)

0.483*** (0.067)

–0.002*** (0.000) 0.483*** (0.066)

Yes Yes 0.606 150

Yes Yes 0.606 150

Yes No 0.541 150

Yes No 0.541 150

Yes No 0.596 149

Yes No 0.598 149

Yes No 0.601 149

Source: For provision of bad loans, loans and advances, equity, total assets: CBN; bank assets/GDP and bank concentration: Beck, Demirgüç-Kunt, and Levine (2009); growth rate, inflation: Author’s calculation, IMF (2010) IFS. Note: Results are reported for pooled OLS models. Robust Driscoll-Kraay standard errors are in parentheses. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Were the Nigerian Banking Reforms of 2005 a Success?

179

management was correlated with Africa’s growth spurt in the early twentyfirst century. The evidence suggests that a larger capital base is correlated with the ability to manage expected bankruptcy costs. The findings from this study are largely in line with previous studies. Only one major result is challenged by the analysis in this chapter—Somoye’s (2008) failure to reject the null hypothesis of significant change in bank efficiency and lending to the private sector. Tests of structural breaks require testing the relative significance of several adjacent years to the candidate year, and earlier tests would not have had post-2005 data available for checking maximum significance. One limitation of this analysis and that of others is that recorded distress may have been artificially low. Recall that CBN reported no banks in distress between 2006 and 2008, despite such signals as a decline of 46 percent in the Nigerian All-Share Index in 2008, which is shown in figure 4.8. Nigerian banks were overextended in 2008 when the loans-to-deposits ratio exceeded 1.5 (see figure 4.6), which was similar to that of Western banks prior to and during the financial crisis of 2008. This should have alarmed bank regulators earlier than it did. In late 2009, two stress tests were executed on the twenty-four banks, and significant distress and poor corporate governance were identified in the banking system. Ten banks failed the tests, and the CBN determined that the system required a capital and liquidity injection of $4.12 billion. It is estimated that rescued banks held NGN 2 trillion ($13.3 billion) in “toxic” loans. Executives at one-third of all banks were forcibly removed and arrested. As aforementioned, while corporate governance was among the reform items, it did not feature prominently in the reform program prior to 2009. These events suggest that the reforms of 2005 may have protected the banking system from a worse crisis than may have evolved, but that additional reform on the part of the banks and the central bank is required. In particular, it seems that better internal controls and bank monitoring are warranted, along with timely and relevant determination of distress. 4.5

Conclusion

This chapter examines the consequences of major reforms undertaken in the Nigerian banking sector in 2005. Firm-level and time-series data allow the extension of analysis conducted shortly after the reforms. A statistically significant break is identified in most financial series at the year 2005. There is increased banking-sector development and greater competition and less government intervention in bank activity. Further, I find that banks are more efficient than prior to the reforms and that this change is correlated with overhead costs generally and from 2005. Asset quality increased between 2005 and 2008, which is correlated with a higher capital base. Favorable macroeconomic conditions likely enhanced the impact of the reforms.

180

Lisa D. Cook

Nonetheless, it is ambiguous whether changes in the microfinance sector aided the poor. While the reforms of 2005 increased safety and soundness by several measures, the analysis suggests that bank distress was un- or underreported after 2005. Along with nonperforming loans, corporate governance, which received less attention than other reforms, may require more examination and oversight than in the recent past.

References Abiad, Abdul, and Ashoka Mody. 2005. “Financial Reform: What Shakes It? What Shapes It?” American Economic Review 95:66–88. Adegbite, E. O. 2005. “Financial Sector Reforms and Economic Development in Nigeria: The Role of Management.” Paper delivered at the inaugural National Conference of the Academy of Management Nigeria. Barth, James R., Gerard Caprio, Jr., and Ross Levine. 2001. “The Regulation and Supervision of Banks around the World, a New Database.” World Bank Policy Research Working Paper no. 2588, Washington, DC, World Bank. http://www -wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2001/06/01 /000094946_01052204005474/additional/112512322_20041117140525.pdf. Beck, Thorsten. 2004. “The Determinants of Financing Obstacles.” Policy Research Working Paper no. 3204, World Bank. Beck, Thorsten, Robert Cull, and Afeikhena Jerome. 2005. “Bank Privatization and Performance: Empirical Evidence from Nigeria.” Journal of Banking and Finance 29 (8–9): 2355–79. ________. 2008. “Banks Regulations are Changing: For Better or Worse?” Policy Research Working Paper Series no. WP 4646, Washington, DC, World Bank. Beck, Thorsten, Asli Demirgüç-Kunt, and Ross Levine. 2000. “A New Database on the Structure and Development of the Financial Sector. World Bank Economic Review, World Bank Group 14 (3): 597–605. ———. 2009. “Financial Institutions and Markets across Countries and over Time: Data and Analysis.” Policy Research Working Paper no. 4943, World Bank. Beck, Thorsten, Asli Demirgüç-Kunt, Ross Levine, Martin Cihak, and Erik H. B. Feyen. 2009. Financial Development and Structure Dataset.” Finance Research Group, World Bank. http://econ.worldbank.org/WBSITE/EXTERNAL/EXT DEC/EXTRESEARCH/0,,contentMDK:20696167~pagePK:64214825~piPK:64 214943~theSitePK:469382,00.html. Beck, Thorsten, and Michael Fuchs. 2004. “Structural Issues in the Kenyan Financial System: Improving Competition and Access.” Policy Research Working Paper Series no. WP 3363, Washington, DC, World Bank. Beny, Laura, and Lisa D. Cook. 2009. “Metals or Management? Explaining Recent Economic Growth in Africa.” American Economic Review 99 (2): 268–74. Bikker, J., and H. Hu. 2002. “Cyclical Patterns in Profits, Provisioning and Lending of Banks and Procyclicality of the New Basel Capital Requirements.” BNL Quarterly Review 221:143–75. Caprio, Jr., Gerard, and Daniela Klingebiel. 1996. “Bank Insolvencies: CrossCountry Experience.” World Bank Policy Research Working Paper no. WPS 1620, Washington, DC, World Bank.

Were the Nigerian Banking Reforms of 2005 a Success?

181

Central Bank of Nigeria. Annual Report and Statement of Accounts, various years, 2000–2008. ________. Banking Supervision Annual Reports, various years 2000–2008. ________. Statistical Bulletins, various years, 2000–2008. ________. 2005. Microfinance Policy, Regulatory and Supervisory Framework for Nigeria. http://www.cenbank.org/out/Publications/guidelines/dfd/2006/micro finance%20policy.pdf. Cook, Lisa D. 2004. “Financial Crisis and Growth in Nigeria: Evidence from the Community Banking System.” Unpublished Manuscript, Stanford University. Demirgüç-Kunt, A., and H. Huizinga. 1999. “Determinants of Commercial Bank Interest Margins and Profitability: Some International Evidence.” World Bank Economic Review 13:379–408. ———. 2010. “Bank Activity and Funding Strategies: The Impact on Risk and Returns.” Journal of Financial Economics 989 (3): 626–50. Demirgüç-Kunt, A., and R. Levine. “Bank-Based and Market-Based Financial Systems—Cross-Country Comparisons.” Policy Research Working Paper no. 2143, World Bank. Ezeoha, Abel. 2007. “Structural Effects of Banking Industry Consolidation in Nigeria: A Review.” Journal of Banking Regulation 8 (2): 159–76. FAO. 2004. “Federal Republic of Nigeria: Assessment of Community Banks.” September 20. http://www.microfinancegateway.org/sites/default/files/mfg-en-paper -federal-republic-of-nigeria-assessment-of-community-banks-sep-2004_0.pdf. EFInA. 2008. National Survey of Nigerian Households on Consumer Finance (data set). Lagos, Nigeria: Enhancing Financial Innovation & Access. Hesse, Heiko. 2007. “Financial Intermediation in the Pre-Consolidated Banking Sector in Nigeria.” Policy Research Working Paper Series no. WP 4267, Washington, DC, World Bank. International Monetary Fund (IMF). 2009. World Economic Outlook. http://www .imf.org/external/pubs/ft/weo/2009/01/index.htm. ———. 2010. International Financial Statistics. Accessed July 2010. Ehttp://www .imfstatistics.org Isern, Jennifer, Amaka Agbakoba, Mark Flaming, Jose Mantilla, Giulia Pellegrini, and Michael Tarazi. 2009. “Access to Finance in Nigeria: Microfinance, Branchless Banking, and SME Finance.” Consultative Group to Assist the Poor. http:// www.cgap.org/publications/access-finance-nigeria-microfinance. Kasekende, Louis, Kupukile Mlambo, Victor Murinde, and Tianshu Zhao. 2009. “Restructuring for Competitiveness: The Financial Services Sector in Africa’s Four Largest Economies.” In the Africa Competitiveness Report 2009, African Development Bank/World Economic Forum, 49–81. King, Robert G., and Ross Levine. 1993a. “Finance, Entrepreneurship and Growth: Theory and Evidence.” Journal of Monetary Economics 35:513–42. __________. 1993b. “Finance and Growth, Schumpeter Might be Right.” Quarterly Journal of Economics 108:717–38. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. “Financial Intermediation and Growth: Causality and Causes.” Journal of Monetary Economics 46 (1): 31–77. Levine, Ross, and Sara Zervos. 1998. “Stock Markets, Banks, and Economic Growth.” American Economic Review 88:537–58. Lewis, P., and H. Stein. 1997. “Shifting Fortunes: The Political Economy of Financial Liberalization in Nigeria.” World Development 25 (1): 5–22. Martinez Peria, Maria Soledad, and Ashoka Mody. 2004. “How Foreign Participation and Market Concentration Impact Bank Spreads: Evidence from Latin America.” Journal of Money, Credit, and Banking 36 (3): 511–37.

182

Lisa D. Cook

Nigerian Deposit Insurance Corporation. Annual Reports, various years, 2000–2006. Ogunleye, G. A. 2003. “The Causes of Bank Failures and Persistent Distress in the Banking Industry.” NDIC Quarterly 13 (4): 21–41. Onaolapo, A. A. 2008. “Implications of Capital Regulation on Bank Financial Health and Nigerian Economic Growth 1996–2005.” Journal of Economic Theory 2 (3): 112–7. PriceWaterhouseCoopers (PwC). 2009. “Balance Sheet Management Benchmark Survey: Status of Balance Sheet Management Practices among International Banks —2009.” http:// www.pwc .com/ gx/ en/ banking-capital-markets/ assets /balance-sheet-management-benchmark-survey.pdf. Rajan, Raghuram G., and Luigi Zingales. 1998. “Financial Dependence and Growth.” American Economic Review 88:559–86. Somoye, R. O. C. 2008. “The Performances of Commercial Banks in PostConsolidation Period in Nigeria: An Empirical Review.” European Journal of Economics, Finance and Administrative Sciences December (14). http://www.european journalofeconomicsfinanceandadministrativesciences.com/ejefas_issues.html. Soludo, Charles. 2004. “Consolidating the Nigerian Banking Industry to Meet the Development Challenges of the 21st Century.” Address delivered to the Special Meeting of the Bankers’ Committee, July 6, Abuja, Nigeria. http://www.bis.org /review/r040727g.pdf. Stulz, René M. 1999. “Globalization, Corporate Finance, and the Cost of Capital.” Journal of Applied Corporate Finance 12 (3): 8–25. World Bank. 2006. Nigeria: Competitiveness and Growth, Appendixes. September 26. Washington, DC, World Bank.

5

Misallocation, Property Rights, and Access to Finance Evidence from within and across Africa Sebnem Kalemli-Ozcan and Bent E. Sørensen

5.1

Introduction

A number of recent studies argue that misallocation of resources across firms is a prime cause of underdevelopment. Standard theory implies that if domestic capital markets are functioning well, the marginal product of capital (MPK) of each firm equals the market interest rate. If firms instead borrow at different interest rates, maybe due to differential access to informal finance or due to political connections, capital is likely to be misallocated and the MPK will differ across firms. Alfaro, Charlton, and Kanczuk (2008), Banerjee and Duflo (2005), Bartelsman, Haltiwanger, and Scarpetta (2009), Hsieh and Klenow (2009), and Restuccia and Rogerson (2008) provide evidence of misallocation in different countries and show that misallocation of resources can explain up to 60 percent of the aggregate total factor productivity (TFP)-differences between poor and rich countries. Differential access to credit may not necessarily lead to severe misallocation if firms with higher MPK invest more, as Banerjee and Moll (2010) point out. However, in the absence of secure property rights owners may not reinvest profits: even if the return to investment is high, government officials may grab a large share of earnings, dilutSebnem Kalemli-Ozcan is professor of economics at the University of Maryland and a research associate of the National Bureau of Economic Research. Bent E. Sørensen is the Lay Professor of Economics at the University of Houston. This chapter is prepared for the NBER-Africa Project. We thank Sevcan Yesiltas for superb research assistance. We also thank Mary Hallward-Driemeier, Simon Johnson, Marianna Sørensen, and participants at NBER-Africa Project conferences in Cambridge and Zanzibar for comments. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13360.ack.

183

184

Sebnem Kalemli-Ozcan and Bent E. Sørensen

ing the incentive of owners to reinvest. Johnson, McMillan, and Woodruff (2002) find exactly such behavior in Russia and Ukraine after the breakdown of communism. We ask two questions in this chapter: What is the extent of capital misallocation within African countries, and why does misallocation vary across these countries? We quantify capital misallocation across manufacturing establishments within ten African countries in 2005 and 2006 using establishment-level data from the World Bank Productivity and Investment Climate Survey.1 This is a unique survey undertaken as part of a major World Bank initiative between 1999 and 2007 in eighty developed and developing countries around the world. The main purpose of the survey was to identify obstacles to firm performance and growth; hence the survey not only asks questions on firm characteristics and outcomes, but also asks many questions on the perceived severity of obstacles such as crime, infrastructure, and financing constraints. Having firms’ own perceptions of financing constraints is a big advantage of the data set because much of the literature infers financing constraints from companies’ financial statements using various modeling and econometric techniques. This data set has been used by, among others, Beck et al. (2006) and Beck, Demirgüç-Kunt, and Maksimovic (2005), who show that these self-perceived constraints actually bind and hurt firm growth. Our data set has information on small and large, as well as listed and private firms, which allows us to control for some important firm characteristics. To the best of our knowledge, there is no systematic study undertaken that calculates the extent of misallocation and its determinants for Africa using comparable firm-level data from many countries. In the literature, there are various approaches employed for calculating the extent of misallocation of capital across firms within a country. As stated above, one of the advantages of our data set is that it allows us to compare the interest rates firms are paying with the market interest rate. This is our starting point because we have data on the interest rates each firm pays on loans. We show that many firms borrow at rates up to 30–40 percent, suggesting that firms have even higher marginal returns to capital.2 We calculate the MPK for each firm using firm-level output and capital stocks under the assumption that the production function is Cobb-Douglas (with parameters calibrated from the literature). Doing so reveals that the distribution of the MPK varies a lot within most African countries. This indicates that capital is inefficiently allocated—a fact that cannot be derived from country-level aggregate figures. We next calculate a measure of mis1. Here, “establishment” refers to a production unit that may be part of a larger firm, but for simpler reading we will also use the term “firm.” 2. Banerjee (2002) displays similar evidence for other developing countries. He emphasizes that these rates must be the rates that firms actually pay because default is rare.

Misallocation, Property Rights, and Access to Finance

185

allocation suggested by Hsieh and Klenow (2009), and this measure also indicates imperfect capital and/or labor allocation.3 Having calculated the extent of misallocation, we seek to explain firmlevel differences in returns to capital within countries and the variation in misallocation across countries. First we show, using multiple regressions, that firms with less access to finance have higher MPK. Small firms have lower MPK (conditional on access to finance and other regressors), indicating that higher efficiency could be attained by allocating more capital to large firms. Moving from a firm where access to finance is no obstacle to a firm where access to finance is a very severe obstacle increases the MPK by 45 percent, revealing that obstacles to credit have important real effects. Second, we find a clear positive correlation between country-level misallocation and the strength of property rights, measured using expropriation risk and investment profile variables from the International Country Risk Guide (ICRG). These variables help explain the variation in misallocation across African countries consistent with the patterns found by Johnson, McMillan, and Woodruff (2002) for former communist countries. Hence, we contribute to the recent debate on “what works in Africa?” in the following sense. Once we calculate the extent of misallocation using different methodologies, we can explain the determinants of this misallocation at the firm level and relate country variation in misallocation to the broader investment climate and business environment. This, in turn, helps us answer why certain countries have better allocation of capital across firms; that is, we can identify relatively successful countries, such as South Africa and Botswana, relative to unsuccessful ones, such as Ghana and Nigeria, and suggest reasons behind their success. We use very simple measures of misallocation. In the process of writing the chapter, a large amount of measures were considered, using different production-function parameters depending on labor and capital type. These more complicated measures produced very noisy patterns and served little purpose. We believe that the lesson from this nonreported work is that fairly underdeveloped economies face many unmeasured obstacles, which obscure patterns in anything but simple straightforward measures.4 It may be the case that some firms pay higher interest rates due to risk premiums, and it may be the case that the simple functional forms we use to measure the MPK are misspecified, making our measures of misallocation noisy. We therefore 3. We attempted a final approach by estimating the correlation between productivity and size (see Alfaro, Charlton, and Kanczuk 2008; Bartelsman, Haltiwanger, and Scarpetta 2009); however, we did not find any clear patterns. 4. We studied alternative measures of labor cost (separating full-time, part-time, temporary, and nonproduction workers), other measures suggested by Hsieh and Klenow (2009), and more narrow indicators of financing constraints, such as use of collateral. We also attempted to include both manufacturing and nonmanufacturing firms.

186

Sebnem Kalemli-Ozcan and Bent E. Sørensen

compare the statistics calculated for African countries to corresponding statistics calculated for a selection of non-African countries at different levels of development— namely, Germany, Ireland, Spain, South Korea, and India. This comparison reveals that standard deviations across firms of all our misallocation measures are much larger in Africa. For example, the standard deviation of the interest rate is 2–5 times higher in African countries than in European countries and the standard deviation of the MPK is about 40 percent higher in African countries (and in India) than in European countries. More than 50 percent of firms in Africa report that access to finance is a severe obstacle, while very few firms in Europe report this as a severe obstacle. The rest of the chapter is structured as follows. Section 5.2 reports on our field trip to Ghana, a country with a high level of misallocation. Section 5.3 describes our data in detail, while section 5.4 presents results from our empirical analysis. Section 5.5 concludes. 5.2

Observations from Investigators’ Trip to Ghana

The authors visited Ghana in May 2011 and interviewed several people familiar with local conditions, such as academics and foreign entrepreneurs. Foreign firms are concentrated in Accra, the capital of Ghana, in a free trade zone that has reliable electricity (although many companies in Ghana rely on generators) and, most importantly, a large modern harbor that allows for easy shipping. Foreign entrepreneurs finance investments with retained earnings or nonlocal financing because contract enforcement in Africa is weak. Most projects are done with a 50 percent down payment up front. The main attraction by far of investing in Ghana (relative to alternative subSaharan countries) is political stability, although a reliable local workforce is another plus. It was mentioned that workers from some other African countries are considered less reliable. One multinational corporation located production in Ghana due to local demand for its product from other foreign companies operating in Ghana and sub-Saharan Africa. This corporation was originally shipping its product from an affiliate outside of Africa but could not keep up with the orders—the motivation for shipping from afar was put as: “Nobody wants to buy something made in Africa because quality is perceived to be poor.” Foreign companies have to obey a 70 percent local content requirement, which means 70 percent of the workforce should be Ghanaians. This constitutes a problem because the local workforce lacks basic skills; for example, plumbers are hard to find. The companies bring in high-tech personnel from India and the Philippines or from the United States (although Americans sometimes do not want to stay) to train the local workforce. This, however, is costly, being very time intensive. Foreign entrepreneurs try to circumvent the 70 percent requirement by other means (one example given was plead-

Misallocation, Property Rights, and Access to Finance

187

ing with officials) in order to get things done. Companies import all capital goods and intermediate goods from the United States and other developed countries. There was general agreement that access to capital through formal channels, such as banks, is severely limited in particular due to lack of clear property rights to land. Being unable to use land as collateral makes it difficult for small businesses to get loans. Microloans (informal) are often available but annual rates are very high, often above 50 percent. One US multinational company owner said that the main reason, more important than infrastructure, for investing in a factory in the free trade zone was that the land is owned by the government—the company paid for a forty-sevenyear lease in advance. Local firms are shut out from financial intermediation and borrow from family or local unofficial lenders. Banks mainly serve the government. Small-scale corruption is another major problem. (Maybe also large-scale corruption, although we did not learn about that.) Mango producers in the north of Ghana were not able to get fruit to the market in Accra without paying prohibitive bribes at police check points, which also slow down trucking on the already inadequate roads (by US standards; according to the foreign entrepreneur, the roads are good by Africa standards). As we understood, police bribes are not particularly large, maybe a few dollars, but with enough checkpoints, it becomes unprofitable to transport low-margin goods over any substantial distance. In the descriptive statistics tables to be discussed later, we show numbers for Ghana and for African countries pooled. 5.3 5.3.1

Data Productivity and Investment Climate Survey

The firm-level data comes from the Productivity and Investment Climate Survey of the World Bank,5 administered in roughly parallel fashion to enterprizes in twenty-one countries in Africa, mostly in face-to-face interviews. The data set provides a basis for making country comparisons of investment climate and severity of constraints affecting firms. It captures firms’ perceptions of key constraints in the business environment that shape operational and investment decisions, as well as several quantitative indices of firm experience. The first roll out of surveys was done in 2006 for thirteen countries: Burundi, Congo, Botswana, Angola, Guinea Bissau, Guinea-Conakry (or Republic of Guinea), Namibia, Gambia, Mauritania, Swaziland, Tanzania, Uganda, and Rwanda. In 2007, a second roll out was conducted in eight additional countries: South Africa, Mozambique, Zambia, Mali, Ghana, 5. The data and related documents are available at http://www.enterprisesurveys.org/.

188

Sebnem Kalemli-Ozcan and Bent E. Sørensen

Senegal, Kenya, and Nigeria. Questionnaires of the two roll outs are not systematically different, except that the second questionnaire generally has more detailed questions. The World Bank also surveys some developed and emerging market countries, but the structure of the questionnaires is somewhat different from that used in the African surveys. For comparison with Africa, we choose Germany, India, Ireland, South Korea, and Spain.6 The data set for African countries, merging the two roll outs, has information on 12,752 establishments. For the comparison countries, we have data for 1,196 German, 2,286 Indian, 501 Irish, 598 South Korean, and 606 Spanish establishments. Enterprizes with five to nineteen, twenty to ninety-nine, and over one hundred employees are labeled small, medium, and large, respectively. The Productivity and Investment Climate Survey comprises four sets of questionnaires, which are particularly designed for the following sectors: manufacturing, retail, residual (out of manufacturing and retail), and micro (also called the informal sector). Each questionnaire has several sections in which detailed information is given. In related surveys, entrepreneurs provided general information including legal status (e.g., proprietorship); the percentage owned by the largest shareholder; private, foreign, or government ownership; sex and ethnic origin of the majority owner; level of education and experience of the top manager; when the firm was established; and whether it was formally registered (section A). The survey also provides information on sales and exports (section C), supplies and import (section D), capacity and innovation (section E), investment climate constraints (section F), infrastructure (G), conflict resolution/legal environment (section H), business-government relations (section I), labor regulation (section J), finance (section K), and productivity (section L). The data was collected using similar survey-sampling methodologies because one of the main objectives in establishing this database is to provide a wide set of measures of firm outcomes and structures that are comparable across countries. The database is mainly a stratified sampling of firms from a representative sample provided by the national statistical offices. If this is not available, stratification is done on a randomly drawn sample. Sample stratification is based on having a third of the data be represented by each size group. Representation of several sectors was also an objective.7 5.3.2

Questions on Obstacles

The main question on obstacles is: Do you think the following (X) present any obstacle to the current operations of your establishment? The answers 6. The World Bank also surveys Brazil, China, and Turkey. However, the structure of those surveys is too different from that of the African surveys to allow us to make comparisons. 7. The World Bank provides sample selection notes giving detailed information on sampling methodologies for the Enterprise Surveys. Some notes are available at http://www.enterprise surveys.org/. Details for the Africa sample are available from the authors by request, but sample selection notes are not available for Germany, India, Ireland, South Korea, and Spain.

Misallocation, Property Rights, and Access to Finance

189

are no obstacle, minor obstacle, moderate obstacle, major obstacle, and very severe obstacle, which are assigned the numerical values 1, 2, 3, 4, and 5, respectively. We have averaged answers to the question stated above into four groups: limited access to finance, weak infrastructure, weak law and order, and red tape. Weak infrastructure is the average of answers to this question where X is “electricity,” “telecommunications,” “transportation,” and “access to land.” Red tape is the average of answers to this question where X is “tax rates,” “tax administration,” “customs and trade regulations,” “labor regulations,” and “business licensing and permits.” Weak law and order is the average of answers to this question where X is “functioning of the courts,” “political instability,” “corruption,” “macroeconomic instability,” “crime, theft, and disorder,” and “practices of competitors in the informal sector.” Weak law and order and red tape are coded such that higher values correspond to less law and order and more red tape. For Indian firms, the answers vary between 0 (no obstacle), 1 (minor obstacle), 2 (moderate obstacle), 3 (major obstacle), and 4 (very severe obstacle). 5.3.3

Construction of Misallocation Measures

The variables we use from the Investment Climate Survey are annual interest rates (self reported), sales, capital stock at current replacement cost, labor, total cost of materials and intermediate inputs, total capital income, and total cost of labor. Variables in domestic monetary values are converted into US dollars using the annual exchange rates from World Development Indicators.8 The definitions are as follows:

• Annual nominal interest rate (R): For annual nominal interest rates, we •

• • •

directly use the information on interest rates paid on loans.9 Annual real interest rate: To calculate real interest rates, we subtract inflation of the year the surveys are conducted. The inflation rate, obtained from the International Monetary Fund, is the annual percent change in consumer prices. Value added (Y): Value added is constructed as total sales minus total cost of raw materials and intermediate goods used in production. Replacement cost value of capital stock (K): Historical cost of replacing all machinery and equipment with new machines. Labor measure (L): We use information on the total number of full-time permanent employees at the end of the survey year to proxy labor used in the production process. Permanent employees are defined as all paid employees that work eight or more hours per day with a contract for a

8. We noticed that monetary values reported in the domestic currency of Ghana are equal to the ones supposedly converted to US dollars. In order to fix that, we multiplied monetary values in the domestic currency of Ghana by 0.00011, the annual dollar exchange rate of Ghana in 2006. 9. The question is as follows: Does your establishment currently have a line of credit or loan from a financial institution? If so, what is the average annual interest rate?

190

Sebnem Kalemli-Ozcan and Bent E. Sørensen

term of one or more fiscal years and/or have a guaranteed renewal of their employment contract. • Total cost of labor (wL): Includes wages, salaries, bonuses, and social payments. • Total capital income (RK): We multiply the replacement cost of capital (K) with R, which is taken as 15 percent. Hsieh and Klenow (2009) use a value of 10 percent, but because the average nominal interest rate for our African sample is about 15 percent, we choose this higher value. For our benchmark samples, the average nominal interest rates are given in table 5.3. Using the above variables, we calculate two measures of misallocation previously used in the literature. We follow Hsieh and Klenow (2009) and outline the pertinent features of their model here. Assume that aggregate output (or, in Hsieh and Klenow, sectoral output) is a CES index of differentiated outputs of firms i = 1, . . . , M; that is, M Yi (s−1)/s )s/(s−1) , with the production of each differentiated product Y = (Si=1 given by a Cobb-Douglas production function Yi = Ai Kia L1−a i , where Ai is firm-level TFP, Ki is capital input, and Li is labor. Profits are p = (1 − τ yi ) PiYi − wLi − (1 + τ Ki ) RKi ,

where Pi is the price of output and R is the rental price of capital; tyi is an output distortion, such as a tax on firm i’s output, which does not affect the relative choice of capital and labor; tyi is allowed to vary by firm and is intended to capture distortions such as corruption or any other impediment to production of firm i, which affects output but is not tied to capital or labor; and tKi captures access to credit. A positive value indicates that a firm pays a higher cost of capital than the official interest rate R, for example, because the firm only has access to informal credit at high rates. Profit maximization gives price as a markup over marginal cost: Pi = [s / (s − 1)](R / a)a [w / (1 − a)](1−a) [(1 − τ Ki )a ] / Ai (1 − τ yi ). The capital-labor ratio is then (1)

a w 1 Ki = , Li 1 − a R 1 + τ Ki

which reflects the relative capital/labor distortion. The marginal revenue product of capital (denoted MRPK) is (2)

MRPKi = a

s − 1 PiYi 1 + τ Ki =R , s Ki 1 − τ yi

which is larger, the larger the output distortion and the larger the capital/ labor distortion.

Misallocation, Property Rights, and Access to Finance

191

Based on these considerations, we use the following measures of misallocation 1. MPK: MPKi = a

(3)

PiYi . Ki

This measure corresponds to equation (2) for σ = ∞, the case of perfect competition. The scaling of PiYi /Ki by any constant will not affect our regressions, where we use the logarithm of MPK, and affects only the descriptive statistics where we focus on the dispersion, rather than the level, of MPK. Because we do not know what would be a suitable value of σ in our sample, we use the perfect competition benchmark. 2. Hsieh and Klenow Measure (HK): For α = 1/3, we calculate the index introduced by Hsieh and Klenow as HKi =

(4)

a (wL)i . 1 − a RKi

This measure directly reflects the relative capital distortion because it, under the assumptions of Hsieh and Klenow’s model, directly measures 1 + τKi as can be seen from equation (1). 5.3.4

Sample Selection Criteria

In our analysis, we use manufacturing firms and limit ourselves to countries with at least thirty-five firms having observations on nominal interest rates. Thus, the baseline sample comprises ten African countries with 4,039 firms, Germany with 221 firms, India with 2,286 firms, Ireland with 175 firms, South Korea with 215 firms, and Spain with 134 firms. We apply the following sample selection criteria to all firms in the baseline sample:

• We drop firms with missing information on key variables such as value added, capital stock, and labor.

• We drop government-owned firms. • We drop firms with negative age, which is calculated as the difference of

the corresponding year that the firm is surveyed and its date of establishment. Thus, if age is negative, we treat the date of establishment as faulty. • We drop firms with negative values of sales, capital stock, labor, total cost of raw materials and intermediate goods. • We drop firms whose replacement cost of capital stock is zero and whose replacement cost is bigger than the net book value of capital. • We drop firms below the 1 percent and above the 99 percent tails of replacement cost value of capital stock.

192

Sebnem Kalemli-Ozcan and Bent E. Sørensen

In the final sample, the total number of firms in African countries (Botswana, Burundi, Ghana, Kenya, Nigeria, Senegal, South Africa, Tanzania, Uganda, and Zambia) is 3,908. The final sample has 168 German firms, 2,129 Indian firms, 140 Irish firms, 178 South Korean firms, and 114 Spanish firms. 5.3.5

Country-Level Data

Our country-level broad institutional measures come from the ICRG Researcher Dataset and World Bank Doing Business databases. The first mentioned data set collects political information and financial and economic data, converting these into risk points for each individual risk component on the basis of a consistent pattern of evaluation. The political risk components are “government stability,” “socioeconomic conditions,” “investment profile,” “external conflict,” “internal conflict,” “corruption,” “military in politics,” “religious tensions,” “weak law and order,” “ethnic tensions,” “democratic accountability,” and “bureaucracy quality.” The main variables used from this data set are “corruption” and “investment profile.” The second data set provides quantitative measures of regulations regarding starting a business, dealing with construction permits, employing workers, registering property, getting credit, protecting investors, paying taxes, trading across borders, enforcing contracts, and closing a business. The main variable we use from this data set is “registering property (days).”

• Registering property: The number of days it takes to register property that an entrepreneur wants to purchase.

• Corruption: This is a measure that assesses actual or potential corrup-

tion in the form of excessive patronage, nepotism, job reservations, “favors for favors,” and secret party funding. Larger values of the index indicate higher risk of conducting business ineffectively. • Investment profile: This is an assessment of factors affecting the risk to investment that are not covered by other political, economic, and financial risk components. The risk rating assigned is the sum of three subcomponents, each with a maximum score of four points and a minimum score of zero points. A score of four points indicates very low risk and a score of zero indicates very high risk. The subcomponents are “contract viability/expropriation,” “profits,” “repatriation,” and “payment delays.” Larger values of the index indicate higher risk of expropriation, payment delays, and so forth. 5.4 5.4.1

Empirical Analysis The Extent of Capital Misallocation

In table 5.1, we display descriptive statistics for our main sample of countries (Burundi, Kenya, South Africa, Senegal, Botswana, Nigeria, Uganda,

Table 5.1

Descriptive statistics Obs.

Mean

Std. dev.

Min.

Max.

Median

Kurtosis

A. African countries, 2005–2006 15.3 6.1 0 8.7 6.2 –11.6 1.5 1.8 –3.7 –0.9 1.4 –7.3 0.2 1.5 –6.3

50 44.4 9.1 5.6 6.1

14 7.7 1.5 –0.9 0.3

6.6 6.3 2.8 3.9 3.4

Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

811 811 3,538 3,533 3,539

Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

52 52 284 284 284

20.5 9.8 0.1 –0.6 0.7

B. Ghana, 2006 7.3 7.3 1.6 1.5 1.4

1.5 –9.2 –3.7 –7.3 –6.3

33.3 22.6 5.9 5.6 5.4

21.1 10.4 –0.1 –0.6 0.7

3 3 3.5 7.1 6

Nominal interest rate Real interest rate log (K/L) log MPK

114 114 158 155

8.5 6.6 4 –0.7

C. Germany, 2005 2.9 2.9 1.1 1.1

4.1 2.2 0 –3.6

17 15.1 6.7 2.7

7.8 5.9 4.1 –0.7

4.2 4.2 4.3 4

Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

694 694 1,495 1,462 1,481

12 8.2 1.3 –1.2 –0.2

D. India, 2004 6.3 6.3 1.5 1.6 1.9

0 –3.8 –6.4 –8.9 –9.2

95 91.2 9.3 12.6 13.7

11.5 7.7 1.4 –1.4 –0.2

94.3 94.3 7.1 9.9 8.6

Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

96 96 132 119 131

4.7 2.3 4.4 –2.4 1.4

E. Ireland, 2005 1.6 2.7 1.6 0.3 1.1 0.2 1.2 –5 1.2 –1.6

9.5 7.1 7 2.4 5.3

4.2 1.8 4.4 –2.5 1.5

3.5 3.5 4.9 6.4 4.4

Nominal interest rate Real interest rate log (K/L) log MPK

106 106 152 144

F. South Korea, 2005 6.4 1.5 2 3.6 1.5 –0.8 4.3 1 1.6 –1 1 –3.8

12 9.2 6.5 1.9

6 3.2 4.3 –1.1

5.6 5.6 2.7 3.7

Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

63 63 90 88 88

G. Spain, 2005 1.5 1.5 1 0.8 0.9

12 8.6 6.8 0.9 3.7

3.7 0.3 4.2 –1.2 1.4

12.5 12.5 2.5 2.3 2.5

4.1 0.7 4.2 –1 1.5

2.5 –0.9 2.1 –2.5 –0.4

Notes: The nominal interest rate is the response to the question “What is the annual cost of loans (i.e., rate of interest)?” To calculate the real interest rate, we subtract the annual inflation rate (percent change in consumer prices) in the year of the survey. The marginal product of capital (MPK) is calculated as α(Y/K) where Y and K are value added and replacement cost of capital stock, respectively. The Hsieh-Klenow (HK) index is calculated as [α/(1 – α)][(wL)/(RK)] where wL and RK stand for total cost of labor and capital, respectively. The standard deviation is calculated for each country and then averaged. The Africa sample comprises Botswana, Burundi, Tanzania, Uganda, Kenya, South Africa, Ghana, Nigeria, Zambia, and Senegal. The firms being surveyed in Germany and South Korea are not asked about the total cost of labor (wL), thus values of the HK-index are not available for those firms; K/L is calculated using the total number of full-time workers at the end of the year of the survey. See the data section for detailed explanations of the variables.

194

Sebnem Kalemli-Ozcan and Bent E. Sørensen

Ghana, Tanzania, and Zambia). These statistics are calculated for each country and then averaged. We display statistics for Ghana separately and— for comparison to emerging and developing countries—Germany, India, Ireland, South Korea, and Spain. The table displays nominal and real interest rates, the logarithm of the capital-labor ratio, the logarithm of the MPK, and the logarithm of the Hsieh-Klenow HK-index. We choose to show the variables in logarithmic form, where the variables are close to normally distributed, because this is how they are used in the regression analysis. For the African countries, nominal interest rates have a mean of 15.3 with a standard deviation of 6.1, have minimum and maximum values of 0 and 50, respectively, and exhibit high kurtosis (compared with the value of 3 for the normal distribution). Real interest rates have a mean of 8.7 percent, a standard deviation of 6.2, a minimum of –11.6, and a maximum of 44.4. Inflation rates may differ widely between rural and urban areas, and in either event such negative, numerically large, real rates are suspicious, so we will focus on nominal interest rates—see Deaton and Heston (2010) for some issues in measuring price levels in developing countries. It is hard to know what is the optimal level of the interest rate in these countries, but large variation in interest rates indicates suboptimal allocation of capital. Ghana seems fairly similar to other African countries, with a large standard deviation of nominal interest rates at 7.3. Szabo (2011) points out that family loans in Ghana are quite common and often carry very small nominal interest rates, and combined with the very high interest rates in the informal market pointed out earlier, this helps explain the enormous spread in interest rates. Interest rates display similar large spreads within India (standard deviation of 6.3) while they are much less dispersed, with standard deviations at 2.9 and below in the developed countries Germany, Ireland, South Korea, and Spain. This indicates that the large spreads observed in Africa do not reflect actuarially fair risk premiums. Capital-labor ratios are approximately log-normally distributed with the log-ratio having a kurtosis of about 3 in Africa. If capital is efficiently allocated, all firms have the same MPK but, obviously, our MPK measures are estimated under highly simplifying conditions and therefore estimated MPKs will vary, maybe due to the simplifying conditions. In order to evaluate if the variation in the MPKs indicates bad allocation of capital, we compare to the spread in estimated MPKs in developed countries. We find low standard deviations of log-MPK of about 1 in developed countries versus 1.4 for the African sample (1.5 for Ghana) and 1.6 for India, indicating misallocation in Africa (and India). The HK-measure takes a theoretical value of unity under efficient allocation and this measure also displays significantly higher variation in Africa and India (due to lack of data, this index is not available for Germany and South Korea). Table 5.2 gives a breakdown of the number of firms into exporters, listed, and small firms while table 5.3 shows the statistics of table 5.1 broken down

Misallocation, Property Rights, and Access to Finance Table 5.2

195

Distribution of firm types N

(%)

All firms Exporters Listed Small

A. African countries 3,908 77 22 1,998

100 1.97 0.56 51.13

All firms Exporters Listed Small

B. Germany 168 4 0 53

100 2.38 0 31.55

All firms Exporters Listed Small

C. India 2,129 374 127 959

100 17.57 5.97 45.04

All firms Exporters Listed Small

D. Ireland 140 16 1 51

100 11.43 0.71 36.43

All firms Exporters Listed Small

E. South Korea 178 25 18 53

100 14.04 10.11 29.78

All firms Exporters Listed Small

F. Spain 114 7 0 30

100 6.14 0 26.32

Notes: The first column reports the number of firms. The second column reports the percent of firm types. Exporters have a ratio of exports to total sales above 50 percent. Listed firms are listed on a stock exchange. Small firms have five to nineteen employees. The Africa sample comprises Botswana, Burundi, Tanzania, Uganda, Kenya, South Africa, Ghana, Nigeria, Zambia, and Senegal.

by type of firm. Large firms have more capital per worker and pay lower interest rates and this holds even more for listed firms. Similarly, exporting and foreign-owned firms have more capital relative to labor, while foreignowned firms pay slightly lower interest rates. There is also less dispersion of interest rates within the group of listed firms, indicating less misallocation of capital within this group of firms. We next study these indicators in graphical form where more information can be shown compactly by country. Figure 5.1, panel (a) displays inflation

196

Sebnem Kalemli-Ozcan and Bent E. Sørensen

Table 5.3

Descriptive statistics by firm types

Small firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index Large firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index Listed firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index Nonlisted firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index Exporting firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index Nonexporting firms Nominal interest rate Real interest rate log (K/L) log MPK log HK-index

Obs.

Mean

Std. dev.

Min.

Max.

Median

Kurtosis

217 217 1,985 1,984 1,985

16.3 9.9 1 –0.8 0.4

6.9 7.1 1.7 1.3 1.4

0 –11.6 –3.7 –6.1 –6.2

50 44.4 9.1 3.5 4.7

15 9.6 0.9 –0.8 0.5

7.4 7.5 3 3.3 3.4

594 594 1,553 1,549 1,554

14.9 8.3 2.1 –0.9 0.1

5.8 5.9 1.8 1.5 1.5

1.5 –9.2 –3.6 –7.3 –6.3

38 30.7 7.9 5.6 6.1

13 7.7 2.3 –1 0

5.5 5 3.2 4.5 3.7

10 10 22 22 22

14.3 5.2 2.8 –0.9 0.1

4.2 3.4 1.2 1 0.9

7.5 0.9 0.1 –2.5 –1.6

21.2 10.5 4.8 1.4 2.5

14.3 4.6 3.1 –1.1 0

2.3 1.8 2.5 2.9 3.9

801 801 3,516 3,511 3,517

15.3 8.8 1.5 –0.9 0.3

6.1 6.3 1.8 1.4 1.5

0 –11.6 –3.7 –7.3 –6.3

50 44.4 9.1 5.6 6.1

14 7.7 1.5 –0.9 0.3

6.6 6.3 2.8 3.9 3.4

34 34 73 71 73

12.4 5.9 2.1 –1 0

4.1 4.2 1.6 1.5 1.6

5 –5.7 –1.6 –4.4 –3.4

30 19.3 7.2 4 5.3

12 6.4 2 –1 –0.2

11.8 5.7 3.5 4.3 4

777 777 3,465 3,462 3,466

15.4 8.8 1.5 –0.9 0.3

6.2 6.3 1.8 1.4 1.5

0 –11.6 –3.7 –7.3 –6.3

50 44.4 9.1 5.6 6.1

14 7.9 1.5 –0.9 0.3

6.5 6.3 2.7 3.9 3.4

Notes: The sample is for Africa only and comprises Botswana, Burundi, Tanzania, Uganda, Kenya, South Africa, Ghana, Nigeria, Zambia, and Senegal. See notes to the previous tables for detailed explanations of the variables.

and the mean and standard deviation of nominal interest rates for South Korea, Spain, Ireland, Germany, Burundi, Kenya, South Africa, Senegal, Botswana, India, Nigeria, Uganda, Ghana, Tanzania, and Zambia, in this order, where we have ordered the countries by the standard deviation of interest rates from low to high. Among the African countries, Burundi displays the lowest variation in interest rates, followed by Kenya and South

Distribution of nominal interest rate and inflation

Notes: Panel (a) displays the mean and standard deviation of nominal interest rates and inflation. The data used are: 2005 data for Botswana, Burundi, Tanzania, and Uganda; 2006 data for Kenya, South Africa, Ghana, Nigeria, Zambia, and Senegal; 2005 data for Germany, Ireland, South Korea and Spain; and 2004 data for India. Panel (b) displays box plots for the distribution of nominal interest rates where the nominal interest rate is the response to the question “What is the annual cost of loans?” (i.e., rate of interest). Inflation rate is the annual percent change in consumer prices. In both panels, countries are ordered according to the standard deviation of nominal interest rates.

Fig. 5.1

198

Sebnem Kalemli-Ozcan and Bent E. Sørensen

Africa, while Zambia has the highest spread, followed by Tanzania, Ghana, and Uganda. Developed countries have much lower variation in interest rates. Figure 5.1, panel (b) shows box plots for the distribution of interest rates (albeit with extreme outliers removed). The main “box” of data for each country shows the range of the 25–75 percentiles. Such plots will reveal if the standard deviations are mainly caused by outliers. Visually, if a low interest rate combined with a low 25–75 spread is considered healthy, as we think it should be, Kenya and South Africa (as well as the developed countries) have the best distribution, while the distributions of interest rates within Zambia, Nigeria, and Ghana are less good. Figure 5.2, panels (a) and (b) display the spread of our two misallocation measures, the MPK and HK indices, respectively. “Spread” is defined as the absolute distance to the country median. Burundi, Botswana, and Nigeria have large spreads in the MPK and, less strongly, in the HK index. The HKindex has very large spreads for Spain and Ireland, which indicates that a high spread of this measure may be driven by outliers and, therefore, may not be a good indicator of misallocation. Figure 5.3 displays self-reported obstacles to growth for the African countries. Typically, access to finance plays the leading role with over 60 percent of all firms mentioning access to finance as a major obstacle in Burundi, Ghana, Nigeria, and Uganda. In South Africa less than 15 percent of firms mention finance, while the number is about 40 percent in Botswana, Kenya, and Tanzania, 50 percent in Senegal, and 30 percent in Zambia. Weak infrastructure is typically mentioned by about 30–35 percent of the firms, although the number is much lower for Botswana, South Africa, and Zambia. Law and order is a problem for 10–20 percent of firms, although the number is higher in Burundi and Kenya. Finally, red tape is mentioned by about 15–20 percent of firms in most countries with a very low number in South Africa. In Kenya, 35 percent of firms point to red tape—Kenya stands out in these figures as having a significant amount of firms mentioning each of the main obstacles, while most other countries have finance dominating other obstacles. 5.4.2

Misallocation, Country-Level Institutions, and Investment Climate

We next turn to the broader policy question of whether good institutions are relevant for performance at the firm level. Our broad institutional variables capture protection of investor rights measured as corruption, the general investment climate, measured as the risk factors affecting the investment and ease of doing business, measured as the days it takes to register a property. These variables are quite correlated among themselves and we show their correlations with the MPK index in figure 5.4 and with the HK index in figure 5.5. In figure 5.4, panel (a), we see a positive relation between misallocation,

The distribution of misallocation measures

Notes: Panels (a) and (b) display box plots for the distribution of the MPK spread and the HK-index spread, respectively. The spread is the absolute value of the difference between the firm-level value of the corresponding variable and its country-level median. In panel (a), the misallocation measure MPK, is calculated as α = Y/K where Y is value added and K is replacement cost of capital. In panel (b), the misallocation measure, the HK-index, is calculated as [(α/(1 – α)][(wL)/wL] where wL and wL stand for total cost of labor and capital, respectively. German and South Korean firms do not provide information on WL, thus the HK-index is not calculated for those firms. In both panels, countries are ranked according to the standard deviation of annual nominal interest rates and outside values are excluded. An outside value is defined as a value that is smaller than the lower quartile minus 1.5 times the interquartile range or larger than the upper quartile plus 1.5 times the interquartile range.

Fig. 5.2

Major obstacle figures

Notes: Limited access to finance, weak infrastructure, weak law and order, and red tape represent the major obstacle groups that establishments face during their operation. The measures are constructed using answers to the question “Do you think that . . . presents any obstacle to the current operations of your establishment?” For example, in the case of limited access to finance, the question is as follows: “Do you think that limited access to finance presents any obstacle to the current operation of your establishment?” Answers to these questions can be no obstacle, minor obstacle, moderate obstacle, major obstacle, and very severe obstacle, which are coded as 1, 2, 3, 4, and 5, respectively. The way we construct our measure of limited access to finance is as follows: We take the number of establishments that answered the question “Do you think that limited access to finance presents any obstacle to the current operation of your establishment?” as “major obstacle and very severe obstacle” and divide by the total number of establishments that answered the question. Hence, our measures represent the percentage of establishments that consider limited access to finance a very important obstacle for their operations. In the case of weak infrastructure, the questions are “Do you think that electricity, telecommunication, transportation, or access to land presents a major/severe obstacle to the current operations of your establishment?” In the case of red tape, the questions we use are “Do you think that tax rates, tax administration, customs and trade regulations, labor regulations, or business licensing and permits present a major/severe obstacle to the current operations of your establishment?” In the case of weak law and order, the questions are “Do you think that the functioning of the courts, political instability, corruption, macroeconomic instability, crime, theft and disorder, or practices of competitors in the informal sector present major/severe obstacles to the current operations of your establishment?”

Fig. 5.3

Fig. 5.3

(cont.)

Fig. 5.3

(cont.)

The relationship between misallocation (MPK) and institutional variables

Notes: Panels (a), (b), and (c) display cross-country correlation plots of the (country) average log(MPK) against the country-level measures property registration (days), investment profile, and corruption, respectively. Log(MPK) is the logarithm of MPK, which is calculated as α(Y /K) where Y is value added and K is replacement value of capital. Property registration is the number of days it takes to register a property that an entrepreneur wants to purchase. Investment profile is an assessment of factors affecting the risk to investment that are not covered by other political, economic, or financial risk components. The risk rating assigned is the sum of three subcomponents: contract viability/ expropriation, profits repatriation, and Payment delays. As that index value increases, the risk of expropriation, payment delays, and so forth, increases. Corruption is a measure that assesses actual or potential corruption in the form of excessive patronage, nepotism, job reservations, “favors for favors,” and secret party funding. As that index value increases, the risk of conducting business ineffectively increases. See the data section for detailed explanations.

Fig. 5.4

The relationship between misallocation (HK-index) and institutional variables

Panels (a), (b), and (c) display cross-country correlation plots of the (country) average log(HK-index) against the country-level measures property registration (days), investment profile, and corruption, respectively. Log(HK-index) is the logarithm of HK-index, which is calculated as [α/(1 – α)] × [(wL)/RK ] where wL and RK stand for total cost of labor and capital, respectively. Property registration is the number of days it takes to register a property that an entrepreneur wants to purchase. Investment profile is an assessment of factors affecting the risk to investment that are not covered by other political, economic, or financial-risk components. The risk rating assigned is the sum of three subcomponents: “contract viability/expropriation,” “profits repatriation,” and “payment delays.” As that index gets bigger, the risk of expropriation, payment delays, and so forth, increases. Corruption is a measure that assesses actual or potential corruption in the form of excessive patronage, nepotism, job reservations, “favors for favors,” and secret party funding. As that index value increases, the risk of conducting business ineffectively increases. See the data section for detailed explanations.

Fig. 5.5

Misallocation, Property Rights, and Access to Finance

205

measured by average log(MPK), and registering property. This implies that the longer it takes to register a property, the higher is misallocation. The implication is that informal lending or retained earnings do not make up for the impediments to formal credit. We see a negative slope for the relation between the (country-mean) level of misallocation and the index for investment profile in figure 5.4, panel (b). Figure 5.4, panel (c), which uses an index for corruption on the X-axis, is very similar. This means that countries with a better investment climate (lower expropriation risk/corruption) have lower levels of capital misallocation on average, which is consistent with the patterns found by Johnson, McMillan, and Woodruff (2002): firms are not likely to reinvest profits when property rights (broadly defined) are weak. For example, according to our field study, we would not expect mango producers in northern Ghana to reinvest profits to increase production for shipping to Accra, because profits would be exhausted by bribes at road checkpoints. The picture is the same for the HK index, as shown in figure 5.5. We proceed with firm-level determinants of the misallocation. 5.4.3

Misallocation and Access to Finance: Firm-Level Evidence

In this section, we investigate the role of various constraints faced by firms in explaining misallocation. Table 5.4 gives descriptive statistics for obstacles averaged into four groups: limited access to finance, weak infrastructure, weak law and order, and red tape, as described earlier. Table 5.4 shows that for African countries and Ghana the most serious obstacle is limited access to finance, which has the highest mean, followed by weak infrastructure. Developed countries have lower means in general for all the obstacles. In developed countries, limited access to finance seems to be equivalent to weak infrastructure for developing countries in terms of importance of obstacles (India is left out of this table because the answers to the questions are scored on a different scale). In table 5.5, we use ordinary least squares (OLS)-regressions to examine determinants of misallocation using log-MPK as the dependent variable. The MPK is equalized across firms under ideal conditions, so in the absence of distortions all regressors should be insignificant and no firm-level obstacle should significantly predict MPK. We interpret positive significant values as determinants of capital market distortions relative to labor market distortions. This is because higher MPK of a firm as a result of a certain obstacle indicates that relatively little capital was allocated to that firm. We find in column (1) that limited access to finance and weak infrastructure are insignificantly correlated with distortions, while the MPK is negative and significantly correlated with weak law and order and red tape. The coefficient of –9.3 to weak law and order implies that an increase of one unit in the weak law and order index (moving from, say, “no obstacle” to “minor obstacle”) is associated with a 9.3 percent increase in distortion in the direction of having too much capital relative to labor. That is, the negative coefficient

Table 5.4

Descriptive statistics of obstacles to firm operations Obs.

Mean

Std. dev.

Min.

Limited access to finance Weak infrastructure Weak law and order Red tape

3,908 3,908 3,908 3,908

A. African countries 3 1.5 2.4 0.9 2.1 0.8 2 0.8

1 1 1 1

Limited access to finance Weak infrastructure Weak law and order Red tape

305 305 305 305

3.6 2.5 1.6 1.8

B. Ghana 1.4 0.7 0.5 0.7

Limited access to finance Weak infrastructure Weak law and order Red tape

168 168 168 168

C. Germany 2.2 1 1.4 0.5 1.7 0.6 2.1 0.7

Limited access to finance Weak infrastructure Weak law and order Red tape

140 140 140 140

1.9 1.7 1.6 1.8

Limited access to finance Weak infrastructure Weak law and order Red tape Limited access to finance Weak infrastructure Weak law and order Red tape

Max.

Median

Kurtosis

5 5 5 5

3 2.5 2 1.8

1.5 2.4 2.6 2.9

1 1 1 1

5 4.5 3.5 4

4 2.5 1.5 1.6

2.4 2.6 3.6 3.1

1 1 1 1

4 4 3.7 4

2 1.3 1.5 2

1.9 9.9 3.9 2.7

D. Ireland 1 0.8 0.6 0.8

1 1 1 1

4 4 4 4

1.5 1.5 1.5 1.7

2.3 2.9 5.8 3.3

177 176 176 177

E. South Korea 2 1.1 1.5 0.8 1.9 0.7 1.7 0.8

1 1 1 1

4 4 4 4

1.5 1 2 1.5

2.1 4.5 2.9 3.3

114 114 114 114

2.3 1.8 1.9 2

F. Spain 1 0.9 0.9 0.9

1 1 1 1

4 3.8 4 4

2 1.5 1.6 1.7

1.8 2.9 3 2.5

Notes: We use 2005 data for Botswana, Burundi, Tanzania, and Uganda; 2006 data for Kenya, South Africa, Ghana, Nigeria, Zambia, and Senegal; and 2005 data for Germany, Ireland, South Korea, and Spain. We average answers to questions about obstacles into four groups: limited access to finance, weak infrastructure, weak law and order, and red tape. The basic obstacle measure is the response to the question “Do you think that X presents any obstacle to the current operations of your establishment?” where X represents various questions whose answers are averaged into these four groups. Answers vary between 1 (no obstacle), 2 (minor obstacle), 3 (moderate obstacle), 4 (major obstacle), and 5 (very severe obstacle). Weak infrastructure is composed of the following Xs: electricity, telecommunications, transportation, and access to land. Red tape is composed of the following Xs: tax rates, tax administration, customs and trade regulations, labor regulations, and business licensing and permits. Weak law and order is composed of the following Xs: functioning of the courts; political instability; corruption; macroeconomic instability; crime, theft, and disorder; and practices of competitors in the informal sector. Limited access to finance is a stand alone question that represents a single X. Weak law and order and red tape are coded such that higher values correspond to less law and order and more red tape.

Misallocation, Property Rights, and Access to Finance Table 5.5 Dependent variable: Log MPK Limited access to finance Weak infrastructure Weak law and order Red tape

Firm-level determinants of misallocation (Africa)

(Africa)

(Africa)

(Africa)

(India)

(S. Korea)

(Europe)

4.5 (1.8) 6.1 (1.3) –9.3** (–2.5) –7.4* (–1.9)

3.2 (1.5) 5.4 (1.1) –8.6* (–2.2) –4.9 (–1.4)

2.6 (1.6) 6.1 (1.2) –8.2* (–2.2) –6.4* (–1.9) –0.2 (–0.7) 14.5 (0.5) –16.0*** (–3.4) –24.8** (–2.5) 3.9 (0.2) 4.4 (0.2) –3.3 (–0.2)

4.9** (2.4) 12.0** (2.4) –4.8 (–0.8) –15.1*** (–4.1) –0.4** (–2.7) 13.5 (0.5) –9.0 (–1.3) –23.6** (–2.8) –4.5 (–0.2) –13.3 (–0.8) 11.3 (0.5)

2.3 (0.5) 16.9** (2.3) –11.8 (–1.3) 11.8*** (3.2) –0.2 (–0.5) 89.1*** (5.8) –61.9*** (–6.9) –15.0 (–1.0) 45.6** (2.7) 3.8 (0.3) –23.6 (–0.8)

13.4 (1.2) 8.5 (0.9) –2.8 (–0.2) –13.6 (–0.4) –0.9 (–1.6) –9.5 (–0.3) –12.4 (–1.0) –50.4** (–2.9) –32.2 (–0.8) –23.9 (–0.6) 84.9** (2.8)

–5.3 (–0.8) –22.1 (–1.6) –0.5 (–0.0) 18.2* (1.9) –0.2 (–1.2) 21.3 (0.8) –26.0* (–1.9) 64.4*** (5.3) 42.9 (1.2) –69.8 (–1.5) –53.3 (–1.2)

Yes No 0.1 3,533

Yes Yes 0.1 3,533

Yes Yes 0.1 3,529

No Yes 0.0 3,529

No Yes 0.1 1,449

No Yes –0.1 141

No Yes 0.4 362

Age Export Small Fin. ∗ export Inf. ∗ export Law ∗ export Red ∗ export Country effects Industry effects Adj. R sq. N

207

Notes: See notes to the previous tables for variable explanations. The interaction variables are constructed by the multiplication of (Export – Export) with (X – X ), where X refers to an obstacle group and X refers to the mean of the corresponding variable over all firms. Export is a dummy, which indicates firms whose percentage of direct exports to total sales exceed 50. The Africa sample comprises Botswana, Burundi, Ghana, Kenya, Nigeria, Senegal, South Africa, Tanzania, Uganda, and Zambia. The Europe sample comprises Germany, Ireland, and Spain. Standard errors are robust and t-statistics are reported in parentheses. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

implies that the MPK is lower when law and order is lacking, indicating that it correlates more with labor market than with capital market distortions (see equation [2]). Column (2) adds industry dummies and column (3) adds age, size, and export status of the firm. Small firms have lower MPKs, meaning higher distortions. This implies that aggregate efficiency could be improved by moving capital to larger, more efficient firms. Limited access to finance now matters for exporting firms; such firms have lower returns and more distortions if

208

Sebnem Kalemli-Ozcan and Bent E. Sørensen

their access to finance is limited. Column (4) drops country dummies. The coefficient of limited access to finance becomes significant in column (4), which implies that limited access to finance in particular explains differences in MPK between countries. Limited access to finance is not significant for developed countries nor for India, even without the country effects. Weak infrastructure is significant when we omit the country dummies in column (4), suggesting that this variable in particular explains cross-country differences in MPK. This variable has strong explanatory power for MPK differences within India, but is insignificant for Korea and Europe. The coefficient to weak infrastructure is positive, indicating that firms invest less in physical capital when infrastructure is lacking. The coefficient to red tape also is negative in column (4), indicating that cumbersome bureaucracy is important in explaining differences in MPK across African countries as well as within countries (recall that a negative coefficient means lower MPK and so higher distortions). The coefficient to this variable is significantly positive for India and Europe, possibly due to high tax rates and a high degree of bureaucracy discouraging investment, although we do not further explore what underlies this. For Africa, the coefficient to red tape is negative, indicating labor market distortion, and it is possible that labor distortions stem from rules such as those imposing minimum numbers of local employees that we described for the case study of Ghana—such rules would likely be referred to as red tape in the surveys. As in column (3), when we include interaction terms, we see that the sign for exporting firms interacted with limited access to finance is negative in all columns (not significant for India and positive for Europe). The interpretation of this is that nonexporting firms are relatively more affected by imperfections in financial markets, leading to high MPKs consistent with such firms having too low a capital/labor ration relative to exporting firms— possibly, exporting firms are able to raise funds through long-term relations with foreign customers. The other interactions are not significant for the African sample on which we focus. Table 5.6 repeats the same exercise using the Hsieh-Klenow index. The HK-index takes a high value in the case of capital distortions (too low capital) and a negative value in the case of labor distortions (too little labor compensation). In contrast to the MPK-measure of table 5.5, it relies on labor costs being correctly measured. A natural first hypothesis is that imperfect financial markets lead to capital distortions, and the estimated coefficient to limited access to finance in column (1) confirms this with a very high level of significance. The coefficient indicates that a firm that moves from, say, “no obstacle” to “minor” obstacle, will have 11 percent higher capital distortion (while moving from “no obstacle” to “very severe obstacle” increases capital distortions by 44.4 percent). Red tape has a negative coefficient, implying that bureaucracy leads to higher labor distortion relative to capital distortions (the opposite holds for India, where capital distortions seems to be

Misallocation, Property Rights, and Access to Finance Table 5.6 Dependent variable: Log HK-index Limited access to finance Weak infrastructure Weak law and order Red tape

Firm-level determinants of misallocation (Africa)

(Africa)

(Africa)

(Africa)

(India)

(Europe)

11.0*** (3.5) –2.6 (–0.5) –5.8 (–1.6) –11.5* (–2.2)

8.6*** (3.4) –2.7 (–0.6) –5.4 (–1.4) –8.5 (–1.8)

7.1*** (4.2) –2.4 (–0.5) –6.2 (–1.8) –8.1 (–1.8) –0.0 (–0.0) 11.5 (0.4) –10.3 (–1.3) –35.1* (–2.2) –3.7 (–0.2) –14.7 (–0.6) 28.0 (1.0)

10.8*** (4.6) 4.2 (0.9) –4.1 (–0.7) –24.9*** (–6.0) –0.3 (–1.4) 12.8 (0.6) –1.4 (–0.1) –29.1** (–3.0) –13.8 (–0.7) –41.0** (–2.7) 42.3** (2.4)

2.9 (0.5) 3.9 (0.4) –15.2 (–1.5) 16.0*** (3.4) –0.1 (–0.3) 96.4*** (3.3) –60.0*** (–7.2) –6.6 (–0.4) 25.5 (1.0) 32.9 (1.2) –49.2 (–1.2)

–2.6 (–0.3) –9.7 (–0.6) –9.4 (–1.2) 11.3 (1.2) 0.0 (0.1) 6.4 (0.3) 3.7 (0.4) –23.4 (–1.2) 20.2 (0.3) 7.3 (0.1) –14.7 (–0.3)

Yes No 0.1 3,539

Yes Yes 0.2 3,539

Yes Yes 0.2 3,535

No Yes 0.1 3,535

No Yes 0.1 1,467

Yes Yes –0.0 219

Age Export Small Fin. ∗ export Inf. ∗ export Law ∗ export Red ∗ export Country effects Industry effects Adj. R sq. N

209

Notes: See notes to the previous tables for variable explanations. The interaction variables are constructed by the multiplication of (Export – Export) with (X – X ) where X refers to an obstacle group and X refers to the mean of the corresponding variable over all firms. Export is a dummy that indicates firms with percentage of direct exports to total sales greater than 50. The Africa sample comprises Botswana, Burundi, Ghana, Kenya, Nigeria, Senegal, South Africa, Tanzania, Uganda, and Zambia. The Europe sample comprises Ireland and Spain. Because German and South Korean firms are not asked to provide information on total cost of labor, the HK-index measure cannot be calculated. Standard errors are robust and tstatistics are reported in parentheses. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

more important). In column (4), we see that for exporters, a high value of limited access to finance implies less relative capital distortion compared to nonexporters. Red tape in particular explains differences between countries as can be seen from the larger coefficient when country dummies are dropped. The interaction effects for exporters reveal negative coefficients, meaning for exporters limited access to finance leads to more labor distor-

210

Sebnem Kalemli-Ozcan and Bent E. Sørensen

tions, though column (4) suggests this pattern might be driven by country differences. Overall, the patterns of table 5.6 are quite similar to those of table 5.5, which indicates that the obstacles we consider distorts capital/labor ratios, rather than output—if output distortions dominated, the regressors would be significant in table 5.5 and insignificant in table 5.6 and not similar as in our results. 5.5

Conclusion

Using establishment-level survey data from the World Bank, we quantify the extent of misallocation within the manufacturing sector for ten African countries. To benchmark our results, we use similar data for India, another developing country, and developed countries such as Germany. Our main measures of misallocation are the MPK and an index suggested by Hsieh and Klenow (2009). Both measures reveal a great deal of capital misallocation in Africa and India and little misallocation in developed countries. Within Africa, capital markets appear to function relatively well in South Africa and, a little less so, in Kenya and Burundi, while Nigeria and Ghana display high levels of misallocation. The firm-level regressions demonstrate that an important determinant of within-country misallocation is limited access to finance. The MPK is 45 percent higher in firms where access to finance is a serious problem compared to firms where access to finance is not a problem. Comparing mean levels of misallocation across countries, we find that countries with higher risk of expropriation that is, lower degrees of property-right protection, display more misallocation. The within-country firm-level results and acrosscountry results both point to the importance of strong property rights and a well-functioning financial system for the efficient allocation of capital.

References Alfaro, L., A. Charlton, and F. Kanczuk. 2008. “Firm-Size Distribution and CrossCountry Income Differences.” NBER Working Paper no. 14060, Cambridge, MA. Banerjee, A. V. 2002. “Contracting Constraints, Credit Markets and Economic Development.” Department of Economics Working Paper no. 02–17 (section 4), MIT. Banerjee, A. V., and E. Duflo. 2005. “Growth Theory through the Lens of Development Economics.” In Handbook of Economic Growth, vol. 1, edited by P. Aghion and S. Durlauf, 473–552. Amsterdam: Elsevier. Banerjee, A. V., and B. Moll. 2010. “Why Does Misallocation Persist?” American Economic Journal: Macroeconomics 2:189–206.

Misallocation, Property Rights, and Access to Finance

211

Bartelsman, E., J. Haltiwanger, and S. Scarpetta. 2009. “Cross-Country Differences in Productivity: The Role of Allocation and Selection.” NBER Working Paper no. 15490, Cambridge, MA. Beck, T., A. Demirgüç-Kunt, L. Laeven, and V. Maksimovic. 2006. “The Determinants of Financing Obstacles.” Journal of International Money and Finance 25:932–52. Beck, T., A. Demirgüç-Kunt, and V. Maksimovic. 2005. “Financial and Legal Constraints to Growth: Does Size Matter?” Journal of Finance 60:137–77. Deaton, A., and A. Heston. 2010. “Understanding PPPs and PPP-based National Accounts.” American Economic Journal: Macroeconomics 2:1–35. Hsieh, C., and P. Klenow. 2009. “Misallocation and Manufacturing TFP in China and India.” Quarterly Journal of Economics 124:1403–48. Johnson, S., J. McMillan, and C. Woodruff. 2002. “Property Rights and Finance.” American Economic Review 92:1335–56. Restuccia, D., and R. Rogerson. 2008. “Policy Distortions and Aggregate Productivity with Heterogeneous Plants.” Review of Economic Dynamics 11:707–20. Szabo, Andrea. 2011. “Family Financing and Productivity among Small Manufacturing Firms in Ghana.” Working Paper, University of Houston.

6

New Cellular Networks in Malawi Correlates of Service Rollout and Network Performance Dimitris Batzilis, Taryn Dinkelman, Emily Oster, Rebecca Thornton, and Deric Zanera

6.1

Introduction

Cellular phone technologies have become increasingly important throughout the developing world and in Africa in particular. Figure 6.1 shows growth in the population serviced by at least one cellular phone network in Africa over the last decade. The number of subscribers has more than tripled during this period and reached as high as 280 million at the end of 2007 (Buys et al. 2009). In contrast, growth in fixed-line infrastructure has been much slower than growth in cellular phone coverage, with approximately 90 percent of all African telephone subscribers consisting of cellular phone users (Paul Budde Communication Pty Ltd. 2009). There are challenges to the expansion of fixed-line infrastructure, including widespread material theft and the need to cover large low-population areas. As a result, cellular network technology may be the best way to bring telecommunication services to most of Africa. Popular media, as well as a growing body of research within economics, Dimitris Batzilis earned his PhD in economics from the University of Chicago. Taryn Dinkelman is assistant professor of economics at Dartmouth College and a faculty research fellow of the National Bureau of Economic Research. Emily Oster is associate professor of economics at Brown University and a faculty research fellow of the National Bureau of Economic Research. Rebecca Thornton is associate professor of economics at the University of Illinois at UrbanaChampaign. Deric Zanera is chief statistician at the National Statistical Office of Malawi. Research for this chapter was funded by the NBER Africa Project on African Successes. We are grateful for research assistance from Eirin Forsund, Cecile Gaubert, Soren Larson, Ajay Shenoy, and Ryan Wang. We also acknowledge the technical insight on cellular phone networks provided by Mung Chiang and Tian Lin in the Electrical Engineering Department at Princeton University. In addition, we thank the National Statistics Office in Malawi for assistance with data as well as the helpful staff at Zain and TNM, Malawi. For acknowledgments, sources of research support, and disclosure of the author’s or authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13366.ack.

215

Cellular phone coverage in Africa

Source: Buys et al. (2009).

Fig. 6.1

New Cellular Networks in Malawi

217

suggest that cellular technologies may be enormously important for improving productive efficiency, for providing a cheap method of transferring information and money across space and, ultimately, for promoting growth.1 Despite this potential importance, we know relatively little about how this technology spreads and who receives access, especially in poor countries with limited infrastructure and limited ability to provide even very basic public goods. We also know little about how quality of cellular services varies within a country: while the presence of cell towers is necessary for cellular technology to have an impact on societies, reliability of these networks is also important. In most developed-country settings we would expect that the timing of access to a new network provided by the private sector would correlate strongly with demand and cost factors. Further, we would expect network quality to be correlated with these factors through network upgrading and strengthening. However, in a developing-country setting it seems possible that other factors, such as political or ethnic influence or the existence of complementary infrastructure, could also drive placement and affect the quality of the service provided to an area.2 How and where infrastructure is provided has potential welfare consequences. In this chapter we take a first step toward addressing these issues using data from Malawi. We focus on estimating the role of demand-side and cost-side factors in driving cell phone access and the performance of these networks across rural areas. Understanding the role of these factors can put an upper bound on the importance of factors like political influence in driving access. Our analysis takes advantage of new data on cell phone access and performance that we collected in Malawi. Our first outcome is cell phone network access. There are two cell phone providers in Malawi. We obtained detailed data from each of the two of them on their exact tower locations and date of construction; our data cover every tower in the country. Using this information, we use geographic information system (GIS) software to construct coverage maps for the country over time. We focus on cellular phone access within a relatively small area, called an enumeration area (EA), of which there are about 9,200 in Malawi. Our data allow us, for each EA, to define coverage by year and to determine the first year in which the area had significant coverage. In addition, for the last two years of our data (in 2008 and 2009) we have some measures of network performance (dropped calls), which allow us to observe one aspect of the quality of service access 1. See, for example, Aker (2008), Aker and Mbiti (2010), Arnquist (2009), Balakrishnan (2008), Hausman (2010), Jack and Suri (2009), Jensen (2007), Klonner and Nolen (2008), Krudy (2009), McGreal (2009), and Ngowi (2005). 2. There is an established literature on the effects of corruption on public good provision and on economic growth more generally. See, for example, Alesina, Baqir, and Easterly (1999), Tanzi and Davoodi (1998), Mauro (1995), Banerjee and Somanathan (2007), Besley et al. (2004), Khwaja (2009), Easterly and Levine (1997), and Kimenyi (2006).

218

Batzilis, Dinkelman, Oster, Thornton, and Zanera

provided to an area. In both cases, we link these data with Malawian Census data from 1998 (demographic proxies for demand) and to information on the geography of the country (variables that proxy for cost). The first cellular towers in Malawi were built in the mid-1990s, and coverage was initially focused around the major cities (i.e., Lilongwe and Blantyre) and other cities covering tourist areas near Lake Malawi (i.e., Salima and Mangochi). However, increases in coverage were rapid: by 2004, 57 percent of the land area of the country had access, and this figure increased to 86 percent by 2008. Conditional on having cell coverage, network quality varies. At the cell-tower level, the fraction of dropped calls reached up to 30 percent in some months of 2008 and 2009, although median monthly dropped-call rates are below 3 percent in most months. We perform three analyses to describe the placement and performance of cellular phone infrastructure in Malawi. In our simplest analysis, we regress network coverage (either a binary measure indicating whether more than 50 percent of the area is covered by a cell tower or a continuous measure of coverage) on the independent variables of interest for three years (2000, 2004, and 2008). This gives us a sense of which variables are important in determining access and how their importance changes over time. Second, we generate Kaplan-Meier survival curves and estimate Cox proportional hazard models to describe each EA’s hazard out of the uncovered state. This allows us to use all of the information on coverage timing together. Finally, we estimate ordinary least squares (OLS) regressions on the percent of dropped calls for towers covering each EA for the month of August 2008 to see whether the performance of the network is related to some of the same demand-side and cost-side variables used in the rollout analysis. We begin with the role of demand. Our analysis suggests that area-level proxies for demand measured in 1998 are strongly correlated with the timing of access between 2000 and 2008. A number of proxies for area income (which we do not observe directly in the census) are correlated with the timing of access. Areas with higher population density in 1998 receive coverage earlier. Higher levels of employment in agriculture correlate with delayed access, and higher levels of education in 1998 correlate with earlier access in the subsequent ten years. The results are similar if we use a binary variable measuring coverage or a continuous measure of the number of towers covering these areas. These correlates correspond with demand-side factors that engineers and other cellular phone industry representatives report as important for defining cell phone market potential. Following this, we move to estimate the correlation between cost-side geographic factors and coverage, in particular the altitude and slope of the location. Both of these variables affect the size of the potential market served by a particular tower (since line of sight may be more or less obstructed, depending on terrain) and affect the cost of building and maintaining towers. We find some evidence that these factors matter. Places with higher

New Cellular Networks in Malawi

219

slope are consistently less likely to have access; results on altitude are more mixed. Not surprisingly, isolation matters: areas that are far from a road are significantly less likely to get access to the network throughout the period. This last correlate may also capture aspects of consumer demand. We find similar results when we estimate hazard models: areas with higher population density receive coverage earlier, as do areas with more education and less agriculture. Areas further from roads and in higher altitude areas have a tendency to get coverage later. The evidence on geographic correlates is more mixed in the hazard models. Turning to the results on network quality, we have two related measures of dropped calls for each cellular network provider in Malawi. Since we only have network traffic data for the later period, we focus on describing how the performance of each network (measured by percent of dropped calls per month) varies in the cross-sectional data. Our measure provides one way of measuring network congestion. A priori, the relationship between market demand and cell network performance could be negative or positive: cell networks may be more congested in areas with more users per tower (i.e., higher market demand), but companies may respond to such congestion by differentially strengthening their infrastructure in these high-demand areas. In Malawi, the latter relationship appears to be the case for both companies. Using the traffic data for each firm separately, we find demand-side factors but not cost-side factors are important for predicting the percent of dropped calls. The EAs with higher population density and a higher fraction of educated adults have fewer dropped calls. We also find that areas with more extensive prior network coverage have a much lower percent of dropped calls in the month. Not surprisingly, the density of the network matters for network performance in Malawi. Taking the two analyses together, we find that within a traditional authority (TA), a larger unit of which there are about 300 in the country, demand and cost factors can explain between 40 percent and 60 percent of the cross-EA variation in cell tower coverage. For the measures of congestion that we use, our control variables account for 85 percent of the variation in dropped-call rates across EAs within the same traditional authority. Although this obviously does not rule out roles for other factors like political influence or complementary infrastructure in determining access to this particular service, it does suggest that these other factors do not play a predominant role in allocation in Malawi. 6.2 6.2.1

Background Country Background

Malawi is a small, landlocked country in southern Africa. It is extremely poor, with a median daily wage for a casual worker of around US$0.40, and

220

Batzilis, Dinkelman, Oster, Thornton, and Zanera

gross domestic product (GDP) per capita (purchasing power parity [PPP] adjusted) of around US$800 in 2008 (CIA 2009). In 1998, almost two-thirds of the population lived below the national poverty line and one-third of the population lived on less than US$0.25 per day (Benson et al. 2002). The large majority of Malawians work in agriculture (82 percent in 1998) and the country has a high population density and a youthful population (Benson et al. 2002). Education levels are low, as are school enrollment rates (although these have been rising in recent years), and the infant mortality rate is 83 per 1,000 (CIA 2009). Most types of public goods, infrastructure, and service provision are lacking in most areas. Roads are in poor condition: only 45 percent of national roads are paved. Although Malawi has an 800 km railway line, this line does not connect with lines in neighboring Zambia or Tanzania (as the line gauge differs). The only connection the country has to a port is through Mozambique’s railroad, part of which has been closed since 1983 due to the civil war in that country.3 Less than 60 percent of the population obtains water from a protected source, while only 4.5 percent of households use electricity for lighting. Despite high poverty rates and limited infrastructure, Malawi compares favorably with many other African countries. In the mid-1990s, the country transitioned from a dictatorship to a multiparty democracy, which has survived since then. Malawi has experienced no major civil conflict over this period, and elections have been held on schedule with only limited accusations of irregularities. Relative to other areas in Africa, tribal tensions have been limited, although postdemocracy tensions have been growing.4 6.2.2

Cellular Telephones in Malawi

Overview of Providers The first cellular operator, TNM (Telekom Networks Malawi Limited), was initially majority-owned by the government of Malawi. This operator was licensed by the government agency MACRA (Malawi Communications Regulatory Authority), a government-appointed regulatory board, in 1995.5 In the early years, growth in the network was concentrated in the four main urban areas and in tourist destinations; network coverage in rural areas was slow and coverage rates were low. The histogram in figure 6.2 illustrates some of this rollout as it occurred in rural parts of the country. It shows the fraction of census enumeration areas (EAs) that gain new access to a 3. CIA World Factbook (2009), Railways Africa: http://www.railwaysafrica.com/category /africa-update/sadc/malawi. 4. For a detailed analysis and history of tribalism and ethnicity in Malawi, see Vail (1989). 5. The Communications Sector Policy Statement establishes MACRA, the Malawi Communications Regulatory Authority, as the body responsible for regulating telecommunications, posts, broadcasting, and the radio frequency spectrum.

New Cellular Networks in Malawi

Fig. 6.2

221

Distribution of first year of cellular coverage at local (EA) level

Note: Range of years is from 1995 (year 1) to 2009 (year 15).

cellular network in each year from 1995 to 2009 (we define “access” in more detail below). The figure clearly shows the initial slow coverage of EAs in the early years; many more rural EAs are covered by a cellular network from 2003 onward. The slow growth and poor coverage prompted reform of this sector in 1998, at which point a second private operator, Celtel, was awarded a government license. Celtel was purchased in 2008 by Zain, which is headquartered in Kuwait and has a strong regional presence in over twenty other African and Middle Eastern countries. By late 1999 both operators were active in establishing their networks across the country, and by 2007, the number of cellular phone subscribers (prepaid and postpaid) had risen to over 1 million people, or 25 percent of the population (ITU 2008). Many more than these subscribers have access to this telecommunications service, as most consumers use a cell phone on a prepaid basis. Currently, over 70 percent of the country has access to the cellular phone network and Zain claims to serve 70 percent of the current Malawian market. Licensing Process As in other countries, companies must have a license to operate a cellular network in Malawi and these licenses are typically awarded after a tendering process. Both TNM and Zain were granted licenses to provide cellular services by MACRA. License fees are required for participation in this market, both at the initiation of the license agreement and on an annual basis after that (annual fees vary between 5 percent and 7 percent of company revenue). Additional fees apply for new services (e.g., Internet service) that each com-

222

Batzilis, Dinkelman, Oster, Thornton, and Zanera

pany chooses to provide. Since 1998, there have been two subsequent tenders announced for the awarding of a third license; however, no third operator has entered the market yet and so we confine our discussion to the combined activities of TNM and Zain in the rest of this chapter.6 One important aspect of the license agreements is that each company was required to build cellular phone towers in certain target areas before specific deadlines. Malawi’s 1998 Communications Sector Policy Statement commits MACRA to “ensure extension of modern telecommunication services to rural areas” and do this “according to a defined program covering rural areas.” In discussions with engineers and managers of both companies in Malawi, many of the target areas in rural parts of the country would not have been viewed as commercially viable sites, at least initially. During the period in which we analyze the rollout, these two companies were therefore expanding the network partly into areas that appeared profitable and partly into areas that were important to connect to meet license obligations. We observe both types of areas in our sample. Cellular Phone Costs and Access Although cellular phone infrastructure has expanded rapidly to cover almost the entire country, ability to access the network in Malawi is still somewhat limited to wealthier individuals. Malawi’s Integrated Household Survey of 2004 indicates that 17 percent of urban households and only 1 percent of rural households owned a cellular phone (although the rates of access to a cell phone are likely to have increased substantially since then).7 The initial cost of buying a handset still represents an important barrier to using the network: in 2009, the cost of the cheapest handset offered by Zain was MK 2,500 = US$18, or about fifty days of day-laborer work. In addition to the cost of a handset, tariffs are high relative to other developing countries, although there is some indication that they have fallen over time. Service prices for the two providers are difficult to keep track of and standardize over time because price-setting responsibility is opaque, tariff changes sometimes occur through temporary promotions and changes are not required to be reported to MACRA (the regulatory authority has no power to set prices). At various points during the period we study, there have been changes in both pre- and postpaid rates, as well as changes in how calls are billed (per minute versus per second), and changes in prices for within and outside network calling. Given this difficulty in standardizing the appropriate “unit of telecommunications” to use to compare prices over time (a 6. How and to whom licenses are awarded may in itself be of interest related to public good and infrastructure provision in Africa. We do not address this specifically in this chapter. 7. In other surveys that have recently been conducted in Malawi in subsequent years, 72.5 percent of urban men had their own cell phone with 75.5 percent of those without a cell phone having access to one; 23.2 percent of rural men had their own cell phone (Godlonton and Thornton 2010).

New Cellular Networks in Malawi

223

difficulty present in all research related to cell phone services, not just in the context of Malawi), no single statistic will capture the true cost of using the cellular phone network. However, to get an overview of how prices have changed on each network over time, we created an average consumption basket (making assumptions about average call durations, call destinations [in or out of network], and average call timing [peak/off-peak] using information from cellular phone traffic data from the later 2008–2010 period) and calculated the price per second of this basket at the relevant tariffs between 1999 and 2010. On average, the price per second of talk time across both companies was about 1.2 US cents in the late 1990s; by 2009, this average price had fallen by 50 percent, to 0.6 US cents per second. Despite this substantial reduction in the price of airtime, cell phone rates in Malawi are still about five times higher than the per-second cost of using a cellular phone in the United States (OECD 2009). As a result of these generally high tariffs in Malawi, individuals often communicate using cheaper options such as text messaging (10 US cents per message), or “flashing”/“pinging” (ringing and hanging up to signal a wealthier party to call back and pay for the call). It is also worthwhile to note that while it costs money to place a call or to send a text message, receiving a call or a text message is free. Thus, even though prices may be high enough to prohibit many people from initiating communication, having access to network coverage may be beneficial, if only to receive calls. Access to a cell tower is clearly necessary to use the new technology, but the network also needs to provide a reliable service in order to be useful for individuals. We know remarkably little about how well cell networks function in Africa. Given the dismal quality of many other types of infrastructure across the continent (roads and railways in particular), it is important to know whether cell infrastructure performs any better and to understand more about the correlates of better performance. In Malawi, network performance varies substantially across the country. Using the monthly data we have access to, we find individual towers report a dropped-call rate of up to 30 percent in some months. Undoubtedly, poor network performance based on this particular measure limits some of the potential effects of cell phones on society and the economy.8 Last, individuals must have access to some means of charging their cellular phone handset. Given the low prevalence of electricity in Malawian homes (according to the NSO [2004]), only 6 percent of households (32 percent urban and 2 percent rural) had access to electricity, individuals in more remote areas must resort to charging cellular phones using car batteries, or must pay for additional charging services in order to use their phones. 8. It is difficult to find accurate statistics on dropped-call rates in the United States for comparison. A recent article in the popular press (Manjoo 2010) reports statistics collected directly from several cell providers that indicate monthly call-drop rates of below 2 percent.

224

Batzilis, Dinkelman, Oster, Thornton, and Zanera

Our analysis in this chapter primarily focuses on the availability of the cellular phone network from cellular towers, rather than use. For the analysis, we assume that an area has coverage if it is reached by cellular phone towers, ignoring the fact that some people may have differing ability to take advantage of the network. We augment the coverage analysis with an examination of how well each company provides a reliable service to an area, where we define one particular measure of service reliability below. 6.3

Data

This chapter uses a number of new data sets, which we discuss in turn below. We have collated four types of data for the description of infrastructure rollout and network performance: (a) administrative data on the placement and timing of new cellular phone towers from both companies, (b) engineering data on cell phone traffic at the tower level for several months in 2008 and 2009, (c) geographic data describing Malawi’s physical terrain, and (d) demographic data from the census. We link these data sets at the enumeration area (EA) level and at times will also refer to the larger traditional authority (TA) in which EAs are nested. 6.3.1

Data on Cellular Phone Network Coverage

We begin with our data on cellular phone network coverage. As mentioned above, there are currently two cellular networks in Malawi: TNM and Zain. To generate information on overall coverage over time, we collected data from both providers. These data include the precise GIS location and construction date for each cellular tower for the period from 1995 to 2009. Using these data on latitude and longitude of the cellular towers, as well as data on elevation, we determined what areas of the country became part of the cellular phone network and when. We used the Viewshed tool of the ArcGIS software, which identifies the points in a map that can be seen from a set of observation points. For a full description of the tool, the reader can consult the help website of ESRI.9 With the Viewshed analysis, we assigned to each point in the map a value that is equal to the number of cellular phone towers (of either network) that are (a) visible from that location and (b) within a distance of 30 km of the tower. The idea is straightforward: there can be no coverage if the tower signal is blocked by physical obstructions, and hence the tower is not “visible,” or if the source of the signal is too far away. Discussions with engineers in Malawi suggested that 30 km is an appropriate range for the towers, although this distance may not be exact in all cases. Each EA consists of many points on the map of Malawi. The cellular 9. http:// webhelp.esri .com/ arcgisdesktop/ 9.2/ index .cfm?TopicName=How%20Viewshed%20works.

New Cellular Networks in Malawi

225

coverage value assigned to an EA is the expected (average) number of towers that cover a randomly chosen point that lies within the area of the EA. In mathematical terms, if x and y are respectively the longitude and latitude of a point in the map, fj (x,y) equals the number of cellular sites that cover the point (x,y) in year j, and Ai is the area of the EA, then the coverage value for the EA can be expressed as: Coverageij = ∫∫ f j (x, y) dx dy . Ai

It is important to note that having a coverage value of, for example, 0.50 does not necessarily imply that 50 percent of the area is covered. For example, if 25 percent of an EA’s area is covered by two towers simultaneously, then the value of the coverage variable is 0.5 as well. In 2008, the average coverage measure ranges from 0 to 47. Perhaps the best way to think of this measure is as an indicator of cellular network intensity in the EA, where EAs with the highest coverage value of 47 have access to the densest network. We use this coverage measure in several ways. First, we run OLS regressions on a binary indicator of coverage in early 2000, middle 2004, and late 2008 periods of our data. For these regressions, we define an entire EA as “covered” in the years when the above coverage value was larger than 0.5. Second, we run OLS regressions on the continuous coverage variable in each period. Finally, when we analyze hazard models, we experiment with various definitions of “covered”: whether the area is covered at all (coverage > 0), whether the coverage is over 0.5, and whether the coverage is over 1.0. Panel A of table 6.1 provides basic information about the fraction of all, urban and rural, EAs that are covered by at least one network in each year, using the binary definition of covered as coverage greater than 0.5.10 As early as 1999, most of the urban EAs were covered by the networks, but only 20 percent of rural areas had access. Coverage in rural areas started to ramp up from 2003 onward. Because of this quick initial saturation of network coverage in the main urban areas and the greater variation in coverage in rural areas, the analysis in this chapter excludes the urban areas and focuses on rural Malawi, where the majority of the population resides. In figure 6.2, as described above, we show the fraction of rural EAs with new cellular network coverage for each year. Maps in figures 6.3 to 6.5 provide a visual representation of the cellular coverage data by year in 1997 (when only TNM was active), 2004, and 2008. 6.3.2

Cell Phone Traffic Data and Network Performance Measures

We collected traffic data from each cell company for several months in 2008 and 2009. We limit our analysis here to the month of August 2008, tak10. Although there are about 9,200 EAs in all of Malawi, we have data for only 8,924, so our analysis focuses on those areas.

226 Table 6.1

Batzilis, Dinkelman, Oster, Thornton, and Zanera Summary statistics for enumeration areas (EAs) Fraction of EA covered

  1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

All EAs

Rural EAs

Urban EAs

A. Cellular phone coverage over time 0.27 0.20 0.27 0.20 0.30 0.23 0.31 0.24 0.45 0.39 0.57 0.53 0.66 0.63 0.74 0.72 0.80 0.78 0.86 0.85

 

 

Min.

Max.

0.002

0.050

0.003

0.023

0.97 0.97 0.97 0.97 0.99 0.99 1.00 1.00 1.00 1.00

N

8,924

8,118

806

 

Mean

S. d.

Num. obs.

B. Summary statistics for traffic congestion data: Rural EAs with cell coverage, by network Fraction dropped calls: Company 1, Aug 2008 0.020 0.010 5,971 Fraction dropped calls: Company 2, Aug 2008 0.009 0.003 8,046 C. Summary statistics for geographic data: Rural EAs only Altitude (meters above sea level) 848.52 342.23 8,118 Slope (% rise) 3.23 3.14 8,118 Distance to road (km) 1.45 2.17 8,118

34 0 0

D. Summary statistics of demographic variables from 1998 census: Rural EAs only Age 22.16 1.59 8,118 16 Education (yrs.) 2.97 1.22 8,118 0 Percent adults married 0.55 0.07 8,118 0 Percent adults in agriculture 0.89 0.17 8,118 0 Population 1,054.01 372.46 8,118 3 Population density (pop/km2) 313.84 770.43 8,118 0.03

1,988 24 62

40 8 1 1 3,661 21,142

Notes: This table presents summary statistics for the primary variables used in the rollout analysis. In all panels, the unit of observation is the enumeration area (EA). Summary measures are calculated over EAs with positive population density. Panels B, C, and D summary statistics are restricted to rural area EAs. Cellular coverage is a combined TNM and Zain coverage measure, constructed as described in the text.

ing this as a representative month during the period. For each cell tower, we calculate a measure of average percent of calls dropped during the month. The measures differ slightly across the two network providers since the level of aggregation differs. One provider’s data was available at the daily level, while the other’s was available at the monthly level. Because of this difference, we do not aggregate the data across providers, and rather present

New Cellular Networks in Malawi

227

statistics separately; because of the potential sensitivity of this information, we anonymize each provider. In this case, Company 1’s drop call rate is computed using monthly data on the number of dropped calls and the number of total attempted calls, while for Company 2, we use the reported fraction of dropped calls by day and average these up to the monthly level. Each tower t is placed in a unique EA. However, because there is a 30 km radius around each cell tower that receives signal from this tower, a typical tower will provide service to more than one EA. We address this by assigning the dropped-call information from tower t to each EA that falls in its radius. For EAs that are covered by multiple towers, we take the average of the dropped-call rate across all towers from which it receives coverage. The EAs that do not receive any network coverage from one or the other company do not have any traffic data assigned to them. In table 6.1, panel B, we see the mean dropped-call rate (fraction) for each cell company for the set of rural EAs with cell coverage. Company 1’s dropped-call rate is higher at 2.0 percent per month, with Company 2 being 0.9 percent per month in August 2008. Note that the fraction of dropped calls is not a perfect measure of network congestion. Dropped calls may occur for reasons unrelated to the quality of the network infrastructure or excess demand for services on the existing network. For example, weather and geographic conditions can also affect the fraction of calls dropped. We use the dropped-call rate as one indicator of network performance, although we recognize that other factors can also contribute to variation in this outcome. 6.3.3

Geographic Data

Since profit-maximizing cell phone providers take the cost of building and maintaining cell towers into account when deciding how to optimally expand the cellular network, we collated important geographic data to be used in our rollout analysis. The source of the elevation data we use is the national map seamless server that contains high-analysis international geographic data.11 As mentioned in section 6.3.1, the elevation data were used in the Viewshed analysis, in order to determine which EAs were covered by towers. Moreover, altitude data were used to calculate the slope of each point in the map. This was done using the slope calculating tool in ArcGIS.12 The average altitude (measured in meters at the EA level) and slope (measured in degrees at the EA level) are used as controls in our regression analysis. The slope is essentially a measurement of the steepness of the terrain. The EAs with precipitous cliffs or mountains have steeper slopes than EAs that are located in plains or plateaus. We created a measure of distance to the nearest road for each EA (using 11. http://seamless.usgs.gov/index.php. 12. Following the process described athttp://webhelp.esri.com/arcgisdesktop/9.2/index .cfm?TopicName=Calculating_slope.

228

Batzilis, Dinkelman, Oster, Thornton, and Zanera

the centroid of the EA as the starting point) where the national roads data from 1998 was provided from the National Statistics Office. Panel C of table 6.1 shows some simple summary statistics on geographic data at the EA level—the average altitude and slope observed in the data. The average height of an EA is almost one kilometer above sea level, although there is a great deal of variation across the country. Malawi’s terrain varies from the high plateaus in the northern and central parts of the country (west of Lake Malawi) to the relatively flatter area around the Shire River in the south. We also see in table 6.1 that rural areas range from being immediately next to a road to more than 50 km distant. 6.3.4

Demographic Data at Baseline

Baseline demographic data comes from the 1998 National Census, provided by the National Statistics Office in Malawi. We obtained a 100 percent sample containing detailed information on each household’s membership, levels of education of adults, occupation, religion, and other variables. One important aspect of the data that is worth mentioning is the limited information on income and labor force participation. These questions were not asked in 1998, possibly due to the fact that the majority of Malawians are subsistence agricultural workers and collecting accurate income and labor data would lengthen the questionnaire. To link the national census to the geographic data, we use spatial files also from the National Statistics Office, which were collected in 1998 during the census. These files contain geographic information on the boundaries of the administrative divisions and the census enumeration areas. The census data follow the administrative structure of the country— regions, districts, and either traditional authorities (TAs) or administrative wards. Malawi consists of three regions—Northern, Central, and Southern. Each region is divided into districts, with a total of twenty-seven in the country. Districts usually contain at least one larger peri-urban center with a district hospital, police unit, and commissioner’s office. Each district contains a number of traditional authorities in the rural areas; in 1998 there were a total of 250 traditional authorities. Traditional authorities are governed by a “traditional authority,” which is a nonelected office and determined by the tribal politics of the area. The four urban centers of Malawi consist of Blantyre, Lilongwe, Mzuzu, and Zomba; we exclude these urban areas from our empirical analyses. Enumeration areas (EA) are defined by the National Statistics Office mainly for data collection, while districts, traditional authorities, and administrative wards are common political divisions used by the government and other institutions. There are approximately 9,200 EAs across the country, and we observe a total of 8,924 of them. We restrict to the 8,118 rural EAs for the analysis. The lowest level of disaggregation consists of villages, governed by a village chief. The EAs do not uniquely contain villages.

New Cellular Networks in Malawi

229

Panel D of table 6.1 shows summary statistics for the census demographic measures at the EA level. As noted above, Malawi has a relatively youthful population with an average age of twenty-two years in rural areas. Education levels are low with an average of under three years of completed schooling for females (males) ages fifteen to forty-nine (fifty-four). Almost 90 percent of adults in rural areas work in agriculture, and the average population density is high, at 314 persons per square kilometer, although this average masks considerable spatial variation in density. An important aspect of the 100 percent census data is that they provide the most complete, albeit crude, set of proxies for market demand that cellular phone network engineers would have had available to them at the beginning of the period. The data also represent the most complete picture of population density and socioeconomic characteristics for most years between 1998 and 2010. A more recent census was conducted in 2008, but the data are still unavailable to the public. As a result, apart from smaller surveys (an income and expenditure survey from 2004 and Demographic Health Surveys), companies would have had to collect their own marketlevel data to supplement the national coverage of the 1998 census or use published population projections in order to construct up-to-date measures of potential market demand. What we will see is that the 1998 census data variables are strong predictors of cell coverage expansion, particularly in the earlier and middle periods of this expansion, and that these demographic variables from 1998 are also important predictors of network congestion ten years later. 6.4 6.4.1

Results: Patterns of Rollout and Correlates of Coverage Graphical Evidence

We begin by illustrating our basic results in graphical form, using maps and estimating survival functions. The maps in figures 6.3, 6.4, and 6.5 show our basic descriptive evidence on rollout of cellular phone coverage over time. Figure 6.3 illustrates the areas that had cellular coverage in 1997, when only TNM was active. Coverage is concentrated around the large cities (Lilongwe, the capital, and Blantyre, the industrial and market center), and several of the main tourist destinations around Lake Malawi (i.e., Salima and Mangochi). By 2004 (figure 6.4) coverage has expanded significantly, moving further out from the cities and, specifically, along the main road network. Figure 6.5 shows coverage in 2008. At this point, coverage has expanded significantly beyond the main road network and the majority of the country has some cellular phone access. We should note that, even in this later period when coverage extends more widely, it still appears to be concentrated around major population centers and tourist centers.

Fig. 6.3

Map of cellular phone coverage in 1997 (TNM only)

Fig. 6.4

Map of cellular phone coverage in 2004 (both networks active)

Fig. 6.5

Map of cellular phone coverage in 2008 (both networks active)v

New Cellular Networks in Malawi

233

In figures 6.6A, 6.6B, and 6.6C, we present estimated Kaplan-Meier survival functions for the time to first cellular coverage within our sample of EAs. These graphs serve as the first illustration of our evidence on correlates of rollout. In these figures, “time of coverage” (on the x-axis) denotes when an EA first received cellular network coverage and ranges from 1995 (year 1) to 2009 (year 15). Each line shows the fraction of EAs that have not yet received cellular phone coverage; as coverage spreads across the country, these “uncovered” lines step downward. We plot survival functions separated by one measure of demand and two measures of cost. We treat population density (measured in 1998) as a proxy for potential market demand, while slope and altitude within the EA affect the cost and feasibility of building cellular towers. Although the three factors are clearly related (e.g., population density is higher in flatter EAs), we examine the relationship between each one individually and the time to first coverage for each EA. We show survivor functions for each of the groups ordered from low to high values of slope, altitude or population density, as well as the 95 percent confidence intervals for each group. Population density is clearly correlated with time-to-coverage, as figure 6.6A shows. Places with the lowest population density get coverage much later in the period, while about 25 percent of EAs in the two highest density groups get coverage by 1999. Although all density groups see more EAs getting coverage after 2003 (year 9), the gap between the survivor functions increases after 2004. This suggests that market demand factors are still important in prioritizing network expansion, even after the backbone structure of this network is in place. Figure 6.6B provides evidence that in the early part of the period (1995 to 2003), areas of different slope appeared equally likely to make the transition from no coverage to coverage (the lines and confidence intervals of the survivor functions overlap). After 2003 (year 9), areas with lower values of the slope variable (i.e., flatter areas) are significantly more likely to make this transition, compared to places in the steepest gradient group. Combining this with the prior graph that indicates the important role for population density in early years, the figures suggest that cost considerations appear more important in discriminating among areas in the second half of the period, once the basic network structure was already established. The message from the survival functions for EAs of different altitude in figure 6.6C is more mixed: in some periods, EAs in different altitude groups have very similar chances of receiving cellular coverage by a certain year. For many of the later years, places with lower altitudes actually have a higher likelihood of transitioning to a covered state. While we might think that locations with the highest altitude would be those places most likely to receive network towers since they offer uninterrupted line of sight, these locations may not be the lowest cost options (e.g., maintenance costs may rise as towers are built on taller and taller mountains) nor are they likely to represent

Fig. 6.6A Arrival of cellular coverage at the local level (by 1998 population density [1 = sparse, 4 = dense]) Notes: Unit of observation is the enumeration area (EA). Graphs show Kaplan-Meier survival functions for EAs with different values of slope, altitude, and population density, where exit is into the “covered” state. Range of years is from 1995 (year 1) to 2009 (year 15). Observations without coverage by 2009 are assigned a year value of 15. Sample is restricted to rural areas with positive population in 1998 census.

Fig. 6.6B Arrival of cellular coverage at the local level (by slope group [1 = flat, 4 = steep]) Notes: Unit of observation is the enumeration area (EA). Graphs show Kaplan-Meier survival functions for EAs with different values of slope, altitude, and population density, where exit is into the “covered” state. Range of years is from 1995 (year 1) to 2009 (year 15). Observations without coverage by 2009 are assigned a year value of 15. Sample is restricted to rural areas with positive population in 1998 census.

New Cellular Networks in Malawi

235

Fig. 6.6C Arrival of cellular coverage at the local level (by altitude group [1 = low, 4 = high]) Notes: Unit of observation is the enumeration area (EA). Graphs show Kaplan-Meier survival functions for EAs with different values of slope, altitude, and population density, where exit is into the “covered” state. Range of years is from 1995 (year 1) to 2009 (year 15). Observations without coverage by 2009 are assigned a year value of 15. Sample is restricted to rural areas with positive population in 1998 census.

areas with large market potential (EAs at lower altitudes have higher population density). Hence, the advantage of building towers in lower altitude EAs seems to emerge only later on in the sample period. 6.4.2

Demographic Correlates

While the figures discussed above are illustrative of some of the important factors related to increased network access, many of these factors are correlated (i.e., altitude and population density).13 We next turn to examining the determinants of access in a multivariate regression framework. We begin by using an OLS framework to estimate the correlates of cell phone access in three different years (2000, 2004, and 2008). We estimate a regression of the following form: (1)

Phonekc = b0 + b1PopDens kc + b2 PctAg kc + b3Educ kc +b4 Age kc + gk + hkc .

In this regression, Phonekc is an indicator for phone coverage (binary or continuous) or for quality of network (dropped calls) in TA k and EA c. The 13. In our sample of rural EAs, the correlation between population density and altitude is –0.10, the correlation between density and slope is –0.17, and the correlation between population density and distance to a road is –0.30.

236

Batzilis, Dinkelman, Oster, Thornton, and Zanera

regressions are estimated at the EA level (denoted by c) with a full set of TA fixed effects γk. This means that we identify off of the EA-level variation. Demographic variables (population density, percent of workforce in agriculture, average education, and average age) are computed by aggregating 1998 census data to the EA level. These variables are proxies for potential market demand in two ways. First, they represent the potential customer base for originating calls, especially in areas with more educated adults earning higher incomes. Second, they capture potential customer base for receiving calls from other parts of the country—therefore, even poor areas with high population density could be more attractive markets than areas with lower population densities. Table 6.2 turns to estimating our first correlates of coverage: demographic variables that proxy for potential market size. Panel A estimates the impact of these demographic variables on a binary for coverage in 2000, 2004, and 2008. Panel B estimates similar regressions using the continuous measure of coverage. We find that most of the demographic variables seem to be important in driving coverage timing. Areas with higher education levels and a lower share of individuals working in agriculture, both typically correlated with higher income, tend to get coverage earlier and are still more likely to have coverage in 2008. The relationship between education and cellular network expansion seems to grow stronger over time. Note that our estimation sample excludes the four main urban areas: so even outside of urban areas, it is the relatively richer rural areas that are more likely to receive early and any cellular network coverage. Higher population density is also associated with greater coverage, and this remains true even through 2008.14 6.4.3

Adding Geographic Factors

In table 6.3 we explore the effect of geographic factors on coverage, controlling for demographics, again at the EA level. The structure of table 6.3 mimics table 6.2: in panel A the dependent variable is a binary for coverage, in panel B it is the continuous coverage measure. We are interested in the coefficients on these cost-side variables as well as how the coefficients for the demand-side variables change between tables 6.2 and 6.3. A number of interesting points emerge from this table. First, even when controlling for demographic variables, EAs with smaller values of slope (that is to say, areas with flatter terrain) have a significantly higher probability of receiving cellular coverage in early years while areas of higher altitude are significantly less likely to receive cellular coverage in the later years. Looking at the interaction term, EAs with steeper slope and at higher altitudes also 14. When we estimate these regressions including total population as well as population density, the coefficient on log population density does not move much at all; the coefficient on total population is negative and significant.

237

New Cellular Networks in Malawi Table 6.2

Cellular phone coverage and demographic proxies for market demand OLS 2000 (1)

Log EA population density Percent adults in agriculture Education (yrs.) Age R-squared N Mean of dependent variable

A. Binary coverage 0.020*** (0.00) –0.033 (0.03) 0.018*** (0.00) –0.001 (0.00) 0.558 8,118 0.20

OLS 2004 (2)

OLS 2008 (3)

0.048*** (0.01) –0.136** (0.04) 0.029*** (0.01) 0.016*** (0.00)

0.027*** (0.00) –0.020 (0.03) 0.036*** (0.00) 0.012*** (0.00)

0.475 8,118 0.53

0.367 8,118 0.85

B. Continuous measure of coverage 0.037*** 0.104*** (0.01) (0.01) Percent adults in agriculture –0.303*** –0.688*** (0.08) (0.13) Education (yrs.) 0.038*** 0.087*** (0.01) (0.02) Age 0.001 0.016**   (0.00) (0.01)

Log EA population density

R-squared N Mean of dependent variable

0.527 8,118 0.20

0.556 8,118 0.53

0.259*** (0.03) –1.575*** (0.36) 0.165*** (0.04) 0.034* (0.02) 0.518 8,118 0.85

Notes: This table shows the relationship between cellular phone coverage in different years and market demand variables measured in the 1998 census. The unit of observation is the EA, and the sample is restricted to rural EAs with positive population according to 1998 census data. All regressions contain TA fixed effects and a constant. Standard errors in parentheses are clustered at the TA level. Binary coverage is defined as an indicator of whether more than 50 percent of the area is covered. Continuous measure of coverage indicates the number of towers covering each area. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

have a higher probability of getting coverage in the early years. This result has a fairly natural interpretation that it may only make sense to locate cellular towers in EAs with steep slopes when there is a high point to locate the tower. If population density is very low in areas with high values of slope, then having a tower at a higher point with a completely uninterrupted line of sight may be required to reach more individuals, conditional on this low density. Second, conditional on geographic factors, the relationship between most of the demographic factors (percent of the workforce in agriculture, average

Table 6.3

Cellular phone coverage, demographic proxies for market demand, and geographic variables OLS 2000 (1)

Log EA population density Percent adults in agriculture Education (yrs.) Age Log altitude Log slope Log altitude ∗ log slope Distance to road (km) R-squared N Mean of dependent variable Log EA population density Percent adults in agriculture Education (yrs.) Age Log altitude Log slope Log altitude ∗ log slope Distance to road (km)   R-squared N Mean of dependent variable

A. Binary Coverage 0.010* (0.00) –0.044 (0.03) 0.013** (0.00) –0.002 (0.00) 0.020 (0.02) –0.138*** (0.04) 0.018** (0.01) –0.014*** (0.00) 0.560 8,115 0.20

OLS 2004 (2)

OLS 2008 (3)

0.027*** (0.01) –0.161*** (0.04) 0.025*** (0.01) 0.012*** (0.00) –0.004 (0.03) –0.216** (0.07) 0.021* (0.01) –0.018*** (0.00)

0.000 (0.00) –0.040 (0.03) 0.031*** (0.00) 0.006* (0.00) –0.101*** (0.03) –0.086 (0.06) 0.004 (0.01) –0.025*** (0.00)

0.484 8,115 0.53

0.390 8,115 0.85

B. Continuous measure of coverage 0.024*** 0.064*** (0.01) (0.01) –0.329*** –0.749*** (0.08) (0.13) 0.029*** 0.073*** (0.01) (0.02) 0.00 0.008 (0.00) (0.01) 0.123*** 0.059 (0.03) (0.06) –0.355*** –1.151*** (0.07) (0.16) 0.047*** 0.152*** (0.01) (0.02) –0.019*** –0.037*** (0.00) (0.01) 0.514 8,115 0.20

0.549 8,115 0.53

0.158*** (0.04) –1.717*** (0.36) 0.126*** (0.04) 0.016 (0.02) 0.183*** (0.12) –2.197*** (0.32) 0.284*** (0.05) –0.108* (0.02) 0.507 8,115 0.85

Notes: This table shows the relationship between cellular phone coverage in different years, market demand variables, and geographic features. The unit of observation is the EA, and the sample is restricted to rural EAs with positive population according to 1998 census data. All regressions contain TA fixed effects and a constant term. Standard errors in parentheses are clustered at the TA level. Binary coverage is defined as an indicator of whether more than 50 percent of the area is covered. Continuous measure of coverage indicates the number of towers covering each area. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

New Cellular Networks in Malawi

239

education, and average age) and cellular coverage remains about the same, or becomes even stronger in the early and middle periods. However, once we control for slope, altitude, the interaction of the two and mean distance to the nearest road, the relationship between population density and cellular coverage is weakened, and in the last year, with the binary measure of coverage, is no longer statistically different from zero. This possibly reflects the fact that a large part of the relationship between geographic factors and cellular coverage indirectly captures the relationship between population density and cellular coverage. We noted at the start of the chapter that one of the motivations for estimating the relationship between demand- and cost-side factors and cell network expansion was to see how large a role these factors play in accounting for the pattern of expansion. Our regressions suggest that, with an EA, these demand and cost factors account for about 50 percent of the variation in coverage. This suggests that although there is some possible role for other variables (e.g., political factors or the availability or lack of complementary infrastructure), likely they do not account for the majority of the variation. 6.4.4

Hazard Models

In table 6.4 we move to estimating Cox proportional hazard models where the outcome is coverage. The model we estimate is: (2)

lc(t|Gc ,b ) = l0(t ) exp (Gc′b )

where, lc (t|Gc ,b) is the EA specific hazard rate, l0(t ) is the baseline hazard rate, t is the year in which the EA obtains cellular coverage, and Gc is the set of EA-specific demographic and geographic variables. The Cox proportional hazards model allows us to estimate the relationships between geographic and demographic variables semiparametrically. This model does not make assumptions about the form of l0(t ) (which is unidentified here) but does assume that time-invariant repressors Gc shift the hazard rate around multiplicatively (Cameron and Trivedi 2005). The advantage of these models over the OLS models is that they provide a more complete picture of which areas get coverage faster, without relying on inference from multiple regressions as in tables 6.2 and 6.3. The hazard models use all of the time variation in rollout between 1998 and 2008, not just the three years that we analyze in the OLS analysis. The evidence in table 6.4 largely echoes what we have found already. In this case, the three columns represent three different cutoff values for coverage. In column (1), the area is defined as covered in the first year of any cell phone coverage. In column (2), it is defined as covered in the first year that our coverage measure exceeds 0.5 (recall this does not necessarily mean that 50 percent of the area is covered; it could be that 25 percent of the area is covered by two towers). In column (3), it is defined as covered in the first year our coverage measure exceeds 1.0.

240

Batzilis, Dinkelman, Oster, Thornton, and Zanera

Table 6.4

Cox regressions: Hazard out of no cellular coverage (1)

Log EA population density Percent adults in agriculture Education (yrs.) Age   N

A. Demand covariates only 0.132*** (0.01) –0.541*** (0.09) 0.002 (0.01) 0.079*** (0.01) 8,118

(2)

(3)

0.280*** (0.01) –0.491*** (0.10) 0.066*** (0.01) 0.099*** (0.01)

0.355*** (0.01) –0.358*** (0.10) 0.070*** (0.01) 0.107*** (0.01)

8,118

8,118

B. Demand and geographic covariates 0 0.5 0.114*** 0.210*** (0.01) (0.01) Percent adults in agriculture –0.574*** –0.578*** (0.09) (0.10) Education (yrs.) 0.001 0.086*** (0.01) (0.01) Age 0.077*** 0.093*** (0.01) (0.01) Log altitude –0.015 –0.054*** (0.02) (0.02) Log slope 0.267* –0.051 (0.14) (0.17) Log altitude * log slope –0.039* –0.032 (0.02) (0.03) Distance to road (km) –0.027*** –0.058***   (0.01) (0.01)

Coverage cutoffs Log EA population density

N

8,115

8,115

1 0.272*** (0.02) –0.457*** (0.10) 0.094*** (0.01) 0.106*** (0.01) –0.031 (0.02) 0.058 (0.19) –0.057** (0.03) –0.088*** (0.01) 8,115

Notes: This table shows the relationship between cellular phone coverage and market demand variables measured in 1998 census and geographic features. The unit of observation is the EA, and the sample is restricted to rural EAs with positive population according to 1998 census data. In column (1), the area is defined as covered in the first year of any cell phone coverage. In column (2), it is defined as covered in the first year that our coverage measure exceeds 0.5. In column (3), it is defined as covered in the first year our coverage measure exceeds 1.0. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level..

Regardless of which coverage definition we use, we find that the demandside factors matter as expected (panel A). More dense areas have faster coverage, as do areas with a higher fraction of educated people; more agricultural areas have slower coverage. When we add in geographic factors (panel B), we find some evidence they matter, but it is fairly weak. Higher altitude areas seem to have slower coverage, and the interaction between altitude and

New Cellular Networks in Malawi

241

slope matters in some specifications. Distance to a road is more consistently important. In these models, the demographic factors continue to be strongly influential, even when we include geographic variables. Together, the demographic and geographic results suggest that both demand-side and cost-side factors drive cellular phone coverage. Even though Malawi is one of the poorest countries in the world, we see coverage expanding into areas where it seems likely there is higher potential market demand—that is, those areas with more income and more potential users. This is in line with what each company’s marketing unit highlighted as one of the key factors guiding rollout: market potential. In addition, controlling for these demand-side factors, we see evidence of more coverage in areas that appear to be easier to reach and build on—that is, those with a less severe slope and areas that are less remote. We see similar patterns in rollouts of other types of infrastructure (for example, television and electricity) in both the developing and developed world (Dinkelman 2011; Gentzkow and Shapiro 2008; Jensen and Oster 2009). 6.5

Results: Correlates of Network Performance: Dropped-Call Rates

In our final piece of analysis, we describe variation in traffic congestion rates across EAs, within a TA, for each of the two network providers. Table 6.5 presents estimates from OLS regressions of the form in equation (1) above where the outcome variable is the percent of dropped calls for each company (defined above) in the month of August 2008. In addition to the prior set of controls, we also control cell phone network coverage within the EA in 2007. For the samples of interest, mean network coverage for Company 1 in 2007 is 0.72, while mean network coverage for Company 2 in 2007 is 0.70. From this analysis, we learn that population density in 1998 is an important predictor of the percent of dropped calls for both companies, and that these relationships are of about the same magnitude. This negative relationship is not obvious a priori: with more people using a given network, higher demand pressure could result in a higher dropped-call rate. What we are seeing in Malawi appears to be the reverse: that in places with higher actual demand (as proxied by higher population density in 1998), the performance of the network is better. This suggests that the cell phone providers are responding to high-demand areas by increasing their coverage quality. In fact, they are more than responding to the increased demand such that coverage is actually better, not just the same, in high-demand areas. This is similar to what we see in the developed world: cell phone coverage is typically at its best in large urban areas and worse in more isolated rural areas, despite the lower demands on the network. It is interesting to note that, in contrast to the analysis of rollout, network congestion as measured by the dropped-call rate is not sensitive to most of

5,971 2.03 0.850 5.25

5,968 2.03 0.850 4.43 3.04

–0.0132* (0.007) –0.200*** (0.054) –0.0231*** (0.007) 0.005 (0.005) 0.000 (0.053) –0.0534 (0.143) 0.003 (0.022) 0.009 (0.006)

(2)

5,968 2.03 0.851 4.52 3.60 22.85

–0.0119* (0.007) –0.216*** (0.055) –0.0218*** (0.007) 0.005 (0.005) 0.000 (0.053) –0.103 (0.143) 0.010 (0.022) 0.008 (0.006) –0.0211*** (0.006) 0.006 (0.012)

(3)

8,046 0.93 0.827 16.61

–0.0128*** (0.002) 0.021 (0.014) –0.003 (0.002) –0.00467*** (0.001)

(4)

8,043 0.93 0.827 14.23 3.48

–0.0131*** (0.002) 0.017 (0.014) –0.003 (0.002) –0.00476*** (0.001) 0.0207** (0.008) –0.012 (0.022) 0.000 (0.003) 0.002 (0.002)

(5)

OLS regressions: Company 2

8,043 0.93 0.828 12.03 3.69 19.32

–0.0125*** (0.002) 0.011 (0.014) –0.002 (0.002) –0.00472*** (0.001) 0.0206** (0.008) –0.0172 (0.022) 0.001 (0.003) 0.002 (0.002) –0.003 (0.002) –0.00692* (0.004)

(6)

Notes: This table shows the relationship between network performance for both of the network providers in August 2008 and a set of variables proxying for market demand (from 1998 census data), cost factors, and prior network coverage. The unit of observation is the EA, and the sample is restricted to rural EAs with positive population according to census data. All regressions contain TA fixed effects and a constant term. Standard errors in parentheses are clustered at the TA level. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

N Mean of outcome R-squared F-stat for demand-side variables F-stat for geographic variables F-stat for prior coverage variables

Company 2 coverage in 2007

Company 1 coverage in 2007

Mean distance to road (km)

Log altitude ∗ log slope

Log slope

Log altitude

Age

Education (yrs.)

Percent adults in agriculture

–0.0120* (0.006) –0.192*** (0.055) –0.0265*** (0.007) 0.006 (0.004)

(1)

OLS regressions: Company 1

Network performance measured as percent dropped calls in August 2008

Log EA population density

 

Table 6.5

New Cellular Networks in Malawi

243

the geographic variables that we use to proxy for cost of actually building the network. Finally, prior network coverage in 2007 in the EA is also a strong predictor of the dropped-call rate. Better, more dense coverage by Company 1’s network reduces the dropped-call rate for Company 1 users, and more dense coverage by Company 2’s network reduces the dropped-call rate for Company 2 users. 6.6

Conclusion

This chapter makes use of new, very detailed data on the history and location of each cellular phone tower constructed in Malawi matched with census data, geographic data, and information on network traffic. The goals of this chapter are twofold. First, we provide detailed evidence on cellular phone rollout in a very low-income context. We show that despite the fact that Malawi is extremely poor, cellular rollout occurs rapidly, and within ten years over half of the country had access to at least one network. This stands in stark contrast to the backlog in access to roads and electricity in this largely rural country. Part of the success of the cell phone rollout may be due to a licensing requirement that cell companies provide access to specific areas by predetermined target dates. However, our evidence shows that both cost-side factors and demand-side market potential variables are important for the timing of initial coverage; cell companies are not simply providing coverage in response to MACRA-mandated target areas. This suggests that even in a setting in which the government may struggle to fairly provide public goods, the timing of infrastructure provision and the quality of the services provided for cell phone networks seem to be driven by market-oriented factors.

References Aker, Jenny C. 2008. “Does Digital Divide or Provide? The Impact of Cell Phones on Grain Markets in Niger.” BREAD Working Paper no. 177, Bureau for Research and Economic Analysis of Development. Aker, Jenny C., and Isaac M. Mbiti. 2010. “Africa Calling: Can Mobiles Make a Miracle?” Boston Review, March/April. http://bostonreview.net/africa-calling -phones-economics-jenny-c-aker-isaac-m-mbiti. Alesina, Alberto, Reza Baqir, and William Easterly. 1999. “Public Goods and Ethnic Divisions.” Quarterly Journal of Economics 114:1243–84. Arnquist, Sarah. 2009. “In Rural Africa, A Fertile Market for Cellphones.” New York Times, October 5. http://www.nytimes.com/2009/10/06/science/06uganda.html. Balakrishnan, Angela. 2008. “Mobile Phone Technology Vital to Growth in the Developing World.” The Guardian (London), February 6, no. 32. http://www.the guardian.com/business/2008/feb/07/globaleconomy.mobilephones.

244

Batzilis, Dinkelman, Oster, Thornton, and Zanera

Banerjee, Abhijit, and Rohini Somanathan. 2007. “The Political Economy of Public Goods: Some Evidence from India.” Journal of Development Economics 82 (2): 287–314. Benson, T., J. Kaphuka, S. Kayanda, and R. Chinula. 2002. Malawi: An Atlas of Social Statistics. Zomba: National Statistics Office Malawi. Besley, Timothy, Rohini Pande, Lupin Rahman, and Vijayendra Rao. 2004. “The Politics of Public Good Provision: Evidence from Indian Local Governments.” Journal of the European Economic Association 2:416–26. Buys, Piet, Susmita Dasgupta, Timothy S. Thomas, and David Wheeler. 2009. “Determinants of a Digital Divide in Sub-Saharan Africa: A Spatial Econometric Analysis of Cellular Phone Coverage.” World Development 37 (9): 1494–505. Cameron, A. Colin, and Pravin K. Trivedi. 2005. Microeconometrics: Methods and Applications. New York: Cambridge University Press. Central Intelligence Agency (CIA). 2009. The World Factbook. Langley, VA: CIA Publication. https://www.cia.gov/library/publications/the-world-factbook/index .html. Dinkelman, Taryn. 2011. “The Effects of Rural Electrification on Employment: New Evidence from South Africa.” American Economic Review 101 (7): 3078–108. Easterly, William, and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.” Quarterly Journal of Economics 112 (4): 1203–50. Gentzkow, Matthew, and Jesse Shapiro. 2008. “Preschool Television Viewing and Adolescent Test Scores: Historical Evidence from the Coleman Study.” Quarterly Journal of Economics 123 (1): 279–323. Godlonton, Susan, and Rebecca Thornton. 2010. “Peer Effects in Learning HIV Results.” Working Paper, University of Michigan. Hausman, Jerry. 2010. “Mobile Phones in Developing Countries.” Unpublished Manuscript, Department of Economics, MIT. International Telecommunications Union. 2009. “World Telecommunications Indicators/ICT Database.” http://www.itu.int/ITU-D/ict/publications/world/world .html. Jack, William, and Tavneet Suri. 2009. “Mobile Money: The Economics of M-Pesa.” Unpublished Manuscript, Sloan School of Management, MIT. Jensen, Robert. 2007. “The Digital Provide: Information (Technology), Market Performance and Welfare in the South Indian Fisheries Sector.” Quarterly Journal of Economics 122 (3): 879–924. Jensen, Robert, and Emily Oster. 2009. “The Power of TV: Cable Television and Women’s Status in India.” Quarterly Journal of Economics 124 (3): 1057–94. Khwaja, Asim Ijaz. 2009. “Can Good Projects Succeed in Bad Communities?” Journal of Public Economics 93 (7–8): 899–916. Kimenyi, Mwangi S. 2006. “Ethnicity, Governance and the Provision of Public Goods.” Journal of African Economies 15:62–99. Klonner, Stefan, and Patrick Nolen. 2008. “Does ICT Benefit the Poor? Evidence from South Africa.” Working Paper, University of Essex, UK. Krudy, John P. 2009. “Africa Rings Up Trade Benefits.” The Washington Times, June 22, B5. Manjoo, Farhad. 2010. “How Often Does Your Phone Drop Calls?” Slate, July 22. http://www.slate.com/articles/technology/technology/2010/07/how_often_does _your_phone_drop_calls.html. Mauro, Paolo. 1995. “Corruption and Growth.” Quarterly Journal of Economics 110 (3): 681–712. McGreal, Chris. 2009. “Front: From Congo to Kathmandu, How Mobiles Have Transformed the World.” The Guardian (London), March 3.

New Cellular Networks in Malawi

245

National Statistics Office. 2004. “Malawi Integrated Household Survey 2004.” http:// www.nsomalawi.mw/ publications/ integrated-household-survey/ 200405 -integrated-household-survey.html. Ngowi, Rodrique. 2005. “Africa’s Cellphone Explosion Changes Economics, Society.” USA Today, October 16. Organisation for Economic Co-operation and Development (OECD). 2009. OECD Communications Outlook 2009. Paris: OECD Publishing. Paul Budde Communication Pty Ltd. 2009. “Global Mobile Communications– Statistics, Trends & Forecasts.” http://www.budde.com.au/Research/2009-Global -Mobile-Communications-Statistics-Trends-Forecasts.html. Tanzi, Vito, and Hamid Davoodi. 1998. “Roads to Nowhere: How Corruption in Public Investment Hurts Growth.” Economic Issues 12. Washington, DC: International Monetary Fund. Vail, Leroy, ed. 1989. The Creation of Tribalism in Southern Africa. Berkeley: University of California Press.

7

Mobile Banking The Impact of M-Pesa in Kenya Isaac Mbiti and David N. Weil

7.1

Introduction

M-Pesa is a money transfer system operated by Safaricom, Kenya’s largest cellular phone provider. M-Pesa allows users to exchange cash for “e- float” on their phones, to send e-float to other cellular phone users, and to exchange e-float back into cash. The story of the growth of mobile telephones in Africa is one of a tectonic and unexpected change in communications technology. From virtually unconnected in the 1990s, over 60 percent of Africans now have mobile phone coverage, and there are now over ten times as many mobile phones as landline phones in use (Aker and Mbiti 2010). Even with the story of mobile phones’ growth as a background, the growth of M-Pesa is startling. Within eight months of its inception in March 2007, over 1.1 million Kenyans had registered to use M-Pesa, and over US$87 million had been transferred over the system (Safaricom 2007). By September 2009, over 8.5 million Kenyans had registered to use the service and US$3.7 billion (equivalent to 10 percent of Kenya’s gross domestic product [GDP]) had been transferred over the system since inception (Safaricom 2009). This Isaac Mbiti is assistant professor of public policy and economics at the University of Virginia. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University and a research associate of the National Bureau of Economic Research. We are grateful to Taryn Dinkelman, John Driscoll, Frederik Eijkman, James Habyarimana, Stephen Mwaura, Benno Ndulu, Pauline Vaughn, Dean Yang, and seminar participants at Tulane University and the NBER Africa Success conference for helpful comments and suggestions. Emilio Depetris Chauvin, Federico Droller, Richard Amwayi Namolo, Angeline Nguyen, Scott Weiner, and Jingjing Ye provided superb research assistance. We are grateful to the Financial Sector Deepening (FSD) Trust of Kenya and Pep Intermedius for providing us with data. Financial support for this research was graciously provided by the NBER Africa Success Project. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13367.ack.

247

248

Isaac Mbiti and David N. Weil

explosive growth was also mirrored in the growth of M-Pesa agents (or service locations), which grew to over 18,000 locations by April 2010, from a base of approximately 450 in mid- 2007 (Safaricom 2009; Vaughan 2007). By contrast, Kenya has only 491 bank branches, 500 postbank branches, and 352 ATMs (Mas and Ng’weno 2009). While the mobile telephone is within sight of becoming a mature business, e-money services like M-Pesa are still in their early days and are continually evolving in response to competitive pressures and customer needs. Despite all the attention M-Pesa has received, there is little quantitative evidence on its economic and social impacts. The combination of widespread cellular communication and the ability to transfer money instantly, securely, and inexpensively are together leading to enormous changes in the organization of economic activity, family relations, and risk management and mitigation, among other things. A decade ago, family members in different parts of Kenya had a very limited scope of communicating with relatives in distant parts of the country, and they faced even greater difficulties in sending or receiving remittances. Now, in many cases, appeals for assistance and the availability of resources can be communicated, and money can be transferred almost instantaneously. Among the changes observers have noted are changes in the the nature, pattern, and impact of remittances. Morawczynski and Pickens (2009) observe that M-Pesa users sent smaller but more frequent remittances, which resulted in overall larger remittances to rural areas. They also observe that urban migrants using M-Pesa visited their rural homes less frequently, potentially weakening the social ties between migrants and their home communities. Researchers have also noted the potential of M-Pesa to affect savings. Morawczynski and Pickens (2009) observe that users often keep a balance on their M-Pesa accounts, thereby using the system as a rudimentary bank account despite the fact that the system does not provide interest. In addition, Vaughn (2007) notes that some individuals stored money in M-Pesa due to safety considerations, especially when traveling across the country. Using ethnographic methods in three communities, Plyler, Hass, and Nagarajan (2010) argue that M-Pesa has enabled small businesses to expand and grow and has also increased the circulation of money in these communities. The explosive growth of M-Pesa has inevitably inspired a great deal of discussion about what the system really is and what it could grow to be. Is it simply a low- cost money transfer system competing with (or replacing) modalities such as cheques and Western Union? Is it a nascent form of electronic money that will someday largely displace cash? Can it be used as a savings account? Is it a means by which financial services can be provided to the unbanked? Jack and Suri (2011) report that three out of four M-Pesa users indicate that they use it to save money. Recently, the potential for M-Pesa to be a savings vehicle has received even more attention, as Safaricom and

Mobile Banking

249

Equity Bank have introduced M-Kesho, an interest- bearing savings account that is directly linked to M-Pesa. In this chapter we examine how M-Pesa is being used in Kenya. We combine data from a number of sources including microlevel survey data (the FinAccess surveys), transaction data from M-Pesa agents, price data from money transfer companies, and aggregate data from Safaricom and the Central Bank of Kenya. We pay particular attention to the question of whether M-Pesa is solely a low- value money transfer system or a nascent form of a means of saving, providing broader financial access for people who are unbanked. The rest of this chapter is organized as follows. In section 7.2, we briefly discuss the structure of M-Pesa. In section 7.3, we examine M-Pesa’s role as a money transfer service. We also examine the characteristics of users, explore data on the distribution of withdrawal and deposit sizes, and analyze the effect of M-Pesa on alternative money transfer modalities. In section 7.4, we examine microeconomic evidence of how M-Pesa affects outcomes such as the propensity of individuals to use financial institutions, as well as to accumulate savings. In section 7.5, we explore the monetary aspects of M-Pesa, including the velocity of e-money circulation. Section 7.6 addresses the question of why people do not store much value in their M-Pesa accounts. Section 7.7 concludes. 7.2 7.2.1

M-Pesa Structure Basic Structure

There are three basic transactions that customers conduct with M-Pesa.

• A customer may deposit money at an M-Pesa outlet in return for e-float

(called a “cash- in” transaction). The customer is required to show a valid identification document, and his identity and the amount of the deposit are logged in a book kept at the outlet. Upon receipt of the money, the M-Pesa agent enters the customer’s telephone number and deposit information into his/her cell phone, and the customer waits at the outlet window until he/she receives a confirmation text message that e-float has been deposited. Unless the system is running slowly (which happens occasionally), the whole transaction takes about a minute or less. • A customer may exchange e-float for cash at an M-Pesa outlet (called a “cash- out” transaction.) Again, the customer must show a valid identification document, and the transaction is logged. The customer tells the the shop clerk how much cash he/she wants, then chooses “withdraw cash” on the M-Pesa menu on his phone, enters the amount to be with-

250

Isaac Mbiti and David N. Weil

drawn (plus the relevant fee), and enters the agent number. The agent then receives a text indicating that the transaction is complete, and the agent then gives the appropriate amount of cash to the customer. This whole transaction takes about one minute. • Finally, a user may transfer e-float from his/her phone to another phone. Our study refers to such a transfer as a “person- to-person transfer,” even though one or both of the parties may be an institution or firm. The user enters the phone number of the recipient and the amount to be transferred on his/her cellphone. The sender and recipient each receive a text message stating that money has been transferred. These three basic transactions can be combined in a number of ways. For example, a user may deposit cash and send the full amount deposited to another user, who can then withdraw the full amount transferred. We refer to this use as “deposit- transfer- withdraw.” Alternatively, a user who receives a transfer from one person may transfer the e-float to some other user instead of withdrawing cash. E-float could circulate in this manner indefinitely, like conventional cash. A third usage possibility is where a user deposits cash and then later withdraws it him/herself without having transferred it. Anecdotally, it is said that people do this for safety when they are traveling (Vaughan 2007; Morawczynski 2009). The usage patterns described above can be mixed in varying ways. For example, a user may receive a transfer and withdraw some of the value while transferring some of the remaining amount elsewhere and leaving some e-float in his account for future transactions. Of particular interest to us is a pattern in which a user might receive a transfer and not withdraw it right away for several reasons: to economize on transaction fees, to economize on the effort of going to an M-Pesa outlet, or to benefit from the safety of storing value on a phone rather than in cash. M-Pesa is safer than cash because a personal identification number (PIN) is required to perform any transaction. If a phone is stolen or lost, the M-Pesa funds are safe unless the PIN has been compromised. If the PIN is compromised and funds are transferred to another account, the legitimate account holder can recover his/her funds if they have not been withdrawn by the fraudulent recipient by initiating a transfer reversal through the customer service department. One of our goals is to better understand such patterns of use. One question in particular is how much of the use of M-Pesa is of the deposittransfer- withdraw type. To the extent that it is used just this way, M-Pesa is primarily a simple money transfer service (which is hardly to say that it is not economically important). By contrast, other uses of M-Pesa suggest other functions. To the extent that e-money circulates among several users between an initial cash-in transaction and a final cash- out transaction, it can be seen as an evolving alternative to currency. Similarly, to the extent that people hold e-float balances on their phones for significant periods of time,

Mobile Banking

251

M-Pesa can be seen as having aspects of banking (as will be seen below, one can even view it as paying interest.)1 All M-Pesa e-float is backed 100 percent by deposits held at three commercial banks in Kenya. Interest earned on these deposits is donated to a charity, which allows Safaricom to avoid being regulated as a bank. An extensive description of the arrangements between Safaricom and the network of agents who service M-Pesa users can be found in Eijkman, Kendall, and Mas (2010) and Jack and Suri (2011). 7.2.2

Pricing

Table 7.1 shows the basic pricing scheme for M-Pesa. To deposit money, a user must register with M-Pesa at an agent location. This is a relatively short process and only requires a valid identification document such as a national ID or passport. Recipients of M-Pesa need not be registered. There is a higher fee for sending money to nonregistered users, but they are not charged any fees to withdraw money and are unable to send the money onward since they are unregistered. The overall transaction fee is far lower for sending to a registered user than to a nonregistered user. In practice 70 percent of users are registered, and approximately 90 percent of transactions are conducted by registered users.2 The pricing structure of M-Pesa is simple and intuitive. However, the pricing structure has a number of “notches” in the terminology of Slemrod (2010). These are points at which incremental changes in customer behavior cause discrete jumps in costs. The incentives around notches are far stronger than those observed at “kinks” in price schedules, such as points where the marginal tax rate changes. For example, in the M-Pesa tariff schedule, the fee for withdrawing up to 2,500 KSh is 25 KSh, while the fee for withdrawing 2,501– 5,000 KSh is 45 KSh. Thus, a person who withdraws 2,600 KSh will be paying a marginal fee of 20 KSh (20 percent) on the last 100 KSh with1. As mentioned above, Safaricom and Equity Bank are now introducing a new service called M-Kesho that allows for mobile phone access to a low- cost bank account. There is no charge for opening the account, no periodic fees, and no minimum or maximum balance (M-Pesa has a maximum balance of 50,000 KSh). Balances from 1– 2,000 KSh (approximately 0.13– 25 USD) receive 0.5 percent interest per year; from 2,001– 5,000 KSh, 1 percent per year; from 5,001– 10,000 KSh, 2 percent per year; and above 10,000 KSh ($125), 3 percent per year. Funds can be transferred without a fee from M-Pesa to M-Kesho, although transfer back to M-Pesa costs 30 KSh. M-Kesho also offers microcredit and insurance services. Microloans can be requested for 100– 5000 KSh, with a 10 percent application fee. Loans are approved or rejected based on a credit score determined by looking at M-Pesa, M-Kesho, and Equity Bank account activity in the last six months, and must be paid back within thirty days (a penalty of 3 percent of one’s outstanding balance is charged for every day after this thirty- day period). Insurance can be obtained for 530 KSh for a year if paid all at once, 830 KSh for the year if paid on a monthly basis, or 1,030 KSh for a year if paid on a weekly basis. For the first year, this insurance is limited to personal accident- related expenses (though this is fairly broadly defined), but after a year it is upgraded to full life insurance (150,000 KSh death or permanent disability benefit plus 20,000 KSh funeral expenses). 2. Refer to the data appendix for details on the computation of this variable.

252

Isaac Mbiti and David N. Weil

Table 7.1

M-Pesa fee schedule Transaction range (KShs)

Transaction type

Minimum

Maximum

Customer charge (KShs)

Deposit cash

100

35,000

0

Send money to a registered M-Pesa user

100

35,000

30

Send money to a non-registered M-Pesa user

100 2,501 5,001 10,001 20,001

2,500 5,000 10,000 20,000 35,000

75 100 175 350 400

Withdraw cash by a registered M-Pesa user at an M-Pesa agent outlet

100 2,501 5,001 10,001 20,001

2,500 5,000 10,000 20,000 35,000

25 45 75 145 170

Withdraw cash by registered M-Pesa user at PesaPoint ATM

200 2,501 5,001 10,001

2,500 5,000 10,000 20,000

30 60 100 175

100 20 —

35,000 10,000 —

0 0 0–30

Withdraw cash by a nonregistered M-Pesa user Buy airtime (for self or other) Pay bill transactions Source: Safaricom Price Schedule (2010).

drawn compared to a fee of 1 percent on the first 2,500 KSh withdrawn. The response of users to the price notches in the M-Pesa tariff schedule should be informative about the optimization problems faced by users. Below, we explore this issue by looking at data on the distribution of withdrawal sizes. 7.3 7.3.1

Uses and Economic Impacts of M-Pesa M-Pesa as a Money Transfer System

Survey Results How Money was Sent in 2006 and 2009. Prior to the introduction of M-Pesa, individuals used a mixture of informal and formal channels to transfer money. Larger bus companies such as Akamba Bus Company or Scandinavia Bus Company offered formal money or parcel transfer services, where recipients would collect the funds at a designated bus terminal. How-

Mobile Banking

253

ever, smaller bus companies or independent minibus operators (matatus) would perform these transactions informally, and in some cases the bus driver would carry the funds with the promise to deliver them. In other cases, individuals would disguise money transfers as packages and place them on the bus for delivery to the designated terminal (Kabbucho, Sander, and Mukwana 2003; Morawczynski 2009). The post office offered a variety of different money transfer products including instant money transfer (postapay) and money orders, which would be delivered to the post office closest to the recipient (Kabbucho, Sander, and Mukwana 2003). Banks and money transfer companies such as Western Union or Moneygram also offered transfer services, although their outlet or branch networks were not as extensive as the post office’s. Figures 7.1 and 7.2 show the change in sending and receiving methods between 2006 and 2009. The figures show that the most common methods to send or receive money were through friends, bus companies, or the post office in 2006. Over 50 percent of people sent money using friends, while close to 50 percent received money via this medium. Approximately 20 percent sent money using the post office, while close to 30 percent received funds this way. Other formal methods such as sending money through banks or money transfer companies like Western Union were less common with less than 10 percent using these methods to send or receive funds.

Fig. 7.1

Sending methods: 2006 and 2009

254

Fig. 7.2

Isaac Mbiti and David N. Weil

Receiving methods: 2006 and 2009

The inception of M-Pesa in 2007 dramatically changed the money transfer market. In less than two years from its inception, M-Pesa was the leading money transfer method with over 50 percent sending money via M-Pesa and over 65 percent receiving funds through the system in 2009 (figures 7.1 and 7.2). The emergence of M-Pesa as the dominant money transfer mechanism virtually eliminated the use of post office products, bus companies, and formal channels such as Western Union and banks, where between 3.5 percent and 0.4 percent of individuals now use these methods to send or receive money (figures 7.1 and 7.2). However, sending and receiving funds through friends remains a popular means of money transfer, where 33 percent of individuals send money via a friend and 22 percent receive funds through a friend in 2009 (figures 7.1 and 7.2). Uses of M-Pesa. Figure 7.3 summarizes the data on M-Pesa use from the 2009 Finacess Survey in descending order of frequency. Close to 42 percent of M-Pesa users reported using the system to purchase mobile phone airtime. Approximately 26 percent of users reported using M-Pesa to save money. While this is a relatively high proportion, it is much lower than the 75 percent saving rate reported in Jack and Suri (2011). Close to 20 percent of users also report using M-Pesa while traveling, presumably for safety concerns as discussed in Vaughan (2007) and Morawczynski (2009). Approximately 6 percent of users made donations via M-Pesa, and our experience

Mobile Banking

Fig. 7.3

255

Uses of M-Pesa

in the field suggests this has grown as currently the majority of calls for donations now include an M-Pesa option. Only 6 percent claim to receive payments on M-Pesa, while only 2 percent claim to receive salaries or wages by M-Pesa. Despite these low levels, 50 percent of M-Pesa users report that they would like to receive their main income by M-Pesa, mainly due to speed and accessibility. The main reported reasons for not wanting to use M-Pesa for the receipt of income was a strong cash preference (30 percent) and a fear of losing their phone (25 percent). Surprisingly, 17 percent of those who did not want to receive their income on M-Pesa were worried they could access the money too easily and thus spend it right away, while another 14 percent claimed their salary would not fit in M-Pesa. Almost 4 percent used an ATM to withdraw cash from their M-Pesa account and 3 percent used M-Pesa to buy goods or pay bills. While the bill- paying prevalence was low in 2009, we expect this to grow as Safaricom has initiated a number of strategic partnerships where customers may now pay for goods and services using M-Pesa. For example, several hospitals, insurance companies, schools, and grocery stores now accept M-Pesa payments. As argued in Aker and Mbiti (2010), these partnerships are part of M-Pesa’s evolution from a pure money transfer system into a payment platform and a formal (regulated) financial service.

256 Table 7.2

Isaac Mbiti and David N. Weil Summary statistics of transactions at M-Pesa agents Withdrawals

N Mean Std. dev. Skewness 10th 25th Median 75th 90th

Deposits

Cyber Center

Katito

Homa Bay

Cyber Center

Katito

Homa Bay

3,477 2,757 4,799 4.03 300 500 1,000 2,850 6,370

6,401 1,402 1,854 5.27 250 475 900 1,680 3,000

2,787 5,762 8,671 2.17 390 700 1,970 6,500 18,500

3,544 3,773 5,949 3.07 300 578 1,500 4,000 10,001

2,524 3,425 6,598 3.21 200 390 1,000 3,000 10,000

3,716 5,240 6,790 2.29 500 1,000 2,500 6,475 14,000

Distribution of Withdrawals and Deposits Our data comes from three M-Pesa outlets. As described in Eijkman, Kendall, and Mas (2010), Cyber Center is an urban outlet in the city of Kisumu, which has a population of 350,000. The outlet is located near one of the city’s markets. Katito is small town with a population of roughly five thousand, located in a rural area about a one- hour trip from Kisumu. It also services surrounding rural areas. Homa Bay is classified as a “district” outlet, meaning that it is in a provincial market town with a population of roughly 20,000 on a main highway. Table 7.2 shows data on the distribution of withdrawal and deposit amounts at the three outlets. Figure 7.4 show the histograms of the distribution of withdrawals from each outlet. The most striking finding in this data is the extent to which a large part of the distribution is composed of very small withdrawals. This is most visible in Katito, the rural outlet, where the median withdrawal is only 900 KSh (about US$13). The 10th percentile of the distibution of withdrawals in Katito is 250 KSh, which implies that one- tenth of users pay a commission of 10 percent or more.3 We can also use figure 7.4 to address the issue of whether there is a large response to the price notches in the M-Pesa tariff discussed above. Although we do not perform a formal test, in most applicable cases we see remarkably little evidence of any response to these notches at all. In the case of Katito, for example, the only price notch that is in the range of an appreciable part of the data is at 2,500 KSh. Although there is indeed a point of mass at this level, it is not out of line with what one would expect given the similar masses at round numbers (500, 1,000, 1,500, etc.). Indeed, there were many fewer 3. Although we do not have data that links withdrawals to transfers, it is likely that in most cases, someone who withdraws 250 KSh has just received this as a transfer, which cost the sender 30 KSh. Thus the overall cost of receiving 225 KSh after fees was 280 KSh, a loss of 19.6 percent.

A

B

C

Fig. 7.4 Frequency of M-Pesa withdrawals by site: A, Homa Bay; B, Cyber; C, Katito Note: Dotted lines represent price notches.

258

Isaac Mbiti and David N. Weil

withdrawals at 2,500 KSh than at 3,000 KSh. We see this for the other outlets as well. It is true that in Homa Bay, which has the largest withdrawals, there are large spikes in the distribution at 10,000 and 20,000 KSh, both of which are price notches. Similarly, there is a spike at 10,000 at Cyber Center. This is consistent with users reacting to the incentives of the price notches, but it is also possible that these large spikes are just due to these figures being round numbers. As another test of whether users of M-Pesa change their behavior in response to characteristics of the pricing structure, we examined data on deposit sizes. Specifically, we are interested in the extent to which people transferring money take into account the fees paid by those who withdraw money (and also the transfer fees that they pay themselves). If such deposits are made as part of a deposit- transfer- withdraw transaction, then the total fees of the transaction will be 55 KSh (that is, 30 KSh for the transfer plus 25 KSh for the withdrawal). A depositor who wanted the recipient to end up with, say, 1,000 KSh would have to deposit 1,055 KSh. We think of small deposits in amounts ending with 55 KSh as being “fee aware.” To the extent that we see deposits of such amounts, it suggests to us that there is a confluence of three factors: first, the depositor intends to transfer the full amount of his/her deposit (minus the transfer fee); second, that he/she expects the recipient to withdraw the amount of the transfer received; and third, that the sender wants the recipient to have access to a round- number amount of money. Table 7.3 shows data on the deposits pooled from the three outlets described above. We consider only deposit amounts below 2,600 KSh, because the withdrawal fee rises after 2,500 KSh.4 Our sample is 6,036 deposits. We tabulate deposits based on the last two digits of the deposit size. The table shows that, not surprisingly, the biggest masses of the distribution are at 00, 50, and 25, which are simply round numbers. However, the fourth largest mass in the distribution (6.8 percent) is indeed at 55, which clearly corresponds to awareness of fees. Further, the fifth largest mass in the distribution is at 30 (4.0 percent), suggesting that some depositors are taking into account transfer fees but not withdrawal fees. Nonetheless, our interpretation of this data is that fee- aware deposits are relatively rare. Of course it is not clear which of the three factors described above is failing in most cases. The Impact of M-Pesa on Money Transfer Companies A number of papers have documented the impacts of mobile phones causing reduced price variation in markets. Jensen (2007) and Aker (2010) find that the introduction of mobile phones reduced price dispersion in fish markets in India and grain markets in Niger, respectively. In these instances 4. The withdrawal fee itself is not counted toward the price of a withdrawal. Thus a customer with a balance of KSh 2,525 in her account could receive KSh 2,500 in cash.

Mobile Banking Table 7.3

259

Distribution of final two digits of deposit amounts Final digits

Percentage

00 05 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 other

58.6 0.3 0.4 0.1 1.1 9.5 4.0 0.1 0.3 0.1 11.7 6.8 2.6 0.1 2.2 1.2 0.4 0.1 0.1 0.0 0.2

Note: M-Pesa deposit data from an agent in Kisumu. Deposits smaller than KSh 2,600. N = 6,036.

the mobile phone technology has increased information flows, which has resulted in price reductions. In contrast, the development and introduction of M-Pesa can be viewed as a “disruptive technology” (Bower and Christensen 1995) or an example of “creative destruction” (Schumpeter 1942; Aghion and Howitt 1992), where M-Pesa revolutionized the money transfer industry. As figures 7.1 and 7.2 show, M-Pesa became the dominant money transfer mechanism within two years of its inception. Ethnographic work by Morawczynski (2009) suggests that M-Pesa’s popularity has been driven by its speed, safety, reliability, extensive network of outlets, and its price relative to the alternatives. Prior to the introduction of M-Pesa, Kabbucho, Sander, and Mukwana (2003) document that the cost of instantly sending US$100 through formal channels ranged between US$12 (MoneyGram) and US$20 (bank wire transfer), while the cost of slower formal channels ranged from US$3 (bus companies) to US$6 (postal money order). Compared to these alternatives M-Pesa offered a significantly cheaper method of instantly transferring funds, where the cost of sending US$100 to a nonregistered user by M-Pesa was approximately US$2.50 in early 2008, while the cost of sending to a registered user was even less. The dominance of M-Pesa can also be observed in the the financial statements of the competitors. Gikunju (2009) examines the financial statements

260

Isaac Mbiti and David N. Weil

of the Postal Corporation of Kenya and finds that revenues and profits for its PostaPay money transfer service declined rapidly after the introduction of M-Pesa and suggests that Western Union’s and MoneyGram’s profits have also declined over the same period. Faced with obsolescence, money transfer companies such as Western Union and MoneyGram have responded by cutting prices, even though they are still unable to match M-Pesa’s superior convenience (Gikunju 2009). Figure 7.5 shows the changes in the money transfer price schedule for Western Union and MoneyGram from the pre– M-Pesa period to the post–M-Pesa period. Overall, these figures show a dramatic reduction in the transaction prices of money transfers. On average, the commission (defined as price to send money divided by the amount sent) charged for money transfers fell from approximately 7 percent in 2003 to 3 percent in 2010. However, we cannot entirely attribute this decline to the competitive pressures induced by the M-Pesa revolution, as other factors such as general technological change could reduce transaction costs and thus reduce prices. Therefore, simple before- and- after comparisons of the price changes will not be sufficient to identify the competitive impact of M-Pesa on the prices of competitors. We employ a difference- in-difference estimation strategy in order to identify the impact of M-Pesa on competitors prices. We construct a database of prices for the main formal competitors in Kenya: MoneyGram and Western Union.5 We obtained the pre–M-Pesa price schedules from Kabbucho, Sander, and Mukwana (2003) and the current price schedules from each provider’s website. As each firm uses different price brackets, we created consistent and comparable price schedules by examining the commissions (price/send amount) for send amounts in 100 KShs intervals ranging from each company’s minimum send amount to each company’s maximum send amount. Our empirical strategy exploits the differences in maximum transaction limits between M-Pesa and its competitors. Central Bank regulations place a maximum transaction limit of 35,000 KShs on M-Pesa, while the transaction limits of MoneyGram and Western Union transactions exceed 500,000 KShs. Given these transaction limits, we would expect to see greater competitive pressures due to M-Pesa on transactions below the M-Pesa threshold of 35,000 KShs compared to transactions above that threshold. Figure 7.5 provides some suggestive evidence of this effect. Focusing on Moneygram, we see that the prices for smaller transactions decreased dramatically, while those for large transactions remained more static. A simple comparison of means above and below the 35,000 KShs threshold and across time is shown in table 7.4. This table shows that there were larger reductions in the prices of transfers below 35,000 KShs compared to those 5. The Postal Corporation of Kenya also has an instant money transfer product called PostaPay. However, we were unable to collect pre–M-Pesa prices. We did have early 2008 prices and we do observe the same patterns as we show in our regressions.

A

B

Fig. 7.5 Changes in prices at money transfer companies: A, Moneygram fees in 2003 and 2009; B, Western Union fees in 2003 and 2009.

262

Isaac Mbiti and David N. Weil

Table 7.4

Average prices of transfers as a percent of transfer amount Transfer amount

Pre–M-Pesa Post–M-Pesa Difference

Less than or equal to 35,000

Greater than 35,000

0.1401 (0.0047) 0.0431 (0.0023) –0.097***

0.0468 (0.0001) 0.0112 (0.0002) –0.0356***

Difference –0.0934*** –0.0319*** –0.0703***

Notes: Standard errors in parentheses. Western Union and Moneygram prices only. Transactions where fees exceed transfer amount are excluded. The P values of the T-test for the difference of means: ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

above. We can formally examine this assertion using the following empirical specification: (1)

p jkt = d0 + d1under35 j + d2 post t + d3under35 j × post t +l j + gk + ´ jkt

where pjkt is the commission, defined as price of sending j shillings, under35 is a dummy variable that indicates whether the transaction amount j is less than 35,000 shillings, post is an indicator variable for the post–M-Pesa period (i.e., 2010), λ is a transaction amount fixed effect, and γ is a company fixed effect. The coefficient of interest is δ3, which captures the impact of M-Pesa on prices. The estimates from equation (1) are shown in table 7.5. These results show that the prices of transactions below 35,000 shillings fell by 6 percentage points, which is approximately a 43 percent reduction in the prices of transactions under 35,000 shillings from 2003. Overall, prices in this segment fell from approximately 14 percent to 4 percent; thus, our estimates imply that competitive pressure from M-Pesa accounts for approximately 60 percent of the decline in prices from 2003 to 2010. A potential concern with our estimation strategy is that we could be simply capturing falling trends in prices. Since we only have two periods of data, we cannot include company specific trends in our analysis. However, we can perform some falsification tests to ensure that our results are not spurious. We create a series of false (and arbitrary) thresholds of 100,000, 125,000, and 150,000 KShs and estimate equation (1) using these fake thresholds and restrict the sample to transactions over 35,000 shillings to avoid M-Pesa effects. Table 7.6 shows the results of this falsification exercise. We do not find negative effects of these false thresholds, but we do find small

Table 7.5

Impact of M-Pesa on prices of competitors Transfer fees (as % of transfer amount)

Transfer ≤ 35,000 Postdummy (Transfer ≤ 35,000) × post Constant Additional controls: Company FE Control for transfer amount and transfer amount squared Transfer amount FE Observations R-squared

(1)

(2)

(3)

0.0863*** (0.0063) –0.0356*** (0.0001) –0.0614*** (0.0057) 0.0626*** (0.0005)

0.0715*** (0.0062) –0.0356*** (0.0001) –0.0615*** (0.0057) 0.0723*** (0.0004)

–0.0356*** (0.0002) –0.0687*** (0.0071) 0.0582*** (0.0008)

Yes No No 18,694 0.494

Yes Yes No 18,694 0.533

Yes No Yes 18,694 0.818

Source: Data from Kabbucho, Sander, and Mukwana (2003), Western Union, and Moneygram. Notes: Preperiod is 2003 and postperiod is 2010. The M-Pesa transfer limit is 35,000 KShs. Robust standard errors in parentheses. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level. Table 7.6

Falsification test on the impact of M-Pesa on prices of competitors Transfer fees (as % of transfer amount)

Postdummy Fake threshold × post Constant Fake threshold Additional controls: Company FE Transfer amount FE Observations R-squared

(1)

(2)

(3)

–0.0366*** (0.0002) 0.00699*** (0.0006) 0.0543*** (0.0003) 100,000

–0.0379*** (0.0001) 0.0114*** (0.0005) 0.0543*** (0.0003) 125,000

–0.0382*** (0.0001) 0.0116*** (0.0004) 0.0543*** (0.0003) 150,000

Yes Yes 17,318 0.912

Yes Yes 17,318 0.92

Yes Yes 17,318 0.922

Source: Data from Kabbucho, Sander, and Mukwana (2003), Western Union, and Moneygram. Notes: Preperiod is 2003 and postperiod is 2010. The real M-Pesa transfer limit is 35,000 KShs. Robust standard errors in parentheses. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

264

Isaac Mbiti and David N. Weil

but positive and significant impacts of this threshold suggesting that trends are not driving our results discussed above. While M-Pesa has forced money transfer companies to lower prices, M-Pesa has also induced these firms and other financial firms to improve their products and services. In some cases, firms have partnered with M-Pesa to offer an integrated service. For example, Western Union recently partnered with M-Pesa to offer international money transfers in which migrants in the diaspora can now send remittances to their friends and family via M-Pesa with Western Union serving as an intermediary. PesaPoint, an independent network of ATMs, allows M-Pesa users to withdraw cash using its large network of ATMs. Commercial banks in Kenya were initially opposed to M-Pesa and lobbied the government to regulate M-Pesa and other mobile money platforms under the commercial banking regulations (Njiraini and Anyanzwa 2008). After these efforts failed banks partnered with M-Pesa to offer better services to customers, and in some cases became M-Pesa agents. There is also suggestive evidence that M-Pesa has increased the efficiency of the banking system. According to a 2009 newspaper article, the advent of M-Pesa has caused commercial banks to work toward speeding up the check- clearing process, which took a minimum of three days.6 7.3.2

Characteristics of Users

We use data from the 2009 FinAccess survey to examine basic characteristics of M-Pesa users and their usage patterns. Overall, our data show that approximately 40 percent of Kenyans have used M-Pesa, with close to 30 percent formally registered with Safaricom. As discussed in Aker and Mbiti (2010), M-Pesa users are more likely to be younger, wealthier, better educated, banked, employed in nonfarm sectors, to own cell phones, and to reside in urban areas (table 7.7). We examine cross- tabulations of M-Pesa use by individual characteristics in table 7.8. Males, urban residents, banked individuals, the wealthy, the better educated, and those employed in the nonfarm sector were more likely to use M-Pesa. Higher socioeconomic status individuals are more likely to use M-Pesa to purchase airtime, save and store money while traveling, and use M-Pesa to pay wages than their respective counterparts. Focusing on saving patterns, table 7.8 shows that 35 percent of banked individuals used M-Pesa to save, while only 19 percent of unbanked individuals used M-Pesa to save. Similarly, 30 percent of wealthy individuals report using M-Pesa to save, while only 15 percent of poor individuals report doing so. Similar gaps are also observed between the more educated and less educated individuals. When we examine the characteristics of users who use M-Pesa as a safekeeping mechanism while traveling, we find very similar patterns to those found 6. “Why central bank position on mobile banking attracts wrath,” 2/6/2009, The Standard http://www.standardmedia.co.ke/InsidePage.php?id=1144015709&cid=457&.

265

Mobile Banking Table 7.7

M-Pesa adoption

Gender Female Male Financial access Not banked Banked Residence Rural Urban Wealth Not poor Poor Age Under 55 Over 55 Education At least primary school Less than primary school Employment status Unemployed Employed Not employed in nonfarm job Employed in nonfarm job Access to cell phone Does not have a cell phone Has a cell phone Observations

Use M-Pesa (%)

Send money with M-Pesa (%)

Receive money with M-Pesa (%)

M-Pesa registered user (%)

35.5 42.5

24.7 35.3

32.1 36.6

24.1 32.7

27.5 71.9

17.8 64.0

23.9 65.2

15.9 64.0

28.8 62.2

18.8 54.8

25.9 54.1

18.3 51.0

52.4 18.7

43.2 9.4

46.0 17.1

41.0 9.0

42.4 20.2

32.9 11.9

37.4 18.4

31.1 12.2

55.2 16.4

44.3 9.4

49.2 14.1

42.4 8.5

35.0 39.4 28.9 56.0

23.7 30.8 18.6 48.7

31.1 34.9 26.0 48.8

22.7 29.2 18.5 44.9

14.2 66.0

7.4 53.8

12.0 59.0

2.6 56.3

6,598

6,598

6,598

6,598

Source: Data are from the 2009 FinAccess Survey. Notes: “Poor” is defined as individuals in the bottom two wealth quintiles of an asset index.

in savings. We find that the wealthier, more educated, and banked individuals are each approximately 2.5 times more likely to report using M-Pesa while traveling when compared to their counterparts. We observe large differences in the frequency of M-Pesa use across demographic and economic groups in table 7.9. Individuals with bank accounts use M-Pesa almost three times as much as those without bank accounts. Urban residents, richer individuals, the more educated, and those in the nonfarm sector use M-Pesa almost twice as often as rural residents, poorer individuals, the less educated, and those employed in the farm sector, respectively. While those with mobile phones used M-Pesa three times as often as those without phones, there are much smaller differences between men and women (with men using M-Pesa 35 percent more frequently than women).

19.4 34.9

21.0 32.9

29.5 14.7

27.9 13.5

29.2 14.5

21.9 27.8 20.1 32.7

7.1 31.3

26.8 60.2

30.4 55.4

48.2 17.6

44.4 18.9

48.1 15.3

36.1 43.6 29.8 53.8

10.4 49.9

6,598

Observations

6,598

2.8 22.2

14.2 19.4 12.4 24.0

20.5 8.8

18.8 13.5

20.7 8.9

14.1 23.1

10.9 27.1

16.7 20.2

Use M-Pesa when traveling (%)

6,598

0.8 7.9

5.4 6.8 4.1 8.7

7.7 1.1

6.7 4.5

7.8 1.2

4.3 9.0

1.9 12.0

4.8 8.4

Use M-Pesa to make donations (%)

Source: Data are from the 2009 FinAccess Survey. Notes: “Poor” is defined as individuals in the bottom two wealth quintiles of an asset index.

6,598

23.6 30.0

Use M-Pesa to save money (%)

36.6 48.5

Use M-Pesa to buy airtime (%)

M-Pesa use by user characteristics

Gender Female Male Financial access Not banked Banked Residence Rural Urban Wealth Not poor Poor Age Under 55 Over 55 Education At least primary school Less than primary school Employment status Unemployed Employed Not employed in nonfarm job Employed in nonfarm job Access to cell phone Does not have a cell phone Has a cell phone

Table 7.8

6,598

2.9 6.4

3.2 6.4 3.8 7.5

6.2 3.4

5.9 4.1

6.6 2.3

4.4 7.2

2.6 9.4

4.7 6.9

Use M-Pesa to receive payments (%)

6,598

0.4 4.6

3.2 3.9 2.1 5.4

4.5 0.6

3.9 2.9

4.6 0.6

1.6 6.3

0.3 7.9

3.0 4.7

Use M-Pesa to withdraw from ATM (%)

6,598

0.4 4.0

2.3 3.5 1.8 4.6

3.7 1.5

3.3 2.5

3.6 1.7

2.3 4.4

1.2 5.7

2.1 4.7

Use M-Pesa to buy goods/ services (%)

6,598

0.4 3.2

2.0 2.9 1.7 3.6

3.0 1.1

2.7 2.0

3.3 0.4

1.5 4.0

0.7 5.0

2.1 3.4

Use M-Pesa to pay bills (%)

6,598

1.4 2.0

1.4 2.0 1.7 2.0

2.0 1.1

1.8 2.0

2.0 1.2

1.8 1.9

1.5 2.2

1.2 2.6

Use M-Pesa to receive wages (%)

6,598

0.8 2.0

0.9 2.0 1.0 2.5

2.0 0.8

1.9 0.8

2.1 0.4

1.4 2.2

0.6 3.1

1.2 2.5

Use M-Pesa to pay wages (%)

Mobile Banking Table 7.9

267

Cross-tabulations of frequency of M-Pesa use Frequency of use (annual)

Use M-Pesa daily (%)

Use M-Pesa weekly (%)

Use M-Pesa monthly (%)

Use M-Pesa irregularly (%)

15.7 21.4

1.2 2.2

12.5 16.6

32.4 33.1

53.9 48.1

10.4 27.8

0.4 3.1

8.1 21.9

30.3 35.6

61.2 39.3

13.5 23.9

0.9 2.5

10.3 19.1

31.3 34.4

57.5 44.0

20.4 10.2

1.9 0.6

16.4 6.6

33.9 28.2

47.8 64.7

18.8 13.4

1.6 1.2

15.1 7.8

32.9 30.7

50.3 60.2

20.2 10.3

1.9 0.4

16.0 7.6

33.3 30.0

48.8 62.0

Gender Female Male Financial access Not banked Banked Residence Rural Urban Wealth Not poor Poor Age Under 55 Over 55 Education At least primary school Less than primary school Employment status Unemployed Employed Not employed in nonfarm job Employed in nonfarm job Access to cell phone Does not have a cell phone Has a cell phone

15.1 19.2 12.9 23.5

1.1 1.8 0.8 2.4

11.7 15.2 9.8 18.9

33.4 32.5 31.5 34.0

53.9 50.5 57.9 44.8

6.2 21.3

0.0 2.0

2.9 17.3

26.9 34.2

70.1 46.5

Observations

6,598

6,598

6,598

6,598

6,598

Source: Data are from the 2009 FinAccess Survey. Notes: “Poor” is defined as individuals in the bottom two wealth quintiles of an asset index.

As columns (2) to (5) suggest, these disparities are mainly driven by differences in daily and weekly use, in which the banked are almost three times as likely to use M-Pesa daily or weekly as the unbanked, and urban residents are almost twice as likely to use M-Pesa daily or weekly compared to rural residents. While daily and weekly users are generally more affluent, educated, and urban, they only account for 1.6 and 14.4 percent of all users, respectively, while 32.7 percent of users are monthly users, and 51.3 percent are irregular users. However, using our annualized measure of M-Pesa usage, we find that daily users (1.6 percent of users) account for 32 percent of transactions, weekly users account for 41 percent of transactions, monthly users account for approximately 21 percent of transactions, and irregular users account for only 6 percent of transactions.

268

Isaac Mbiti and David N. Weil

Overall, these simple cross- tabulations of the intensity of M-Pesa use and the main uses of M-Pesa by individual characteristics reveal that the most intense users generally have higher socioeconomic status. Moreover, theses higher SES individuals are also more likely to use M-Pesa in ways that could reap large economic gains, such as savings. Taken together, these patterns perhaps suggest that more affluent members of society are among the biggest beneficiaries of M-Pesa. This, of course, does not preclude poorer and more vulnerable members of society from reaping significant economic and social benefits from M-Pesa. More research will be needed to examine the extent to which M-Pesa benefits are distributed across socioeconomic strata. 7.4

Economic Impacts of M-Pesa: Microlevel Evidence

Morawczynski and Pickens (2009) find that M-Pesa has changed the patterns of remittances. This observation is supported by the 2009 FinAccess surveys that show that almost 35 percent report that they have increased the frequency of sending transfers due to M-Pesa, while 31 percent report an increase in the receipt frequency of transfers due to M-Pesa (figure 7.6). Surprisingly, 18 percent report a decrease in the sending frequency, while

Fig. 7.6

Change in transfer frequency due to M-Pesa usage

Mobile Banking

Fig. 7.7

269

Change in amount sent or received via M-Pesa

22 percent report a decrease in the receiving frequency, with the remainder reporting no change in transfer frequency. Figure 7.7 shows the change in the amount of transfers received. Almost 35 percent of users claim that they sent larger transfers due to M-Pesa, while 30 percent claim to have received larger transfers because of M-Pesa. In contrast, roughly 20 percent report decreases in the amount of transfers sent or received, with the remainder reporting no change in the amount of transfers received or reported. We find very strong correlations between reported changes in transfer frequency and reported changes in the amount transferred. Over 85 percent of individuals report the same effect for both changes in frequency and changes in transfer amount (for both sending and receiving). For example, 87 percent of individuals who claim to have received transfers more frequently report that the amount of transfer has also increased, and a very small percentage report sending smaller transfers more frequently. This suggests that people do not, in fact, send smaller transfers more frequently as reported in Morawczynski and Pickens (2009). However, as we have no data on the extent or magnitude of these changes, we are unable to examine the magnitude of changes in the frequency or size of transfers due to M-Pesa.

270

Isaac Mbiti and David N. Weil

The qualititative studies on M-Pesa such as Morawczynski and Pickens (2009) have suggested that M-Pesa serves as a partial substitute for the formal banking system. Prior to the introduction of M-Pesa, most Africans were excluded from modern financial services. Using data ranging between 2001 and 2005, Beck, Demirgüç-Kunt, and Peria (2007) show that African countries lagged in financial access. During this period they show that Ghana had 1.6 branches per 100,000 and Kenya had 1.3 branches per 100,000, while Uganda and Tanzania both had less than 0.6 branches per 100,000. The ATM penetration of these countries was even lower—ranging from 1 per 100,000 in Kenya to less than 0.20 per 100,000 in Tanzania. In contrast, the United States had thirty- one bank branches and 120 ATMs per 100,000 people during that period. Perhaps partly as a result of the small banking networks in many African countries, a low proportion of individuals have a bank account. On average the FinScope surveys show that 30 percent of East and Southern African adults have a formal bank account (FinMark Trust 2008). These proportions range from a high of 63 percent in South Africa to low of 9 percent in Tanzania. With the low levels of financial development in many African countries, many observers have identified the potential for systems such as M-Pesa to expand the reach of the financial system and provide a platform to deliver financial services to the poor and excluded. Burgess and Pande (2005) show that the expansion of rural banking in India significantly reduced rural poverty rates. While this was mainly driven by increased access to credit, mobile systems such as M-Pesa could facilitate the expansion of branchless banking, in which banks increase the financial reach using agents as intermediaries to provide services to clients in rural and remote areas where the fixed costs of opening a branch would be prohibitive (Pickens, Porteous, and Rotman 2009). This possibility, however, is contingent upon banks’ willingness to serve poorer clients and upon government regulations that promote or hinder branchless banking. A number of qualitative studies such as Morawczynski and Pickens (2009) and Mas and Morawczynski (2009) have explored the economic and social impacts of M-Pesa in Kenya. For instance, Morawczynski and Pickens (2009) find ethnographic evidence that M-Pesa has changed savings behavior and the pattern of remittances, and has increased rural livelihoods. While these studies provide suggestive evidence of the impacts of M-Pesa, they are generally unable to quantify the effects of the system and are limited by their small sample sizes. An exception is Jack and Suri’s (2010) empirical study that shows that M-Pesa improves the ability of households to smooth risks. We contribute to the literature by providing quantitative estimates of the impact of M-Pesa in Kenya on a variety of economic and social outcomes including financial access and usage. We combine the 2006 and 2009 FinAccess surveys and create a balanced panel of the 190 sublocations that were surveyed in both rounds in order to examine the economic impact of M-Pesa on various outcomes pertaining to remittances, financial access, and

271

Mobile Banking Table 7.10

Summary statistics from estimation sample 2009

Frequency of domestic transfers received Frequency of domestic transfers sent Frequency of international transfers received Frequency of international transfers sent Banked Uses informal saving product Hides money for savings Uses formal saving product Has a formal loan Has an informal loan Employed Employed in nonfarm sector Percent that feel sending money with a friend is risky Percent that feel sending money with postal service is slow Percent that own a cell phone Observations

2006

Mean

SD

Mean

SD

4.303 2.547 0.298 0.0762 0.285 0.729 0.522 0.342 0.101 0.331 0.755 0.366 0.335

(3.544) (3.063) (0.840) (0.295) (0.233) (0.196) (0.239) (0.251) (0.120) (0.211) (0.153) (0.261) (0.208)

1.125 1.079 0.17 0.0478 0.186 0.518 0.251 0.271 0.0903 0.255 0.799 0.364 0.335

(1.580) (1.678) (0.565) (0.225) (0.200) (0.275) (0.256) (0.241) (0.131) (0.262) (0.157) (0.267) (0.208)

0.245 0.294

(0.193) (0.273)

0.245 0.294

(0.193) (0.273)

190

190

190

190

Notes: Data from the FinAccess survey aggregated to the sublocation level. Only sublocations that were in both rounds were included in the estimation sample.

economic livelihood.7 Sublocations are the smallest administrative unit in Kenya and consist of two to three villages in rural areas or a large neighborhood in a city. The summary statistics of this estimation sample is shown in table 7.10. We examine the relationship between M-Pesa and various economic and social outcomes at the sublocation level using the following specification: (2)

y jt = b0 + b1mpesa jt + X ′jtb2 + b3Tt + m j + y jt

where mpesajt is the proportion of individuals that use M-Pesa in sublocation j at period t; X is a vector of controls including education, gender, age, marriage rate, and wealth; T is a time fixed effect; μ is the sublocation fixed effect that captures time invariant unobservable variables at the sublocation level; and υ is an idiosyncratic error term. Also, y is a set of outcomes variables that includes frequency of sending and receiving transfers, possession of a bank account, saving methods, and employment. Simple regression estimation of equation (2) will lead to biased and inconsistent estimates if the time- invariant unobservables (μ) or the timevarying unobservables (υ) are correlated with M-Pesa use and our set of outcome variables. To circumvent this we employ a sublocation fixed effects 7. See data appendix for more details on the construction of the estimation sample.

272

Isaac Mbiti and David N. Weil

instrumental variable (FE-IV) procedure to eliminate the time- invariant heterogeneity and biases due to endogenous M-Pesa adoption. Specifying Δ as the sublocation first- difference operator, we can estimate the following fixed effect regression: (3)

Dy jt = b0 + b1Dmpesa jt + D X ′jtb2 + Dy jt .

While biases due to time- invariant unobservables are eliminated in equation (3), the estimates will still be biased and inconsistent if Δ is correlated with Δmpesa. We need an instrument (or a set of instruments) that predicts M-Pesa use but does not directly impact our set of outcomes. Both rounds of the data contain perceptions of the most common money transfer methods; however, we focus solely upon the 2006 perception data as the 2009 perceptions would be influenced by M-Pesa. Respondents are asked to identify the riskiest, slowest, and costliest money transfer method. We focus on the proportion of residents that identify sending money with a friend as the riskiest method, the proportion of residents that identify the post office as the slowest, and the proportion that identify money transfer companies as the most expensive. If more respondents in a sublocation feel that their alternative means of transferring money are inefficient, they would be more likely to adopt M-Pesa. Moreover, conditional on the sublocation fixed effect, this 2006 perception should have no direct impact on outcomes (or the change in outcomes). The identification assumption is conditional on the vector of controls (such as wealth and education) and the sublocation fixed effect, the perceptions of the alternative methods will only indirectly affect the set of outcomes (such as banking) through M-Pesa adoption. We can specify the set of estimating equations for the FE-IV regression as: (4)

Dy jt = b0 + b1Dmpesa jt + D X ′jtb2 + Dy jt

(5)

Dmpesa jt = a0 + Z ′j 0 a1 + D X ′jt a2 + Dn

where Z is the set of instruments: the proportion that rank friends as the riskiest method to transfer money in 2006, the proportion that rank the post office as the slowest method in 2006, and the proportion that rank money transfer companies as the most expensive option in 2006. The extent to which transferring funds through friends is risky will be mostly determined by social capital and crime. In terms of financial access, the most plausible concern is that banks are less likely to locate in these areas due to security concerns. Since these areas are more likely to adopt M-Pesa, then this would lead to an underestimate of the impact of M-Pesa adoption on financial access. There are a number of factors that could determine efficiency of money transfers via the post office. First, these could reflect the motivation of post office employees. Employee motivation could be driven by the quality of supervision. If better supervisors were located in faster growing areas (which were more likely to see expansions of financial services), then this would also lead to underestimates of the impact of

273

Mobile Banking

M-Pesa adoption. Alternatively, the speed of the post office could reflect the quality of transportation links or local infrastructure (e.g., electricity, telephone links). If financial institutions were less likely to expand to these more “isolated” areas, then this would again lead to an underestimate of the impact of M-Pesa adoption. However, if these institutions were more likely to expand in these areas, then our methodology would overestimate the impact of M-Pesa adoption. However, we feel that the costs of operating in isolated areas may be prohibitive for banks and thus we feel that they are unlikely to expand in these areas. Since the price schedule for money transfer companies does not vary within Kenya, the perceptions of cost are likely driven by marketing and word of mouth. If these companies target their marketing in faster growing areas (which were more likely to see expansions of financial services), then this would also lead to underestimates of the impact of M-Pesa adoption. The results from equations (2) and (3) are shown in table 7.11. The estimates from the random effects specifications show a positive relationship Table 7.11

Impact of M-Pesa on transfers, employment, and financial access M-Pesa use Random effects (1)

Fixed effects (2)

Dependent variable

Coefficient

Standard error

Coefficient

Standard error

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

0.286*** 0.374*** 1.03 3.386*** 0.179*** 0.209*** –0.0785 –0.187** 0.152*** –0.011 –0.00328 0.115** 0.0841

(0.064) (0.053) (1.069) (0.669) (0.040) (0.069) (0.080) (0.083) (0.043) (0.033) (0.069) (0.053) (0.054)

0.176*** 0.262*** 0.956 3.553*** 0.189*** 0.161** –0.277*** –0.357*** 0.152*** 0.00334 –0.0892 0.133** 0.084

(0.061) (0.064) (1.184) (0.670) (0.044) (0.080) (0.094) (0.105) (0.045) (0.035) (0.089) (0.061) (0.062)

Percent receive a transfer Percent sent a transfer Frequency of transfers received Frequency of transfers sent Banked Belong to group Informal saving Hide money for savings+ Formal savings Formal loan Informal loan Employed Employed in nonfarm job

Notes: Each row is the coefficient on M-Pesa use from separate regressions. Data from the FinAccess survey aggregated to the sublocation level. Only sublocations that were in both rounds were included in the estimation sample. Sublocation fixed effects and sublocation random effects estimates. + hide money for savings is a subset of informal savings, and frequency of transfers only includes domestic transfers. Additional controls: male, married, education, age, wealth, and year. There are 380 observations in each regression specification. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

274

Isaac Mbiti and David N. Weil

between M-Pesa adoption and frequency of sending and receiving transfers, although only the estimate of sending transfers is statistically significant. The estimates also show a strong positive association between M-Pesa adoption and bank use, formal savings, and employment. In addition, the estimates show a negative and statistically significant relationship between M-Pesa adoption and saving money using secret hiding places. Similar patterns are observed in fixed effect specifications. The point estimates on sending remittances, bank use, formal savings, and employment are very similar when compared to the random effects specifications. However, we do observe the larger negative correlations between M-Pesa and informal savings and using a secret hiding place to save money. We estimate equations (4) and (5) in order to obtain causal estimates of the impacts of M-Pesa. Table 7.12 shows first- stage relationship between our set of instruments and the endogenous variables. The estimates show that M-Pesa adoption was positively correlated with greater proportions of individuals who rank using friends as the riskiest money transfer method. Similarly, perceived slowness of transferring funds using the post office in period 0 and the perceived cost of money transfer companies have positive and significant effects on M-Pesa adoption. This set of instruments is highly significant, with a joint F-test of 26, which is well above the weak instrument thresholds. Table 7.12

First-stage relationship M-Pesa adoption (1)

Percent that rate sending money with a friend is riskiest Percent that rate sending money with post office is slowest Percent that rate sending money with MTC is costliest Constant Observations R-squared First-stage F

0.169** (0.0838) 0.298*** (0.0988) 0.589*** (0.0873) 0.111*** (0.0430) 190 0.351 26.09

Notes: Data from the FinAccess surveys aggregated to the sublocation level. Only sublocations that were in both rounds were included in the estimation sample. First difference/ fixed effect estimates. Robust standard errors clustered at sublocation level in parentheses. There are 190 observations rather than 380 as the data is first differenced. Additional controls: male, married, education, age, wealth, and year; MTC is a money transfer company such as Western Union or MoneyGram. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Mobile Banking Table 7.13

275

Impact of M-Pesa on transfers, employment, and financial access M-Pesa use

Dependent variable

Coefficient

Standard error

(1) Percent receive a transfer (2) Percent sent a transfer (3) Frequency of transfers received (4) Frequency of transfers sent (5) Banked (6) Belong to group (7) Informal saving (8) Hide money for savings+ (9) Formal savings (10) Formal loan (11) Informal loan (12) Employed (13) Employed in nonfarm job

0.171 0.299*** 2.749 5.300*** 0.279*** 0.432*** –0.383** –0.772*** 0.273*** 0.00385 0.0456 0.308*** 0.094

(0.109) (0.108) (2.442) (1.257) (0.082) (0.125) (0.184) (0.202) (0.084) (0.080) (0.159) (0.116) (0.112)

Falsification exercise: (14) Frequency of international transfers received (15) Frequency of international transfers sent

0.295 –0.0861

(0.661) (0.130)

Notes: Each row is the coefficient on M-Pesa use from separate regressions. Data from the FinAccess survey aggregated to the sublocation level. Only sublocations that were in both rounds were included in the estimation sample. First difference/fixed effect-instrumental variable estimates. Robust standard errors clustered at sublocation level in parentheses. ^ Denotes Endogenous Variables. Excluded instruments are percent rate transfers by post office are slower, percent rate transfers by friend are riskiest, percent rate money transfer company are most expensive; + hide money for savings is a subset of informal savings, receipt and frequency of transfers in rows (1) to (4) only includes domestic transfers. There are 190 observations rather than 380 as the data is first differenced. Additional controls: male, married, education, age, wealth, and year. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

The FE-IV estimates of equation (4) and (5) are shown in table 7.13. These estimates show that M-Pesa adoption led to increases in the frequency of sending transfers. The point estimate shows that if M-Pesa were universally adopted, individuals would send five more remittances per annum. Evaluating this point estimate using the mean M-Pesa adoption rate of 40 percent, we see that M-Pesa increased the frequency of sending remittances by two, which is more than double the 2006 level. Our estimates imply that M-Pesa accounts for almost the entire increase in the sending frequency of transfers between 2006 and 2009 (table 7.10). This is consistent with figure 7.6, which shows that 35 percent report increases in the frequency of sending transfers due to M-Pesa. While we observe significant increases in the sending frequency of transfers, we surprisingly do not find any effect of M-Pesa on the

276

Isaac Mbiti and David N. Weil

frequency of receiving transfers, even though 30 percent report increases in frequency of receiving transfers due to M-Pesa. While M-Pesa has been touted for banking the “unbanked,” there are no estimates of the direct impact of M-Pesa on people adopting bank accounts. Row 5 of table 7.13 provides this evidence. These estimates show that increased M-Pesa adoption leads to greater bank use. The point estimates imply that universal adoption of M-Pesa would increase the proportion banked by 28 percentage points. Evaluated at the mean adoption rate of 40 percent, we see that M-Pesa has increased the proportion banked by almost 11 percentage points, which represents a 58 percent increase over the 2006 banking level. As the data was collected prior to the integration of M-Pesa with banks, this result could be driven by increases in money (or cash) by users. It could also be driven by the complementarity between M-Pesa and banks. If M-Pesa were more valuable or useful in combination with a bank (or vice versa), then increases in demand for M-Pesa would also increase the demand for banking. This evidence provides some evidence that M-Pesa does not entirely serve as a substitute for the formal banking system, but, rather, is viewed (or used) as a complementary tool by individuals. Qualitative evidence from Morawczynski and Pickens (2009) suggests that M-Pesa is used as a saving instrument. This notion is supported by the 2009 round of the Finaccess survey in which over 25 percent of individuals report using M-Pesa as a saving device. While we do not have data on the amount saved, we do have information on the methods used to save and can therefore examine the impact of M-Pesa on savings methods. Row 7 of table 7.13 shows the impact of M-Pesa on the use of informal saving mechanisms. Informal saving mechanisms include rotating saving and credit associations (ROSCA), saving with a group of friends, savings given to a family or friend for safekeeping, and saving by storing funds in a secret place. While the summary statistics show that the proportion of individuals using informal methods to save has increased from 52 percent to 72 percent, our estimates show that M-Pesa decreases the use of informal saving mechanisms. Evaluated at the mean M-Pesa adoption rate, M-Pesa would reduce the prevalence of informal saving by 15 percentage points, approximately a 30 percent reduction from the 2006 level. We observe similar effects for the use of secret hiding places to save money. Row 8 of table 7.13 shows that for the average adoption rate, M-Pesa would reduce the proportion of people saving money in secret places by 30 percentage points, which is slightly greater in magnitude than the 2006 level. Since we do not observe any changes in the use of formal savings methods (which do not include M-Pesa), these results suggest that users are shifting savings from informal tools to M-Pesa, perhaps due to the superior security of M-Pesa. M-Pesa could also affect economic activity directly by increasing access to funds and indirectly by increasing savings and banking rates. Plyler, Hass, and Nagarajan (2010) argue that M-Pesa has promoted the growth rates of

Mobile Banking

277

(small- scale) firms in the communities they studied, and they argue that this was largely driven by the increased circulation of money in these communities. Figure 7.7 provides some supportive evidence of the increase in funds due to M-Pesa, in which almost 35 percent of individuals report that they sent larger transfers due to M-Pesa, while close to 30 percent report that they received larger transfers due to M-Pesa. We use employment as a measure of economic activity and examine the impacts of M-Pesa on employment. We use a measure of employment that incorporates farm labor (own- farm and on others farm), nonfarm labor (such as civil service employment), and self- employment (such as owning a shop). Individuals are considered employed if they are actively engaged in any of these activities. Row 12 of table 7.13 shows that M-Pesa is associated with increases in any type of employment. For the average M-Pesa adoption level, M-Pesa would increase employment by 12 percentage points, approximately a 15 percent increase from the 2006 employment level. While this is encouraging, column (7) shows no impact of M-Pesa on nonfarm employment. This suggests that the increases in employment due to M-Pesa are driven by changes in farm employment. One possible explanation is that the increased resource flows due to M-Pesa are channeled toward farming, thus boosting the demand for labor and increasing employment. Unfortunately, we do not have the data to investigate these underlying mechanisms further. We perform some falsification tests to boost the credibility of our empirical methodology. At the time the 2009 survey was collected, the international money transfer feature of M-Pesa was not yet available. Thus, M-Pesa should have no impact on international money transfers. Rows 14 and 15 of table 7.13 show that we do not find any significant impact of M-Pesa on international transfer patterns. This provides some reassurance that our methodology is not flawed. 7.5 7.5.1

M-Pesa Velocity and the E-Money Loop Velocity

As a measure of how people are using M-Pesa, and also for the purposes of understanding where M-Pesa fits into a broader monetary framework, we are interested in calculating the “velocity” of M-Pesa. In standard monetary economics, “transactions velocity” is defined as the frequency with which the average unit of money is used in transactions. Transactions velocity is different than the more frequently measured income velocity of money, which is simply nominal GDP divided by the relevant money stock. In the case of M-Pesa, the potentially relevant transactions are deposit of money (creation of a unit of M-Pesa), transfer, and withdrawl of money (extinguishing of a unit of M-Pesa). In this respect, M-Pesa differs from cash, which, in a simple monetary system, would circulate in transactions

278

Isaac Mbiti and David N. Weil

with only rare instances in which it is created or liquidated (although a piece of cash may enter and leave the banking system many times over the course of its life). As our measure of M-Pesa velocity, we focus only on transfers, which are the closest analogue to purchases using money in a simple monetary system—indeed, if e-money is eventually used in a money- like fashion, such transfers would play the role of transactions using money. Our measure of M-Pesa velocity is thus the total value of person- to-person transfers (per unit time) divided by the average outstanding balance of e-float. For example, if 100 units of e-float are created at the beginning of the month, transferred from person to person five times in the month, and extinguished at the end the month, then monthly velocity will be five. Notice that having 100 units of e-float transferred from person to person five times in the month could happen either because the people receiving transfers then transferred the e-float to someone else or because each time a transfer was received, the recipient withdrew his cash and a new user deposited cash and received e-float. We discuss this issue (the length of the “e- money loop”) in the next section. Of the two numbers required to measure velocity, the harder one to obtain is the outstanding balance of e-float. As discussed above, all money deposited to create e-float is held by a trust fund that holds deposits in commercial banks. Thus, the outstanding balance of e-float is in principle perfectly observable at any point in time, both to Safaricom and to regulators. This information is not always freely available, however. Weil, Mbiti, and Mwega (2012) were able to obtain monthly data on the size of the trust balance from July 2007 through December 2011.8 The trust balance is by construction identical to the quantity of e-float outstanding. However, in the calculation of transfer velocity it is not clear how to treat e-float that is held by M-Pesa agents. Conceptually, one might want to think of the relevant aggregate for calculating transfer velocity to be e-float held on the phones of customers only. We can construct an estimate of this quantity by subtracting estimated e-cash held on the phones of M-Pesa agents from the trust balance. Eijkman, Kendall, and Mas (2010) report end of day e-float for different types of M-Pesa outlets. These range from 90,000 KSh for rural stores to 40,000 KSh for city stores. Rural stores have particularly high end- of-day float because they engage in primarily cash- out transactions. City stores did more balanced business, though with an excess of cash-in over cash- out. These end- of-day figures do not corre8. An earlier version of this chapter used a different source of data regarding the size of outstanding e-float. Specifically, we used an audit of M-Pesa conducted by the Ministry of Finance in January 2009. That audit states that “whereas the system transacted about 17 billion KShs in August 2008, the net deposite/residual value per customer (i.e., deposit less withdrawals) was KShs 203.” We interpreted the figure of 203 KSh as outstanding e-cash per customer. We now think that this interpretation was incorrect. Similarly, we now think that the value of velocity that we derived based on this measure, which was between eleven and fourteen transactions per month, was incorrect.

Mobile Banking

279

spond to beginning- of-day figures, of course. In our calculations we chose a value of 50,000 KSh per M-Pesa agent. Multiplying this by the number of M-Pesa agents gives our estimate of total e-cash held by M-Pesa agents. From the Safricom website, we have data on the number of agents monthly from April 2007 through April of 2011. For most of the period for which we have data, the fraction of e-cash held by M-Pesa agents has been relatively stable, and there is no discernible trend. For the last year in which we have data, it fluctuates narrowly within the range of 10– 12 percent. The other piece of information required for the calculation of transfer velocity is the monthly value of person- to-person transfers. This is reported by Safaricom for the period April 2007–April 2010. Using these data, figure 7.8 shows our calculated value of transfer velocity monthly. We show velocity both using the full size of the trust balance (labeled unadjusted) and subtracting our estimate of e-cash held by M-Pesa agents. Both series show a significant upward trend. For example, adjusted velocity rises from roughly two transfers per month in the first year of M-Pesa’s operation to roughly four in the last few months for which we have data. The calculated values of velocity seem to indicate that M-Pesa is functioning as a hybrid of a money transfer system, on the one hand, and a means for storing value, on the other. Velocity of four, for example, implies that the average unit of e-cash was transferred once per week. If M-Pesa were purely being used as a money transfer system, we might expect that velocity would be significantly higher. For example, a simple deposit- transfer-

Fig. 7.8

Transfer velocity of e-cash

280

Isaac Mbiti and David N. Weil

withdraw transaction might involve e-cash being created (in the sense that it is transferred from an agent to a customer), transferred, and extinguished (transferred back to an agent’s phone) in much less than a day. This would imply a velocity of over thirty transfers per month. Since we know anecdotally that at least some users indeed do not keep e-cash on their phones for very long, our estimates of velocity imply that some other users are keeping their cash on phones for significantly longer than one week. To give an example, velocity of four would be consistent with thirty users each making one transfer of 1,000 KSh per month where the e-cash existed for only one day, while at the same time nine users made one transfer of 1,000 KSh per month but held on to e-cash for an entire month. Note that in this example, most e-cash at any point in time is held by nonfrequent transactors, even though most transfers are done by frequent transactors. The fact that velocity is trending upward over time suggests that the balance of users within the system is moving in the direction of people who are less inclined to hold e-cash on their phones and more inclined to use the system solely for transfers. This idea can be tested to some extent by looking at the trends in balances per customer and monthly transactions per customer. To construct balances per customer, we use the trust fund balance along with our estimate of e-cash held by M-Pesa agents. Figure 7.9 shows our calculated value. Balances of e-cash per customer are remarkably stable, in the neighborhood of 700 KSh. (Note, however, that there is an interesting decline between September of 2009 and April 2010. Since the latter month is the last for which we can currently calculate velocity, this decline does explain some of our measured rise in velocity.) This average figure represents a distribution of cash balances about which we have no data, although presumably it is highly skewed with most customers at any point in time having balances at or near zero. In the future we hope to get

Fig. 7.9

Average balance per customer

Mobile Banking

Fig. 7.10

281

Value of monthly transfers per customer

data on this distribution. Figure 7.10 shows the value of monthly transfers per customer, which is stable at around 2,500 KSh for the second half of the period in which we have data. 7.5.2

The E-Money Loop

Irving Fisher defined the “cash loop” as the number of transactions that a unit of currency goes through between being withdrawn from a bank and returning to a bank. Analogously, we can think of the “e- money loop” as the number of transfer transactions that the average unit of M-Pesa goes through between being transferred onto a customer phone and being transferred back from a customer phone to the phone of an M-Pesa agent. The length of the e-money loop is not necessarily related to the velocity of e-money. To see this, think about the following two scenarios:

• Scenario 1: Mr. A deposits 100 KSh into M-Pesa on January 1. He transfers the money to Mr. B on January 15. Mr. B withdraws the money on February 1. Also on February 1, Mr. C deposits 100 KSh into M-Pesa, and transfers it to Mr. D. on February 15. Mr. D withdraws the money on March 1. In this case, velocity is one transaction per month, and the length of the e-money loop is also one. • Scenario 2: Mr. A deposits 100 KSh into M-Pesa on January 1. He transfers the money to Mr. B on January 15. Mr. B leaves the e-float on his phone until February 15, at which point he transfers it to Mr. C. The money is repeatedly transferred on the 15th of every month, and never withdrawn. In this case, velocity is one transfer per month while the length of the e-money loop is infinite. As with velocity, we can put together available scraps of information to get an estimate of the length of the e-money loop. An audit of M-Pesa conducted by the Ministry of Finance in January 2009 says that “the system transacted about KShs 17 billion” in August 2008. What does this number

282

Isaac Mbiti and David N. Weil

mean? According to Safaricom, the volume of person- to-person transfers that month was 8.32 billion KSh. It thus seems likely that 17 billion is the volume of cash-in plus cash- out transfers. A different source (Kimenyi and Ndung’u 2009) gives the value of total transactions for August 2008 at just under 15 billion KSh, and in this case the phrase “monthly transactions” is explicitly defined as “deposits plus withdrawals.” Given that one of the authors of that study is chairman of the Central Bank of Kenya, it is likely that the figure is based on the same (nonpublicly available) data as the Ministry of Finance audit. Given the similarity in magnitude and the similar phrasing, we take the 17 billion to be similarly referring to the value of deposits plus withdrawals in the month. Given this observation, what is notable is how close the total of deposits and withdrawals is to twice the value of person- to-person transfers. The relationship between deposits, withdrawals, transfers, and the length of the e-money loop is9 (6)

loop length =

2 × transfers . deposits + withdrawals

Thus the data indicate that the length of the e-money loop is roughly one. This would be true if all transactions took the form of deposit- transferwithdraw. The total for deposits and withdrawals would be less than twice transfers, and the length of the loop greater than one, if there were some appreciable fraction of people who received a transfer and then sent the money on somewhere else without doing a withdrawal. Similarly, the total for deposits plus withdrawals would be more than twice monthly personto-person transfers if an appreciable number of people used their phone to store money without transferring it. Of course, it is possible that there was a good deal of both these activities (receiving money and transferring it onward without taking money out, on the one hand, and depositing and withdrawing without transferring, on the other), but the data are suggestive, at least to us, of the overwhelming majority of use being of the deposittransfer- withdraw type. Using data from Safaricom (for monthly person- to-person transfers) and from Kimenyi and Ndung’u (for monthly deposits to withdrawals) we can calculate the implied length of the e-money loop for the period July 2007– July 2009. This is shown in figure 7.11. It is interesting to note that in the data the e-money loop starts out at slightly less than one before trending up 9. The key assumption required to derive this equation is that the system is in a steady state, where monthly deposits are equal to monthly withdrawals. In this case (deposits + withdrawals)/2 is just equal to the quantity of deposits. Also, in this case, transfers made in a given month would be equal to transfers that would eventually be made with the e-money created in a given month (which in turn would be equal to that month’s deposits). The formula is not fully accurate, since M-Pesa was in fact growing over time. Given information on the rate of growth of M-Pesa and M-Pesa velocity, one could construct a better estimate, but our sense is that it would not differ significantly.

Mobile Banking

Fig. 7.11

283

Length of the e-money loop

to almost exactly one. It is possible that the lower figure represents a different use of M-Pesa in the program’s early days (more cash storage and fewer multiple transfers), but it is also possible that this is some sort of measurement error—recall that the figure for total transfers given in the Ministry of Finance audit was about 7 percent higher than the figure in Kimenyi and Ndung’u (2009). 7.5.3

Implications for Measuring the Money Supply

As M-Pesa and other forms of electronic money have become more prevalent, economists have turned their attention to the implications for measurement of monetary aggregates and the relationship between money, prices, and real variables. To the extent that e-float is a form of money, failure to measure it in monetary aggregates could lead policymakers astray. For example, if the stock of e-float grew while conventional money did not, monetary policy would be looser than policymakers thought. A natural initial approach to this problem would be to simply add the stock of e-money into the measures of, say, M1. This is problematic for two reasons. First, at least in the case of M-Pesa, the existing stock of e-money is backed 100 percent by transactions accounts held at commercial banks. If these accounts are subtracted from M1 while M-Pesa balances are added, the net effect is zero. Second, however, the transactions velocity of e-money may be higher than the transactions velocity of other components of M1, such as cash. Put differently, a small amount of M-Pesa, by circulating fre-

284

Isaac Mbiti and David N. Weil

quently, provides the same transaction (and transfer) services as a much larger quantity of cash. If one had estimates of the transactions velocities of M-Pesa and the other components of a monetary aggregate, it would then be possible to create a velocity- weighted index, in which those components with higher velocity received a higher weight (see Spindt [1985] for a discussion). As shown above, getting a rough approximation of the velocity of M-Pesa is not difficult, and with better data one could get a truly precise estimate. Unfortunately, measuring the velocity of other monetary aggregates—a problem on which monetary economists have been working since the time of Jevons—is much harder. For this reason, and also out of curiosity of how M-Pesa compares to other monies, we have pulled together the few estimates of transaction velocity. The estimates span a number of countries and historical eras and, therefore, pertain to a variety of institutional structures and transaction technologies. This may explain some of the vast variation in the data. A common measure of the velocity of demand deposits is the “demand deposit turnover rate,” defined as the ratio of debits to demand deposits in a period to the average value of demand deposits. In the United States, between 1919 and 1941, the annual turnover rate on demand deposits at commercial banks varied between 19.4 and 53.6 (Board of Governors of the Federal Reserve System 1976). In more recent data, the turnover rate for banks, excluding major New York banks, rose from 135 to 475 per month over the period 1980– 1995 (US Census Bureau 1996). Engberg (1965) presents data on demand deposit turnover in East Africa between 1950– 1963, over which period it rose from 4.1 to 9.9 per quarter. Using data from Cletus (2004), the demand deposit turnover rate in Gambia between 1983 and 1993 varied between two and eleven transactions per month. In Taiwan, in 2007, the annual turnover rate on demand deposits was 328 (Republic of China 2009). In Thailand, monthly demand deposit turnover in 2009 averaged forty- one.10 As far as currency goes, there are even fewer estimates of velocity. Irving Fisher’s calculations for the years around the beginning of the twentieth century in the United States found that transactions velocity of cash was in the neighborhood of twenty per year. Spindt (1985) applies a method suggested by Laurent (1970) to look at the velocity of circulation of currency. His estimate is that the velocity of currency in the United States ranged between seven and ten transactions per month over the period 1970– 1985. A study by the US Federal Reserve based on household surveys (Avery et al. 1986) estimated the velocity of currency in 1984 at between fifty and fifty- five transactions per year. Feige (1987) estimates the length of the cash 10. Bank of Thailand online data query. http:// www2.bot.or.th/statistics/ReportPage .aspx?reportID=31&language=eng.

Mobile Banking

285

loop in the Netherlands at approximately four transactions in data from the 1960s and 1970s. A preliminary, conclusion from this exercise is that the transactions velocity of M-Pesa (roughly four transactions per month) does not stand out as being much higher than that of other monetary components that are held by households, particularly cash. This is at least somewhat surprising, since technologically, it would be possible for M-Pesa to have enormous transactions velocity. In any case, for the present, even if a very large velocity adjustment were appropriate, M-Pesa does not compare with other parts of the monetary aggregate. The average over the period January–June 2008 of currency (M0) was 85.2 billion shillings, while currency plus demand deposits (M1) was 393 billion shillings (Central Bank of Kenya 2008). By contrast, our calculated value of outstanding e-float in August 2008 was 3.3 billion shillings. 7.6

Why Isn’t M-Pesa Used for Storing Value?

Much of the evidence presented in our chapter is strongly suggestive of the conclusion that M-Pesa is only rarely used for storing value for any significant period of time. This can be seen in the low value of average M-Pesa holdings at a point in time (700 KSh, or about ten dollars), the velocity of four transactions per month, and in the short length of the e-money loop. Although a significant fraction of users report that they use their M-Pesa accounts for storing money, such storage is of relatively small amounts of money or for relatively short periods of time. Why don’t people store more value on M-Pesa? One possible reason is that it does not pay interest. If this is the case, then the implementation of M-Kesho or some other scheme to pay interest on transactions could lead to a significant change in behavior. To gather insight into this question, we could ask: At what interest rate would M-Pesa users store significant value on their accounts? Part of the answer to this question can be gleaned by looking at behavior with respect to withdrawals. Although M-Pesa balances do not pay explicit interest, holding money in M-Pesa does yield interest in the form of reducing transaction costs. Consider the problem of an individual who receives periodic transfers into his M-Pesa account. One strategy would be to withdraw each transfer as it is received. An alternative would be to group two or more transfers together and withdraw them all at once. The latter strategy holds money on the M-Pesa account for longer, but involves lower costs. A general analysis of alternative withdrawal strategies would be enormously complex, given the complexity of the price schedule as well as the stochastic nature and varying sizes of transfer receipt. Here, we examine an extremely simple version of the problem to get a feel for the magnitudes involved.

286

Isaac Mbiti and David N. Weil

Consider an individual who receives a transfer of 1,000 KSh on the first of every month. We will allow for only two strategies: first, she can take out the money each time she receives a transfer. Alternatively, she can wait until she has accumulated 2,000 (that is, every other month) and take the money out then. On the M-Pesa price schedule, the price of withdrawing 1,000 KSh is the same as the price of withdrawing 2,000 (i.e., 25 KSh). Let W be the amount withdrawn, and C be the cost. The monthly interest rate r at which an individual would be indifferent between these two strategies is given implicitly by the equation (7)

W −C +

W − C 2W − C = 1+ r 1+ r

where the left- hand side is the present value of withdrawals net of costs using the first strategy and the right- hand side is the same thing using the second strategy. The solution is (8)

r=

C . W −C

For the example just given (W = 1,000; C = 25), the solution is r = 2.6 percent. On an annual basis this is 36 percent—certainly a high interest rate. Using a smaller value of the amount withdrawn, W, would yield a higher implicit interest rate, as would considering an individual who received transfers more frequently than every other month. For example, an individual who received and withdrew 500 KSh every two weeks—a scenario that seems like it might be consistent with what we see in the data—would be demonstrating a discount rate of at least 240 percent per year! The information on the distribution of withdrawal sizes can also be brought to bear here. Although we do not solve the full- scale problem, it is clear that for moderate interest rates there should be a significant amount of bunching of withdrawals at the high end of price ranges—that is, just below the price notch. An individual who withdraws only a little more than the price notch (say, 3,000 KSh when the price notch is at 2,500 KSh) and who is going to be receiving another transfer in the next few months, is paying an enormous price to get his/her money early. And yet the striking observation from our data on the distribution of withdrawal amounts is that there seems to be no bunching at all at the price- notch points. There are also, obviously, a very large number of withdrawals of amounts that are far lower than, say, half of the price notch. Unfortunately, we do not have the data to be completely formal in this analysis. Above, we described the distribution of withdrawal sizes and the frequency of withdrawals, but we do not have these data at the individual level, and so we do not know their joint distribution. We know that most withdrawals are made by individuals who withdraw frequently (every month or more frequently), and that a good fraction of withdrawals are small enough (medians around 1,000 KSh) that two or more of them would fit under the

Mobile Banking

287

2,500 KSh price notch. We also know that there is not a very large mass of withdrawals at the price notch—at least no more than would be expected given the fact that price notches are at round numbers. From this data is seems reasonable to conclude that a significant fraction of withdrawals are made by people who are applying high time discount rates, since otherwise they would be grouping their withdrawals into more economical chunks. We can extend this example further by considering the costs borne by the sender as well. Although we do not have data on senders, it is a reasonable supposition that in many cases a monthly withdrawal of 1,000 KSh is matched by a monthly transfer of the same amount. As discussed above, the fee for transfers of any size is 30 KSh. Thus there would be the possibility that a husband or son working in the city could transfer KSh 2,000 once every two months, instead of KSh 1,000 every month. The total cost of such a transfer and withdrawal of either 1,000 or 2,000 KSh is 55 KSh (30 KSh for transfer and 25 KSh for withdrawal). Plugging this cost figure into the equation above, the implicit interest rate at which a family would be indifferent between transferring 1,000 KSh every month and transferring 2,000 KSh every two months is thus 5.8 percent per month. A final observation that suggests that users of M-Pesa have high financial discount rates comes from a discussion we had with an employee of Kenya Power and Light Corporation, the country’s electricity supplier. Electricity customers receive monthly bills, and must pay them within a fixed time window or their power will be cut off. A bill- pay service was recently established, whereby M-Pesa users could pay their bills through their cell phones, rather than by directly visiting a KPLC office, post office, or bank, all of which involve waiting in a long line. Despite the superior convenience of M-Pesa, the take-up of the service was relatively low; only about 12 percent of the 1.2 million customers paid by M-Pesa, and we were curious as to why. The employee’s theory was that it had to do with the delay involved in paying with M-Pesa. The M-Pesa payments were batch processed overnight and thus required between twenty- four and forty- eight hours to clear, more time than paying in person where the payments were reflected instantly. Therefore the person paying the bill by M-Pesa would have to have the money one or two days earlier than otherwise. Evidently, this extra one or two days was, to most potential users, more valuable than the huge convenience of not having to pay the bill in person. In fact the KPLC employee stated that M-Pesa use for paying electric bills was actually declining due to this lag in processing. This is again suggestive of very high time discount rates. It is important to note that the high financial discount rates that households apply to cash that moves through M-Pesa do not necessarily imply that housholds highly discount the future consumption flows or utility. As in a standard Baumol-Tobin model of cash management, another reason to hold small cash balances is if there is a high cost of holding cash itself. Such a cost could be due to theft in a conventional sense, which can be viewed as a tax on cash balances. However, crime rates would have to be extremely

288

Isaac Mbiti and David N. Weil

high to justify the behavior we see. A more likely cost of holding cash is the high implicit tax represented by the ability of other family members to request either gifts or loans from one’s available cash balances. This is notion is supported by Ashraf (2009), who reports that women in Kenya often form secret saving societies to hide income from their husbands. Finally, and somewhat similarly, holdings of cash may simply raise temptations to spend that individuals find impossible to resist. The inability to save cash holdings has been shown to be a constraint to fertilizer adoption in western Kenya (Duflo, Kremer, and Robinson 2011) and promotes participation in ROSCAS that can act as a commitment saving device (Gugerty 2007). It could be that the extra transaction costs associated with holding small cash balances are a price worth paying to avoid giving in to these temptations. These observations might be taken to suggest that the types of interest rates potentially offered through cell phone banking will do little to alter the amount of money that people store on their phones. However, recent literature on the financial lives of the poor may suggest otherwise. Collins et al. (2009) find that the world’s poor utilize a vast range of financial instruments to meet different needs, and prioritize different qualities of these instruments based on how they use them. For instruments used to smooth day- to-day consumption, they find that it is most important to the poor that these keep their money secure and easily accessible, but pay little attention to the interest they might earn. However, when the poor seek to accumulate what Collins and colleagues refer to as “usefully large sums” to pay for life- cycle events (such as weddings or funerals) or other larger expenses, they do take into account the interest that different financial tools can offer them, along with their security, reliability, and structure (for example, requiring them to make periodic deposits to help ensure that they will succeed in building up a larger sum of money). Based on this, we might say that M-Pesa has found a niche in the former realm of day- to-day cash management, but not as much in the accumulation of larger sums. The introduction of a program encouraging saving and offering interest might allow mobile banking to find an additional niche as a simple and secure financial tool for the accumulation of usefully large sums. One survey found that 38 percent of respondents said that the feature they would most like to see added to M-Pesa was the ability to earn interest on their accounts, making this the most popular response (Jack, Pulver, and Suri 2009). This suggests that interest will be an attractive feature of M-Kesho. (As discussed above, M-Kesho will also offer insurance and microloans, which may also be attractive features.) 7.7

Conclusion

In this chapter we have examined M-Pesa from a number of different perspectives. Using firm- level data from competing money transfer services we find that the introduction of M-Pesa has led to significant decreases in

Mobile Banking

289

the prices of competitors. In addition, we examine microlevel data from the FinAccess surveys, where we find that frequent M-Pesa users are more likely to be urban, educated, banked, and affluent. Our analysis of the 2006 and 2009 rounds of the FinAccess surveys reveal that M-Pesa use increases frequency of sending transfers, decreases the use of informal saving mechanisms such as ROSCAS, and increases the probability of being banked. This suggests that M-Pesa is complementary to banks, whereby the adoption of M-Pesa has increased the demand for banking products. Although a significant number of survey respondents indicate that they use their M-Pesa accounts as a vehicle for saving, our analysis of aggregate data suggests that the overwhelming use of M-Pesa is for transferring money from individual to individual, with extremely little storage of value. This can be seen in many ways. Our estimates of M-Pesa velocity, the number of transactions per month for the typical unit of e-float, is roughly four transactions per month, depending on some auxiliary assumptions. We also estimate the length of the “e- money loop,” that is, the average number of person- to-person transactions that take place between the creation and destruction of a unit of e-float. Our estimate is quite near one. Although we cannot be certain, we take this as evidence that the vast majority of M-Pesa use is of the form of a cash deposit, followed by a single person- to-person transfer of e-float, followed by a cash withdrawal. Our analysis of data on the size and frequency of M-Pesa withdrawals also suggests that M-Pesa users have relatively high opportunity costs of holding funds on their phones. For example, there seems to be little evidence of users bunching several transfer receipts together into a single withdrawal in order to economize on fees. This suggests that even if M-Pesa were to pay interest at the same rate as banks, there would not be a significant change in the saving behavior of users.

Data Appendix FinAccess Surveys The FinAccess surveys, conducted in 2006 and 2009, are nationally representative household surveys that were designed to measure financial access in Kenya. The surveys were collected by Financial Sector Deepening Trust Kenya (FSD Kenya), with financial and technical support from a variety of partners including the Central Bank of Kenya, donors, and a number of commercial banks in Kenya.The 2006 round consisted of approximately 4,400 individuals, while the 2009 round consisted of close to 6,600 individuals. A unique feature of this data is that it aimed to capture access to a wide range of both formal and informal financial tools. Moreover, the consis-

290

Isaac Mbiti and David N. Weil

tency of the surveys enable reliable comparisons across time of the changing nature of financial access. Using sampling weights we can aggregate the data to the sublocation level. Sublocations are the lowest administrative unit in Kenya and consist of two to three villages in rural areas or a large neighborhood in a city. We combine the 2006 and 2009 FinAccess surveys and create a balanced panel of the 190 sublocations that were surveyed in both rounds. We constructed the measure of transfer frequency as follows. We converted the categorical responses into annual numerical values as follows: Daily = 365 times a year, weekly = fifty- two times a year, monthly = twelve times a year, irregularly/once in a while = once a year. We then used these conversion factors to change categorical responses on transfer sent and received frequencies, as well as M-Pesa use frequencies, into annualized numerical values. Our wealth measure is constructed by using principal component analysis on the household assets and durable goods such as televisions and refrigerators. We then create wealth quantile dummies based on the principal component analysis. Transfer Prices Kabbucho, Sander, and Mukwana (2003) document the prices of various money transfer methods. We use their data from 2003 as the baseline and compare it to current (2010) prices of Moneygram and Western Union. The Moneygram fee schedule is documented online, while the Western Union rates were collected in person by research staff. These fees are converted into a database that contains prices for a series of transfer amounts in 100 KShs intervals. This allows us to compare the prices across comparable set of prices. M-Pesa Transaction Data We collect M-Pesa transaction data from an agent in Kisumu. These data contained transaction type and transaction amount over a three- month period in 2010 for three M-Pesa shops. Katito is in a rural area, Homa Bay is in a small town, and Cyber is in an urban environment. Further details of these stores can be obtained from Eijkman, Kendall, and Mas (2010).

References Aghion, Philippe, and Peter Howitt. 1992. “A Model of Growth through Creative Destruction.” Econometrica 60 (2): 323– 51.

Mobile Banking

291

Aker, Jenny C. 2010. “Information from Markets Near and Far: The Impact of Mobile Phones on Agricultural Markets in Niger.” American Economic Journal: Applied Economics 2 (July): 46– 59. Aker, Jenny, and Isaac Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic Perspectives 24 (3): 207– 32. Ashraf, Nava. 2009. “Spousal Control and Intra-Household Decision Making: An Experimental Study in the Philippines.” American Economic Review 99 (4): 1245– 77. Avery, Robert B., Gregory E. Elliehausen, Arthur B. Kennickell, and Paul A. Spindt. 1986. “The Use of Cash and Transaction Accounts by American Families.” Federal Reserve Bulletin February:87– 108. Beck, Thorsten, Asli Demirgüç-Kunt, and Soledad Martinez Peria. 2007. “Reaching Out: Access to and Use of Banking Services across Countries.” Journal of Financial Economics 85 (1): 234– 66. Board of Governors of the Federal Reserve System. 1976. “Banking and Monetary Statistics 1941– 1970.” Section 5, Table 55, September. http://fraser.stlouisfed.org /publications/bms2/issue/60/download/106/section5.pdf. Bower, Joseph L., and Clayton M. Christensen. 1995. “Disruptive Technologies: Catching the Wave.” Harvard Business Review January–February. https://hbr.org /1995/01/disruptive- technologies- catching- the- wave/ar/1. Burgess, Robin, and Rohini Pande. 2005. “Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment.” American Economic Review 95 (3): 780– 95. Central Bank of Kenya. 2008. “Statistical Bulletin, June 2008.” http://www2.central bank.go.ke/downloads/statistics/bulletin/Junsb08.pdf. Cletus, Agu. 2004. “Efficiency of Commerical Banking in the Gambia.” African Review of Money Finance and Banking 2004:31– 50. Collins, Daryl, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven. 2009. Portfolios of the Poor: How the World’s Poor Live on $2 a Day. Princeton, NJ: Princeton University Press. Duflo, Esther, Michael Kremer, and Jonathan Robinson. 2011. “Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya.” American Economic Review 101 (6): 2350– 90. Eijkman, Frederik, Jake Kendall, and Ignacio Mas. 2010. “Bridges to Cash: The Retail End of M-PESA.” Savings and Development 34 (2). 219– 52. Engberg, Holger L. 1965. “Commercial Banking in East Africa, 1950– 1963.” Journal of Modern African Studies 3 (2): 175– 200. Feige, Edgard L. 1987. “The Theory and Measurement of Cash Payments: A Case Study of the Netherlands.” In The Practice of Econometrics: Studies on Demand, Forcasting, Money and Income International Studies in Economics and Econometrics, vol. 15, edited by R. D. H. Heijmans and N. Neudecker. Dordrecht, the Netherlands: Kluwer Academic Publishers. FinMark Trust. 2008. Finscope In Africa, 2008. http://www.finscope.co.za/documents /2008/FSAfricaBrochure08.pdf. Gikunju, Washington. 2009. “Mobile Money Transfers Edge out Rival Operators.” Business Daily, September 14. http://www.businessdailyafrica.com/-/539552/657466 /-/view/printVersion/-/2gufmtz/-/index.html. Gugerty, Mary Kay. 2007. “You Can’t Save Alone: Commitment and Rotating Savings and Credit Associations in Kenya.” Economic Development and Cultural Change 55:251– 82. Jack, William, Caroline Pulver, and Tavneet Suri. 2009. “The Performance and Impact of M-Pesa: Preliminary Evidence from a Household Survey.” Power Point

292

Isaac Mbiti and David N. Weil

presentation. http://technology.cgap.org/technologyblog/wp- content/uploads /2009/10/fsd_june2009_caroline_pulver.pdf. Jack, William, and Tavneet Suri. 2010. “The Risk Sharing Benefits of Mobile Money.” Working Paper, MIT. ———. 2011. “Mobile Money: The Economics of M-Pesa.” NBER Working Paper no. 16721, Cambridge, MA. Jensen, Robert T. 2007. “The Digital Provide: Information (Technology), Market Performance and Welfare in the South Indian Fisheries Sector.” Quarterly Journal of Economics 122 (3): 879−924. Kabbucho, Kamau, Cerstin Sander, and Peter Mukwana. 2003. “Passing the Buck— Money Transfer Systems: The Practice and Potential for Products in Kenya.” MicroSave Africa Report. http://microfinancegateway.org/content/article/detail /19594?PHPSESSID=332fab3a7849fc635883a38e113c62da. Kimenyi, Mwangi S., and Njuguna S. Ndung’u. 2009. “Expanding the Financial Services Frontier: Lessons from Mobile Phone Banking in Kenya.” Brookings Institution, October 16. http://www.brookings.edu/articles/2009/1016_mobile _phone_kimenyi.aspx. Laurent, Robert D. 1970. “Currency Transfers by Denomination.” PhD diss., University of Chicago. Mas, Ignacio, and Olga Morawczynski. 2009. “Designing Mobile Transfer Services: Lessons from M-Pesa.” Innovations 2009 (Spring): 77– 91. Mas, Ignacio, and Amolo Ng’weno. 2009. “Three Keys to M-Pesa’s Success: Branding, Channel Management and Pricing.” Unpublished Manuscript, Bill and Melinda Gates Foundation, December. Ministry of Finance of Kenya. 2009. “Ministry of Finance Audit Findings on M-Pesa Money Transfer Services.” January 26. http://kenyapolitical.blogspot .com/2009/01/ministry- of-finance- audit- findings- on- m.html. Morawczynski, Olga. 2009. “Exploring the Usage and Impact of ‘Transformational’ Mobile Financial Services: The Case of M-PESA in Kenya.” Journal of Eastern African Studies 3 (3): 509– 25. Morawczynski, Olga, and Mark Pickens. 2009. “Poor People Using Mobile Financial Services: Observations on Customer Usage and Impact from M-PESA.” CGAP Brief Online. http://www.cgap.org/gm/document- 1.9.36723/BR_Poor _People_Using_Mobile_Financial_Services.pdf. Njiraini, John, and James Anyanzwa. 2008. “Unmasking the Storm Behing M-Pesa.” East African Standard, December. http://www.standardmedia.co.ke/business /article/1144002826/unmasking- the- storm- behind- m- pesa?articleID=114400 2826&story_title=unmasking- the- storm- behind- m- pesa&pageNo=2. Pickens, Mark, David Porteous, and Sarah Rotman. 2009. “Scenarios for Branchless Banking in 2020.” Focus Note no. 57, Consultative Group to Assist the Poor (CGAP), Washington, DC. Plyler, Megan, Sherri Hass, and Geetha Nagarajan. 2010. “Community-Level Economic Effects of M-PESA in Kenya: Initial Findings.” IRIS Center Report, University of Maryland. http://www.fsassessment.umd.edu/publications/pdfs /Community-Effects-MPESA-Kenya.pdf. Republic of China. 2009. “Monthly Bulletin of Statistics.” August. http://eng.dgbas .gov.tw/public/data/dgbas03/bs7/bulletin_eng/PDF/eng- month9808.pdf. Safaricom. 2007. “M-Pesa Update.” Press Release, December 7. ———. 2009. “Industry Update, March 12. http://www.safaricom.co.ke/fileadmin / template/ main/ downloads/ investor_relations_pdf/ Industry%20Update%20 120309.pdf.

Mobile Banking

293

Schumpeter, Joseph A. 1942. Capitalism, Socialism, and Democracy. New York: Harper and Brothers. Slemrod, Joel. 2010. “Buenos Notches: Lines and Notches in Tax System Design.” Unpublished Manuscript, University of Michigan. Spindt, Paul A. 1985. “Money Is What Money Does: Monetary Aggregation and the Equation of Exchange.” Journal of Political Economy 93 (1): 175– 204. US Census Bureau. 1996. “Statistical Abstract, 1996, Section 16, Banking, Finance, and Insurance.” http://www.census.gov/prod/2/gen/96statab/finance.pdf. Vaughan, Pauline. 2007. “Early Lessons from the Deployment of M-PESA, Vodaphones’s Own Mobile Transactions Service.” In The Transformational Potential of M-Transactions, Vodaphone Policy Paper Series no.6. http://www.vodaphone .com/m- transactions. Weil, David N., Isaac Mbiti, and Francis Mwega. 2012. “The Implications of Innovation in the Financial Sector on the Conduct of Monetary Policy in East Africa.” Report submitted to the Interanational Growth Centre Tanzania Country Programme.

8

African Export Successes Surprises, Stylized Facts, and Explanations William Easterly and Ariell Reshef

Over the last fifteen years or so the world has experienced a sharp increase in export activity (until the recent financial crisis). Africa has not lagged in this respect. From 1994 to 2008 exports of goods per capita from subSaharan Africa have increased more than fourfold, or 13 percent per year, on average.1 This is compared with 4 percent for the United States, 8 percent for Germany, 13 percent for India, and 19 percent for China.2 Given the well-known difficulties in exporting from Africa (let alone running business there), 13 percent annual growth rates of exports per capita are no small feat.3 This motivates a closer examination of the patterns and determinants of African export success. Broadly speaking, the composition of exports from sub-Saharan Africa has remained relatively constant over time, with a relatively low share of William Easterly is professor of economics at New York University and a research associate of the National Bureau of Economic Research. Ariell Reshef was assistant professor of economics at the University of Virginia at the time this research was done. He is currently a research director at the CNRS and an associate member of the Paris School of Economics. We wish to thank Nathan Nunn and participants of the NBER African Development Successes conference in Accra, July 18–20, for excellent comments on a previous draft. We are grateful for funding from the NBER Africa Project. We thank Shushanik Hakobyan for excellent research assistance. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters /c13365.ack. 1. We do not address destinations in this chapter. However, it is worth noting that much of the African export growth is regional. The share of exports from the average sub-Sahara African country to destinations outside sub-Saharan Africa has steadily declined from 71 percent in 1994 to 53 percent in 2008. 2. Data from World Bank World Development Indicators database. 3. The World Bank Doing Business database exhibits glaring differences in the ease of export activity between African countries and the United States and Germany, but also versus India and China in almost every measured dimension.

297

298

William Easterly and Ariell Reshef

manufacturing exports and high shares of all other export categories (agriculture, food, fuel, and ores and metals). However, on a closer, detailed examination of export activity, it becomes evident that these broad categories mask much heterogeneity. For example, not all agricultural exports are the same. From our examination of export activities in three East African countries that we visited, we witness price variation based on quality differentiation of products that are exported, sometimes exclusively to Europe and the United States, other times regionally. We also witness quality upgrading and attempts to capture larger proportions of the value chain. In this chapter we demonstrate that several common views about exporting activity from Africa are not accurate at best, and in some cases simply wrong. Perhaps exaggerating a bit the traditional view held for many years, Africa is seen primarily as a commodity exporter, and commodity exports are perceived not as “prestigious” as other exports (such as manufacturing) because commodity revenues are thought to reflect mainly endowments and world prices rather than domestic success. These views in turn predict that (a) the distribution of export revenue (“success”) across goods will be different in Africa relative to the rest of the world, (b) that African nations simply export a fixed set of low-value-added goods that correspond to their individual commodity endowments, and as such, (c) that revenues by good are determined by world prices. While many researchers probably now already have a more subtle view of African exports, we think that the polar extreme of this traditional view still has some influence in academic and policy circles, and hence is worth testing as a set of hypotheses. One sign of the influence of this traditional view is the large amount of policy discussion and research literature about commodity “booms” and “busts,” which are defined by large increases or decreases in world commodity prices. In fact, we largely reject these hypotheses. We demonstrate that in many dimensions African exporting is in line with the rest of the world. As in the rest of the world, export success is dominated by a small number of Big Hits. On average, Big Hits are no more and no less as rare in Africa as anywhere else: they follow a power law that is broadly similar in Africa to outside Africa. Moreover, it is not correct that worldwide commodity price movements determine export revenues in Africa. Nor is there much of a difference in the role of world prices between commodities and manufactured products. The Big Hits change by a surprising amount from one period to the next, but the changes are not driven by global prices. In order to establish these stylized facts we use detailed HS4 product-level data from the Comtrade database. However, Comtrade gave us concerns as well as great access to detail. We noticed signs of significant measurement error problems, to which we devote a whole section below. In the worst case scenario, some of our results could be driven by measurement error. Other results are less sensitive to measurement error because they compare results across groups of countries or products, and there is usually no a priori rea-

African Export Successes

299

son to expect measurement error to be systematically different between these different groups (although, of course, this possibility still remains). With this new and better understanding of the statistical data, we traveled to a set of East African countries and interviewed several exporting entrepreneurs in booming export industries, as well as government officials and non-governmental organization (NGO) personnel, with one broad question in mind: What are the determinants of export success in Africa? We have come up with a set of answers that, in many respects, would not be very different from what we might expect to find elsewhere. African exporting entrepreneurs perform very similar activities to those that exporters are expected to do anywhere else. This is in line with Tybout (2000), who concludes that manufacturing firms (not only exporters) in developing countries are not inefficient relative to their counterparts elsewhere. If there are differences, they are driven by low incomes in target markets, detrimental macro policies, high transportation costs, bureaucracy, and poor rule of law. It is comforting that our stylized facts are consistent with findings from our interviews. Given our interviews, we classify determinants of success into two broad categories: conventional and idiosyncratic. We document the following conventional determinants: moving up the quality ladder, utilizing strong cases of comparative advantage, responding to trade liberalization, investing in technological upgrades, foreign ownership, exploiting ethnic networks, and relying on personal foreign experience of the entrepreneur. Some determinants are idiosyncratic in nature: Rwanda’s coffee quality upgrade was a foreign aid success despite the usual poor record of aid, and a Rwanda handicraft export success defied the long odds that cause most handicraft projects to fail. Other idiosyncratic features include sheer passion of the entrepreneur (Uganda roasted coffee), luck (Nile perch from Lake Victoria), and cost shocks (rising aviation fuel costs killed off cut-flower exports from Uganda). We find that there is a role for international aid organizations in bridging the gaps between Africa and markets in the West, but that only careful implementation of aid in partnership with local producers (or farmers) and exporters works well in raising exports. This is in line with Roberts and Tybout (1997), who argue that due to informational externalities, as well as externalities that arise from more efficient delivery of supporting services to exporters, there is a role for public-sector intervention.4 Although far from being the majority, some of the exporters we interviewed cite the importance of government support in accessing trade fairs, or complain about lack thereof. Our chapter also corroborates the conclusions of Artopoulos, Friel, and 4. See also Rauch and Watson (2003) for another example of how informational asymmetries shape the relationship between buyers in developed countries and suppliers in developing countries.

300

William Easterly and Ariell Reshef

Hallak (2010), which find that successful exporters in Argentina who either pioneer a new industry or participate in a new booming one have a particular mind-set, exposure to the world, and apply the correct “export business model” (as opposed to a domestic mode of operation). We find this to be true in several of our export success stories, and in particular in the cases in which entrepreneurs are pioneers. Less surprisingly, important factors contributing to export success are regional free trade zone agreements and low duties for imports into Europe. Tybout (2000) reports overall efficiency improvements due to removal of trade barriers, but not productivity gains at the plant level. Our interviews tend to corroborate this in a qualitative way. Although not the focus of the chapter, we contribute to the international trade literature more broadly, by providing several case studies on export decisions. One of the most important questions in the trade literature is whether exporting improves productivity or if exporters are simply selecting into exporting based on existing productivity (see Clerides, Lach, and Tybout 1998; Bernard and Jensen 1999; Melitz 2003). In our interviews we observe deliberate export decisions that are taken together with specific investments, but also unintentional exporting entry that happens by chance. Therefore, both views may coexist in reality. Another regularity that we have observed is that quality matters.5 Practically every exporter that we interviewed told us this, invariably in the beginning of our interview.6 This is particularly true when exporting to rich markets (European Union and the United States). However, when exporting regionally, and given the relatively low incomes of consumers in Africa, cost seems to matter, in some cases more than quality. There seems to be a trade-off between cost and quality, and when incomes are low, costs trump quality. Hence, which model is right depends on context. The rest of the chapter is organized as follows. We first document that export success is dominated by a few “Big Hits,” both in Africa and elsewhere. The value of the Hits approximately follows a power law. Next, we document that the Big Hits do not remain the same: the successful goods change a lot from one period to the next (again both in Africa and elsewhere). We then explore whether this export success instability is explained by world prices, and find that they play only a small role. We note that measurement error may be contributing to the measured instability of export values by good, although we find that aggregation alleviates the problem. The stylized facts that we establish do not match that traditional view that sees African 5. See Baldwin and Harrigan (2011). 6. The quality question was not the first we asked, though. After explaining who we are, we started each interview by stating our research question and then allowing the entrepreneur to start talking freely about her business. Almost invariably it was at that stage that quality came up.

African Export Successes

301

commodity exports as a passive endowment, with changes driven mostly by global commodity prices. In the final section, we explore pathways to Big Hits with a series of case studies.7 8.1

Success is Rare and Dominated by a Few Big Hits

Success in exporting is rare, but it can be very big. This is manifested in the data by concentration of export revenue on a small number of Big Hits. An easy way to summarize this rareness of export success and the relative size of Big Hits is the following statement: African exports approximately follow a power law—the top-ranked exports are vastly larger than lower-ranked exports. We calculate the average export share of the top-ranked export product for all thirty-seven African countries for which we have data, then the second, down to the twentieth product—all at the four-digit Harmonized System (HS) code level. The results are reported in table 8.1. Figures 8.1A and 8.1B display the power law graphically. Hausmann and Rodrik (2006) had previously pointed out the phenomenon of hyperspecialization, although only for a few countries and products. In contrast, the scope of our work is comprehensive. We also make a very significant addition to the Hausmann and Rodrik findings, in that we characterize the probability of “Big Hits” as a function of the size of the hit by a power law. In Easterly and Reshef (2009) we document and analyze this phenomenon more deeply for a broad international sample. Besedes and Prusa (2007) make a complementary point to ours. They find that most new trade relationships fail within two years and that the hazard rate of such failure is higher for developing countries. Nevertheless, developing countries have the highest increase in trade relationships: there seems to be a lot of attempts in discovery as it is.8 Table 8.1 also shows how the rankings are affected by excluding extractables (oil and minerals) and commodities. Then the table compares the pattern of African “Big Hits” to that for non-African countries (all other countries in the world). In addition, the table shows in the last line the coefficient to the approximate power law, which is calculated by regressing log rank on log export share for the top twenty products in each column. The common perception of African countries as undiversified, monoexporters is partially confirmed by this data. Concentration levels at the top of the distribution are somewhat higher than those in other countries. However, the comparison shows that both African and non-African exports have the 7. Summaries of all interviews are available in an online appendix on the authors’ web pages. 8. Bernard et al. (2007) document concentration across US exporting firms, while Eaton et al. (2007) find that Colombian exports are dominated by a small number of very large exporters. Arkolakis and Muendler (2008) make a similar point for Brazilian and Chilean exporting firms and also use a power law to approximate the distribution of exports.

302 Table 8.1

William Easterly and Ariell Reshef Average shares of top twenty goods for all countries in group shown Export shares, average of 37 African countries

Export shares, average of 130 non-African countries

All goods (%)

Excluding extractables (%)

Excluding extractables and commodities (%)

All goods (%)

Excluding extractables (%)

Excluding extractables and commodities (%)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

47.6 13.7 7.8 4.1 2.9 2.3 1.9 1.5 1.3 1.1 1.0 0.9 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4

42.6 15.5 7.5 4.6 3.2 2.7 2.1 1.7 1.5 1.3 1.1 1.0 0.9 0.8 0.7 0.6 0.6 0.5 0.5 0.5

34.9 14.0 7.4 5.2 4.0 3.0 2.5 2.1 1.8 1.5 1.4 1.2 1.1 1.0 0.9 0.9 0.8 0.7 0.7 0.6

27.5 11.6 6.3 4.5 3.6 2.7 2.2 1.9 1.7 1.5 1.3 1.2 1.0 0.9 0.9 0.8 0.8 0.7 0.7 0.6

21.4 10.5 6.7 4.8 3.8 2.9 2.4 2.1 1.8 1.6 1.5 1.3 1.2 1.1 1.0 0.9 0.9 0.8 0.8 0.8

20.7 10.6 6.5 4.8 3.6 2.9 2.5 2.1 1.9 1.6 1.5 1.3 1.2 1.1 1.0 0.9 0.9 0.8 0.8 0.8

Power law coefficient

–0.64

–0.67

–0.75

–0.79

–0.87

–0.88

Export rank of good

same tendency of very fat-tailed distributions and Big Hits (which in the tail is approximately a power law). Africa can then be seen as simply having a somewhat more extreme power law, rather than being completely unique in having high concentration of the top exports. The surprising and interesting point is that the top twenty export products are ranked on an almost perfect straight line (in logs), which shows both the rareness of Big Hits, as well as their relative size. The fact that there is a linear relationship between rank and export share in logs shows two things. First, the distribution of exports exhibits fat tails: although Big Hits are uncommon, they would be drastically smaller in a normal distribution of export values across goods. The power law also implies a fatter tail than another common fat-tailed distribution, a log-normal distribution. Second, the probability of observing a Big Hit of size x declines exponentially with

Fig. 8.1A

Power laws, all goods

Note: Horizontal axis is log base 10 of average export share corresponding to each rank. Vertical axis is log base 10 of rank from one to twenty.

Fig. 8.1B

Power laws, excluding extractables and commodities

Note: Horizontal axis is log base 10 of average export share corresponding to each rank. Vertical axis is log base 10 of rank from one to twenty.

304

William Easterly and Ariell Reshef

the expected size of the hit. In other words, the probability of observing a hit of size x is proportional to x-p, where p is the power coefficient.9 8.2

The Big Hits Change from One Period to the Next

If Big Hits were there to stay forever, then this would simplify the discussion. It would follow that some countries are simply better at producing some products and they export those for which they have a comparative advantage. In other words, the simple static Ricardian model is a good description of the world. However, as we demonstrate here, Big Hits do not remain so big relative to other Big Hits for long. The composition of export Big Hits changes quite a bit over time. Tables 8.2A, 8.2B, 8.2C, and 8.2D demonstrate this phenomenon for selected countries. In those tables we report the value of the top ten exports (at the HS four-digit level) in the start year and in the end year and compare their ranks in one year versus the other. It is evident that there is much churning of Big Hits. Tables 8.3A and 8.3B make this argument more formal and general. In table 8.3A we report the low correlation of the ranks of the export basket in a start year with that in the end year of a sample of focus countries. In panel A of table 8.3B, we show that these results are not far from the average African country. Restricting to nonextractables and noncommodities does not change things materially. In addition, panel B suggests that the results are not dissimilar for non-African countries. Although the rank correlation over all goods is higher for non-African countries, restricting to the top fifty or one hundred goods brings Africa in line with non-African countries. The top fifty products account for over 80 percent of export value on average, so the similarities at the top of the distribution are also those that matter most. Thus, the phenomenon of churning Big Hits is not unique to Africa, and is in fact similar quantitatively to non-African countries. Table 8.4 shows the changing nature of success in another way. We decompose export growth (∆X) into intensive export growth in products that are exported both in the start and end year (∆B), new products that are not exported in the first year (N), and lost products that exported in the first year but not in the end year (L): DX = X (t ) − X (t − 1) = B (t ) − B (t − 1) + N (t ) − L (t − 1) = DB + N (t ) − L (t − 1). 9. The power coefficients are estimated at less than unity in absolute value. As is well known, when the power coefficient is less than unity, the implied theoretical Pareto distribution does not have finite moments. This is mostly a concern for the structural empirical trade literature, which relies on productivity distributions that are Pareto. Arkolakis (2008) deals with this by adding marketing costs, while Eaton, Kortum, and Kramarz (2008) add to their model demand and entry shocks.

1 2 3 4 5 6 7 8 9 10 12 14 20 67

277,491 220,638 141,285 111,920 84,100 62,896 31,759 29,809 19,676 15,568 13,848 709

Rank

976,646 965,567

1996 Export value ($) Gold, unwrought, semimanufactured, powder form Cocoa beans, whole or broken, raw or roasted Wood sawn, chipped lengthwise, sliced or peeled Coconuts, Brazil nuts, and cashew nuts, fresh or dried Plywood, veneered panels, and similar laminated wood Veneers and sheets for plywood etc. < 6 mm thick Oil seeds and oleaginous fruits nes Oils, petroleum, bituminous, distillates, except crude Manganese ores, concentrates, iron ores > 20 percent Manganese Cocoa butter, fat, oil Natural rubber and gums, in primary form, plates, etc. Prepared or preserved fish, fish eggs, caviar Aluminum ores and concentrates Unwrought aluminum

Description

Ghana top ten exports in real value in 1906 and 2008 (thousand US$)

Cocoa beans, whole or broken, raw or roasted Gold, unwrought, semimanufactured, powder form Wood sawn, chipped lengthwise, sliced or peeled Veneers and sheets for plywood etc. < 6 mm thick Cocoa butter, fat, oil Unwrought aluminum Oils, petroleum, bituminous, distillates, except crude Prepared or preserved fish, fish eggs, caviar Natural rubber and gums, in primary form, plates, etc. Aluminum ores and concentrates Manganese ores, concentrates, iron ores > 20 percent manganese Oil seeds and oleaginous fruits nes Plywood, veneered panels, and similar laminated wood Coconuts, Brazil nuts, and cashew nuts, fresh or dried

Description

Table 8.2A

9 10 14 16 18 132

49,852 47,549 23,735 20,904 17,937 447

7 8

6

58,597 55,475 53,047

5

2 3 4

1,031,154 131,018 119,222 75,002

1

Rank

1,814,192

2008 Export value ($)

173,663 64,935 44,925 31,675 24,839 21,492 19,830 16,311 13,226 12,148 5,413 1,794 174 2

2 3 4 5 6 7 8 9 10 12 20 54 182

Value ($)

1

Rank

2001

Coffee, coffee husks and skins, and coffee substitutes Oil seeds and oleaginous fruits nes Vegetables, leguminous dried, shelled Cut flowers, dried flowers for bouquets, etc. Vegetables nes, fresh or chilled Gold, unwrought, semimanufactured, powder form Vegetable products, nes Raw skins of sheep or lambs Sheep or lamb skin leather, without wool on Meat of sheep or goats, fresh, chilled, or frozen Goat or kid skin leather, without hair Niobium tantalum vanadium zirconium ores, concentrate Raw hides and skins of bovine, equine animals Buckwheat, millet, and canary seed, other cereals

Description

Ethiopia top ten exports in real value in 2001 and 2008 (thousand US$)

Coffee, coffee husks and skins, and coffee substitutes Vegetable products, nes Oil seeds and oleaginous fruits nes Sheep or lamb skin leather, without wool on Vegetables, legumous dried, shelled Buckwheat, millet, and canary seed, other cereals Goat or kid skin leather, without hair Raw hides and skins of bovine, equine animals Raw skins of sheep or lambs Niobium tantalum vanadium zirconium ores, concentrate Gold, unwrought, semimanufactured, powder form Meat of sheep or goats, fresh, chilled, or frozen Cut flowers, dried flowers for bouquets, etc. Vegetables nes, fresh or chilled

Description

Table 8.2B

250,275 116,997 104,740 83,462 79,898 30,526 30,053 28,816 28,054 22,721 857 338 39

101 229

562,263

Value ($)

2 3 4 5 6 7 8 9 10 12 58

1

Rank

2008

1 2 3 4 5 6 7 8 9 10 12 35 50 126 149 250 397

720 413 81 59 16 2

Rank

487,662 39,211 36,160 26,199 19, 426 13,384 12,378 11,432 8,506 3,970 3,171

Value ($)

1995

Coffee, coffee husks and skins, and coffee substitutes Fish fillets, fish meat, mince except liver, roe Cement, (portland, aluminous, slag, or hydraulic) Tobacco unmanufactured, tobacco refuse Documents of title (bonds, etc.) unused stamps, etc. Tea Live plants nes, roots, cuttings, mushroom spawn Tube, pipe of iron or steel, except seamless > 406.4 m Flat-rolled iron/steel, > 600 mm, clad, plated, or coated Animal and vegetable fats or oils, hydrogenated only Soaps Oil seeds and oleaginous fruits nes Vegetables, leguminous dried, shelled Gold, unwrought, semimanufactured, powder form Maize (corn) Raw hides and skins of bovine, equine animals Cotton, not carded or combed

Description

Uganda top ten exports in 1995 and 2008 (thousand 2008 US$)

Coffee, coffee husks and skins, and coffee substitutes Fish fillets, fish meat, mince except liver, roe Gold, unwrought, semimanufactured, powder form Maize (corn) Vegetables, leguminous dried, shelled Raw hides and skins of bovine, equine animals Tobacco unmanufactured, tobacco refuse Cotton, not carded or combed Oil seeds and oleaginous fruits nes Soaps Flat-rolled iron/steel, > 600 mm, clad, plated, or coated Tea Live plants nes, roots, cuttings, mushroom spawn Cement (portland, aluminous, slag, or hydraulic) Animal and vegetable fats or oils, hydrogenated only Documents of title (bonds, etc.), unused stamps, etc. Tube, pipe of iron or steel, except seamless > 406.4 m

Description

Table 8.2C

1 2 3 4 5 6 7 8 9 10 13 15 18 24 29 81 154

25,888 20,584 15,796 13,569 8,439 6,256 1,161 268

Rank 403,138 107,942 77,644 66,216 55,739 47,209 42,470 37,080 32,125

Value ($)

2008

135,279 97,329 59,398 54,617 49,832 31,190 30,160 26,980 25,874 18,354 12,113 4,103 —

($)

1998

1 2 3 4 5 6 7 8 9 10 13 27 676

Rank

Tanzania top ten, 1998 and 2007 (real thousand US$)

Coconuts, Brazil nuts, and cashews Coffee, coffee husks and skins Fish fillets, fish meat, mince Tobacco unmanufactured, tobacco Gold, unwrought, semimanufactured Tea Cotton, not carded or combed Cloves (whole fruit, cloves and stems) Diamonds, not mounted or set Mounted precious or semiprecious Vegetables, leguminous dried Wheat or meslin flour Precious metal ores and conc.

Description

Table 8.2D

Gold, unwrought, semimanufactured Precious metal ores and concentrates Fish fillets, fish meat, mince except Coffee, coffee husks and skins Tobacco unmanufactured, tobacco Vegetables, leguminous dried, shelled Mounted precious or semiprecious Wheat or meslin flour Cotton, not carded or combed Tea Coconuts, Brazil nuts, and cashews Diamonds, not mounted or set Cloves (whole fruit, cloves and stems)

Description

565,616 205,280 152,969 118,679 96,839 52,559 45,503 41,940 41,263 39,979 28,371 17,574 8,804

($)

2007

1 2 3 4 5 6 7 8 9 10 12 19 30

Rank

309

African Export Successes Table 8.3A

Rank correlations of top exports today with past, focus countries Rank correlations between start and end year

Country

Start

End

Top 50

Top 100

All goods

N

Ethiopia Ghana Rwanda Tanzania Uganda

2001 1996 2003 1998 1995

2008 2008 2008 2007 2008

0.261 0.362 0.443 0.000 0.247

0.407 0.318 0.503 0.333 0.307

0.405 0.557 0.292 0.529 0.458

775 1,031 572 1,138 1,087

Table 8.3B

Rank correlations between start year and end year within countries All

Top 50 in start year

Top 100 in start year

A. Average for 33 African countries All export goods 0.540 0.248 Excl. extractables 0.544 0.249 Excl. extractables and commodities 0.543 0.227

0.293 0.290 0.273

B. Average for 101 non-African countries All export goods 0.786 0.200 Excl. extractables 0.786 0.195 Excl. extractables and commodities 0.788 0.194

0.292 0.291 0.289

Notes: Start year varies for African countries, median is 1998; end year is usually 2008, occasionally 2007. Start year is 1998 for non-African countries and 2008 for end year. Data: HS four-digit, Comtrade.

By dividing by ∆X we have 1 = DB / DX + N (t ) / DX − L (t − 1) / DX . Table 8.4 reports this decomposition, as well as X(t), X(t – 1), and the average annual growth rate of exports (not all start and end years are the same). All values are in 2008 prices. The analysis is performed at the sixdigit level, which is appropriate for describing products. While the median growth due to the intensive margin is 70 percent, new products account for 43 percent, while lost products account for –4 percent of export growth (these numbers need not sum to 100 percent because the median is applied to each category separately). The table shows that much of the changes in success are attributable to new goods and that there is a lot of churning. Yet another way to demonstrate that large changes in composition of success are typical is the following. Using data on top forty products for each of the thirty-three sub-Saharan countries in the Comtrade data, we identify products with negative change in share and take sum of all of those,

Botswana Burkina Faso Cameroon Cote d’Ivoire Ethiopia Gabon Ghana Guinea Kenya Lesotho Madagascar Malawi Mali Mauritania Mauritius Mozambique Namibia Niger Nigeria Rwanda S. Tome & Principe Senegal Seychelles South Africa Sudan Tanzania Uganda Zambia Zimbabwe Median

Exporter

Table 8.4

2000 1995 1995 1995 1997 1993 1996 1995 1997 2000 1990 1990 1996 2000 1993 2000 2000 1995 1996 1996 1999 1996 1994 1992 1995 1997 1994 1995 2000

First year

2008 2005 2006 2008 2008 2006 2008 2008 2008 2004 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2008 2007 2008 2008 2007

Last year 3,368,768 197,667 1,944,587 3,640,389 710,709 3,186,509 3,215,205 900,479 2,398,136 366,938 432,044 600,499 507,907 272,311 2,034,127 350,126 1,612,501 251,825 14,869,750 12,712 2,740 392,542 32,230 17,121,042 911,502 745,552 143,064 1,392,485 2,304,749

First year 4,825,800 329,378 3,399,945 9,674,154 1,595,059 6,015,203 4,029,949 1,486,836 4,629,977 967,758 1,483,924 878,699 1,913,799 1,081,147 2,086,809 2,332,100 4,682,885 439,178 79,574,670 346,110 5,618 1,776,324 149,709 73,102,248 9,466,236 1,962,557 1,338,063 5,070,833 3,169,664

Last year

Exports (thousands of US$ in 2008 prices)

Decomposition of export growth, HS six-digit level

70 83 89 52 67 86 71 65 81 74 81 73 –7 85 –47 11 85 31 95 75 83 53 55 58 –4 53 33 40 87 70

Products exported in both years (%) 222 27 26 50 40 18 43 43 23 56 22 42 109 15 296 90 20 179 6 25 19 51 48 43 105 49 68 65 57 43

New products (%)

Growth decomposition

–192 –10 –15 –2 –7 –3 –14 –8 –4 –30 –3 –15 –2 0 –149 –1 –5 –110 0 0 –3 –3 –4 –2 –2 –1 –1 –4 –44 –4

Lost products (%)

4.5 5.1 5.1 7.5 7.3 4.9 1.9 3.9 6.0 24.2 6.9 2.1 11.1 17.2 0.2 23.7 13.3 4.3 14.0 27.5 8.0 12.6 11.0 9.1 18.0 9.7 16.0 9.9 4.6 8.0

Export growth per year (%)

African Export Successes

311

separately for each of thirty-three countries. We then identify products with positive change in share and take sum for all of those, separately for each of thirty-three countries. Then we take averages across thirty-three countries (same start date and end date within each country). On average the sum of the negative changes is –26 percent and the sum of the positive changes is 31 percent. This implies much turnover in the shares of the top forty products. 8.3

Changes in Export Shares Are Not Driven by Prices

In this section we demonstrate that the surprisingly large changes in export shares are driven primarily by quantity changes rather than price changes. Since Comtrade does not include data on prices, we use unit values as proxy for prices. Unit values are just the weighted average of prices within a particular product category. We use the following decomposition of changes in export shares. The export share of product i in time t, s(i,t), is given by s (i,t ) = r (i,t ) / R (t ) , where r(i,t) is the revenue of product i in time t and R(t) is total revenue in time t. In logs this can be written as ln s (i,t ) = ln r (i,t ) − ln R (t ) = ln p (i,t ) + lnq (i,t ) − ln R (t ) where p and q represent price and quantities, respectively. Taking differences, this becomes Dln s (i ) = Dln p (i ) + Dlnq (i ) − Dln R and thus 1 = Dln p (i ) / ( Dln s (i ) + Dln R ) + Dlnq (i ) / ( Dln s (i ) + Dln R ). We use this decomposition to gauge the relative importance of changes in prices and quantities to export shares, controlling for the growth in overall export revenue. For each country we computed the median percent of changes due to prices and quantities, then we computed medians across countries. Table 8.5 reports the results of this exercise. Price changes account for much less than quantity changes—only 10 percent of changes in shares for the median country, when all products are taken into account. This result is robust to restricting to the top forty products, only commodities, or only noncommodities. Although price changes have the largest role among the top forty products, it still explains only about 19 percent of the percent change in export shares on average. This is evidence against that traditional view that sees African export performance as explained mainly by world prices. African countries are not just passively exporting their commodity endowments. To drive our point further, we demonstrate that global forces (prices or

312

William Easterly and Ariell Reshef

Table 8.5

Category All products Top 40 Commodities Noncommodities

Decomposition of export share growth between price and quantity

Median first year

Median last year

Median no. of HS4 products

1998 1998 1998 1998

2008 2008 2008 2008

247 40 5.5 242

ln s (median) (%)

Δ ln p/ (Δ ln s + Δ ln R) (median) (%)

Δ ln q/ (Δ ln s + Δ ln R) (median) (%)

39 73 54 43

10 19 9 10

90 81 91 90

Note: Numbers are medians across thirty African countries for HS4 products.

other) are not important forces in determining commodity export revenues and in changes in Big Hits. We find that global year fixed effects do not explain much of the time variation in individual commodities exported by multiple African countries. There is a high share of idiosyncratic time variation in total time variation. Moreover, we find a very small difference between commodities and noncommodities. We fit the following fixed effects regressions: r ( c,t ) = a ( c ) + d (t ) + e ( c,t ) , where r(c,t) is export revenue from some product that is exported by many countries c in many time periods t, a(c) capture country effects, t(t) capture time effects, and e(c,t) is an idiosyncratic error. Given the estimates of such regressions for several products, we decompose the variance V ( r ) = V (C ) +V (T ) +V ( e ) , where C denotes country fixed effects, T denotes global year dummies, and e is the residual. The purely intertemporal variation in each export good is V(r) – V(C). Table 8.6 reports the results of this variance decomposition for a set of products that are prevalent in African exports. For each product the sample is all countries that export it. If commodities’ variation over time were driven by global prices, V(r) – V(C) would be largely accounted for by global price movements, which would be captured by V(T). However, the share of V(T) in explaining intertemporal variation, that is, V(T)/(V(r) – V(C)), is small. Moreover, there are no statistically significant differences between commodities and noncommodities in this regard. The role of country endowments, V(C)/V(r), is statistically larger for commodities, but the difference is not economically large (we reject the hypothesis that V(C)/V(r) has the same in both groups of products, at standard levels of significance). This means that there is substantial specialization across countries in differentiated products, not much less than in commodities. The role of global price movements in commodity export success is much smaller than what one traditional view of commodity exports would

African Export Successes Table 8.6

313

Variance decomposition of export revenues by country and global factors V(C)/V(r) (%)

V(T)/V(r) (%)

V(T)/[V(r) – V(C)] (%)

Tea (HS 902) Gold (HS 7108) Coffee (HS 901) Sugar (HS 1704) Diamonds (HS 7102) Cotton (HS 5201) Cocoa (HS 1801) Tobacco (HS 2401) Oil (HS 2709 and 2710)

85 72 93 75 88 76 90 86 78

0 4 1 3 1 2 1 2 1

3 14 11 12 12 10 8 14 7

Median

85

1

11

Commodities

Noncommodities Mixed odoriferous substances for industrial use (HS 3302) Cut flowers, dried flowers for bouquets, etc. (HS 603) Goat or kid skin leather, without hair (HS4106) Wood in the rough or roughly squared (HS 4403) Polymers of ethylene, in primary forms (HS 3901) T-shirts, singlets and other vests, knit or crochet (HS 6109) Oral and dental hygiene preparations (HS 3306) Men’s or boys’ suits, jackets, trousers, etc., not knit (HS 6203) Fish, frozen, whole (HS 303) Prepared or preserved fish, fish eggs, caviar (HS 1604) Printed reading books, brochures, leaflets, etc. (HS 4901) Vegetables nes, fresh or chilled (HS 709) Woven cotton fabric, >85% cotton, < 200g/m2 (HS 5208) Median

  76 84 61 84 57

2 0 3 2 10

7 3 8 13 24

87 70

1 2

5 6

78 70 78

1 4 2

6 13 11

74 74

3 5

10 19

71 74

1 2

5 8

Notes: The table reports the variance decomposition of export revenue into country factors (C), global time factors (T) and residuals (e), that is, V(r) = V(C) + V(T) + V(e). The purely intertemporal variation in each export good is V(r) – V(C).

predict. Within-product decompositions for each country show that changes in export shares are driven more by quantity changes than by price changes. Finally, the role of country endowments and global prices is not different between commodities and noncommodities. 8.4

Measurement Error Concerns

Some of our results are sensitive to the existence of measurement error. We do notice potential measurement problems, first by observing spottiness of coverage of export product data by country, both at the six-digit and four-digit level. In particular, there are many blanks for products in years

314

William Easterly and Ariell Reshef

that earlier and/or later had significant positive values. Therefore, in all of the analysis above we choose the start year for each country at a point when the coverage becomes extensive; usually there is a clear dividing line between very spotty coverage and consistent coverage. However, this procedure does not guarantee that coverage is complete in the later years. In this section we report a few examples that indicate that there are indeed serious data problems in the Comtrade data. We did not exhaustively check all data. We found these errors in the course of closer examination of Comtrade data that pertains to (potential and actual) African export case studies that we report in the final section of this chapter. For these goods, we first investigate measurement error at the six-digit HS code level and then examine the data at the four-digit level to see whether aggregation alleviates measurement error. In some cases the importer and exporter data roughly agree. Aggregation may alleviate discrepancies: it seems preferable to use four-digit over six-digit data. Measurement error is evident in discrepancies between importer reports and exporter reports on the same trade flows by year and by good, and there are discrepancies in blank entries between importer reported data and exporter reported data. The modest improvements when aggregating sixdigit to four-digit level indicates that there are classification disagreements at the six-digit level, but we were disappointed it did not improve more than it did. Is it possible that one of the sides systematically misses some of the trade flows, some of the time? Except for the case of Rwanda leather products reported below, it did not seem obvious which side was underreporting on average. This is the case whether we use cost, insurance, and freight (CIF) or free on board (FOB) export data. One reason that there are many discrepancies for both countries examined above is that they are landlocked; they do not have their own port, and hence do not ship anything by sea directly. Rwanda often exports via the Mombasa port in Kenya. Exports are documented as being shipped to Kenya (or Uganda, which is on the way), but the final destination is not Kenya. This is the case for coffee exports from Rwanda. Of course, landlocked countries could export some products by air directly, but even this is not always the case. For example, Tanzania (not a landlocked country) sometimes exports fresh (chilled) fish by air from Entebbe or even Nairobi (a bit less than twenty-four hours, driving).10 Much of the exports of fresh fish from Tanzania are documented in the data as being exported to Kenya and Uganda, but little is consumed there and almost all finds its way to Europe.11 10. This is because the airstrip at Mwanza, on the shore of Lake Victoria, where most fish processing occurs, is too short for some large cargo planes. 11. Another source of discrepancies is the fact that since relative peace has been achieved in southern Sudan, regional exports to that destination have boomed, but most of this is informal and does not show up in statistics. This has been indicated by Dr. Adam Mugume from the Bank of Uganda.

African Export Successes

315

We are therefore worried that instability of exports could just be reflecting measurement error: a possible caveat for our results about changing in composition of Big Hits. However, we would expect measurement error to be the same for commodities and noncommodities. Therefore, the results that commodity exports are not systematically more volatile over time—nor more driven by global prices—still hold. Since measurement error may be more serious in poorer regions, the results comparing African and nonAfrican countries are somewhat more questionable (although an offsetting effect might be the greater number and complexity of products traded in rich countries). We do not see any obvious solution to the measurement error problem. Limiting the analysis to products in which importer and exporter reports match closely may induce a selection bias to certain types of products in which such agreement is more likely. So far we see aggregation to the fourdigit level as the only way to alleviate the problem. Our hope is that examining the data from many different angles may alleviate measurement error problems, but we have no way of knowing whether such hopes are justified. In the end, we are left with the usual irreducible helplessness in working with the data that are available. 8.4.1

Leather and Hides in Ethiopia and Rwanda

Six-Digit Analysis Our first exercise is to compare blanks and nonblanks in exporter and importer data in the leather and hide industry in Ethiopia and Rwanda. In Ethiopia there are thirty-two six-digit goods under this category in the years available, 2001–2008. Table 8.7 summarizes the data. The two sources match 70 percent of the time. When the importer reports a nonblank, the exporter does so 78 percent of the time; the reverse calculation shows when the exporter reports a nonblank, the importer does also 72 percent of the time. The off-diagonal elements show a slight tendency for importers to be more likely to report blanks when an exporter does not, compared to the other way around. This calculation does not suggest that any one source can be identified as underreporting. This is confirmed by comparing export revenues for the 112 observations that both have nonblanks. Exporter quantity is greater than importer quantity in fifty-five observations, that is, in almost exactly half of the cases. The correlation of the magnitudes for these 112 observations is only .47, which suggests there is some signal there but also a lot of noise. The Rwanda Comtrade data before 2003 is very patchy and unreliable, especially in the exporter-reported data, with obvious signs of severe underreporting. Therefore all the tables in this chapter for Rwanda begin in 2003. In Rwanda there are twenty-three six-digit goods under the leather and hide group in the years available, 2003–2008. Table 8.8 summarizes the data.

316

William Easterly and Ariell Reshef

Table 8.7

Ethiopia, six-digit leather sector, 2001–2008 Importer blank

Importer not blank

Sum

68 44

32 112

100 156

112

144

256

Importer blank

Importer not blank

Sum

Exporter blank Exporter not blank

72 30

16 20

88 50

Sum

102

36

138

Exporter blank Exporter not blank Sum

Table 8.8

Rwanda, six-digit leather sector, 2003–2008

Rwanda has a more serious problem of inconsistency. Although the two sources match 67 percent of the time, this mainly reflects the high number of blanks in both sources. When the exporter reports a nonblank, the importer does so only 40 percent of the time. When the importer reports a nonblank, the exporter does so 56 percent of the time. In sum, there are more nonblanks reported by exporters than by importers. This suggests the importer data is the one that tends most to underreport. This conclusion for importers from Rwanda underreporting is confirmed by the twenty observations for which both sources report nonblanks. The exporter quantity exceeds the importer quantity in fifteen of these cases. The correlation of magnitudes for the twenty observations is basically zero. Four-Digit Analysis Aggregation may help the error problem. Exporters and importers may classify correctly broad product categories at the four-digit level, but might not pay as much attention to the fifth and sixth digits. A casual examination of some product descriptions confirms that the six-digit classification can be quite subtle when it comes to manufactured goods. In Ethiopia the matching of blanks and nonblanks in exporter and importer data at the four-digit level increases to 82 percent, as can be seen in table 8.9. It is somewhat puzzling that now the exporter seems to be underreporting relative to the importer as far as the blanks matrix. However, the exporter quantity is greater than the import quantity in 57 percent of the cases where both are nonblank, so it is not clear on which side there is underreporting. The correlation between the two sources rises slightly to 0.54 relative to the 0.48 correlation at the six-digit level. Figure 8.2 reports the result of adding all nonblank entries in the leather and hides group for each year to form the highest level of aggregation for

African Export Successes Table 8.9

317

Ethiopia, four-digit leather sector, 2001–2008 Importer blank

Importer not blank

Sum

Exporter blank Exporter not blank

16 2

14 56

30 58

Sum

18

70

88

Importer blank

Importer not blank

Sum

Exporter blank Exporter not blank

12 14

11 17

23 31

Sum

26

28

54

Fig. 8.2

Ethiopian leather and hide exports

Table 8.10

Rwanda, four-digit leather sector, 2003–2008

this group. Both exporter and importer data are of the same order of magnitude and exhibit similar trends, although in the exporter data the decline in 2001–2004 and the increase in 2004–2008 are much more pronounced. In Rwanda, on the other hand, the data inconsistencies do not improve at the four-digit level relative to the six-digit level, as can be seen in table 8.10.

318

Fig. 8.3

William Easterly and Ariell Reshef

Rwandan leather and hide exports

The underreporting still seems to be on the importer side, because thirteen of the seventeen observations with nonblank entries in both exporter and importer data are greater in the exporter-reported data. This is even clearer when we aggregate all leather and hide exports by year for Rwanda. As can be seen in figure 8.3, importer-reported data are consistently below exports. This is likely due to the fact that Rwanda is landlocked. 8.4.2

Ethiopian Shoes

The data for Ethiopian shoes are also problematic. The exporter and importer data match blanks and nonblanks only 66 percent of the time. Table 8.11 suggests some underreporting by exporter data. However, when both have data, the exporter quantity is greater than the importer quantity 57 percent of the time. At the four-digit level of aggregation the impression that exporters are the ones who are underreporting is strengthened, since importers have fewer blanks than do exporters, as seen in table 8.12. At the most aggregate level, aggregating over all shoe products the importer and exporter data on shoe exports match closely year by year, as illustrated in figure 8.4. In this case, aggregation substantially solves measurement problems.

African Export Successes Table 8.11

319

Ethiopia shoes, six-digit categories, 2001–2008 Importer blank

Importer not blank

Sum

90 25

48 53

138 78

115

101

216

Exporter blank Exporter not blank Sum

Table 8.12

Ethiopia shoes, four-digit categories, 2001–2008 Importer blank

Importer not blank

Sum

Exporter blank Exporter not blank

1 3

11 33

12 36

Sum

4

44

48

Fig. 8.4

8.4.3

Ethiopian shoe exports

Rwanda Coffee

Like most other Rwanda Comtrade data, the coffee exporting data before 2003 is very patchy and unreliable. However, as figure 8.5 illustrates, beginning in 2003 the exporter and importer reporting on unroasted coffee from Rwanda coincides remarkably well.

320

William Easterly and Ariell Reshef

Fig. 8.5

8.5

Rwanda exports of unroasted coffee, HS 090111

Pathways to Big Hits

Subject to caveats about measurement error, we have rejected that traditional view that sees African exports as reflecting mainly a passive endowment by nature and driven by global commodity prices. In that case, what are the other paths to success? To answer these questions we traveled to Africa to interview exporting entrepreneurs from successful industries. Our methodology is as follows. First, we examined four-digit HS code exports from all countries to try to detect success stories. The data are from the same Comtrade database that we used to establish the stylized facts above. We identify success stories as exports of products that are increasing export revenues dramatically and increasing their share in total exports, and/or increasing unit values; all must have attained a significant size. We do not examine extractable commodities and their derivatives (oil, gold, ores, iron bars, etc.). Given a set of candidate products, we traveled to three East African countries—Rwanda, Uganda, and Tanzania—to meet entrepreneurs that export them. The choice of countries was dictated by preexisting contacts. The sample of entrepreneurs that were interviewed was dictated by those contacts. We restricted attention to those industries identified above as export success stories. In Rwanda, our initial contact was with one coffee exporter, who introduced us to other entities in this industry, including one American importer. An economist at the Rwanda Development Board helped us get

African Export Successes

321

in touch with entrepreneurs in other successful export industries from our set of candidates. In Uganda, our initial contact was a journalist; we drew on his personal contacts to reach entrepreneurs. This explains the smaller number of entrepreneurs we met there. Finally, our contacts at the Bank of Tanzania gave us a list of entrepreneurs drawn from the set of successful export industries there. We provide excerpts of interviews in an appendix.12 Clearly, the sample is not representative of economic or even exporting activity; it includes only successful exporters. As such, it is impossible to test the importance of the broad factors, as well as transportation costs, financial constraints, and so forth—although we still document entrepreneurs’ account of the importance of such factors. But the goal of the interviews was to identify why export of a particular product took off. In that sense, the sample suits our purposes. In the process we obtained firsthand accounts of the business model of the firms we visited, as well as difficulties facing exporters in Africa. We did not manage to interview entrepreneurs in all candidate-exported products, but the picture that emerges illustrates many reasons for success and for Big Hits. The interviews always started with an introduction to our mission, followed by an open discussion about the firm: how and when it was founded, product range, when it started to export, and so forth. In this first part we let the entrepreneur speak freely, while we asked for clarifications along the way. Later in the interview, in order to focus better on the topic at hand, we asked questions from a list that we prepared in advance. Not all of these questions had to be answered directly, but they served as guidance to facilitate a deeper understanding of the characteristics of the exporting activities of the firm. We organized the discussion around broad determinants of export success versus idiosyncratic determinants. The former include removal of trade barriers, well-known Heckscher-Ohlin labor cost advantages for labor-intensive exports, specific resource endowments, moving up the quality ladder for “traditional” low-quality export products, superior technology foreign ownership, and ethnic networks. One particularly interesting aspect of successful exporting entrepreneurs in many cases is personal foreign experience outside of Africa, which is not necessarily related to business. In addition, other general issues like the importance of quality and transportation costs and finance for exporting were evident in the interviews.13 In almost all cases we find that entrepreneurs are actively conducting market research and feasibility studies to determine where to invest and where to export. But personal contacts are important in staring exporting and in exporting to new destinations. We report separately a few cases in which idiosyncratic determinants played a particularly strong role. 12. See http://www.nber.org/data-appendix/w16597/ER_AfricaExportSuccess_interviews_ appendix.pdf. 13. Indeed, Freund and Rocha (2010) find that land transport delays are the most detrimental factors that constrain African trade, much more than tariff reductions.

322

William Easterly and Ariell Reshef

A major theme is that exporting requires a particular mind-set, an exportoriented business model (Artopoulos, Friel, and Hallak 2010). Relative to serving the domestic market, exporting requires better planning, meeting deadlines, dealing with a lot more paperwork, better quality control, maintaining product consistency, and so forth. All these, in turn, require modern organization and management methods, investment in information and communications technology (ICT), and appropriate marketing strategies— which imply a modern way of doing business. This was evident in most of the firms we studied. In addition to the broad determinants, we documented interesting cases in which the reasons for export success are very idiosyncratic. In two cases— quality coffee and handicrafts exports from Rwanda—financial assistance and help in penetrating foreign markets in the West from aid organizations, and to some degree government assistance, was instrumental. Note that these are the exception in our sample and concentrated in Rwanda. Although aid can help some people some of the time, overall, the evidence for a causal link from aid to growth does not pass empirical scrutiny (see Easterly [2003], Rajan and Subramanian [2005], and references therein).14 If all aid assistance was as successful as what we found in Rwanda coffee and handicrafts, then we would have expected a causal link would be established in the aggregate. This is why we consider the success of aid-assisted exports as idiosyncratic, rather than include aid as a broad category for export success. In the handicrafts case, this success in Rwanda may be even more idiosyncratic, as some experienced aid practitioners find that handicraft projects face severe challenges and often fail.15 Luck—that is, being in the right place at the right time with the right knowledge and connections—played a particularly important role in the case of chilled fish exports from Lake Victoria. We also find that cost shocks can reverse a success in one location (fresh-cut flowers from Uganda), while another location may gain at its expense (Ethiopia). In addition, we document the persistence and passion of one pioneering entrepreneur as the main determinant for successfully exporting roasted coffee from Uganda. Finally, since each firm is different, we list a few factors that were instrumental in the success some of the remaining firms.16 We expand on each of these idiosyncratic cases below.

14. Easterly (2003) also criticizes this causal link on theoretical grounds. 15. See Saundra Schimmelpfennig, “Problems with Selling Handicraft Projects Internationally,” September 2009, at http://goodintents.org/common-aid-problems/selling-handicraftinternationally. 16. Freund and Pierola (2010) report three cases from Peru with similarities to ours. Fresh asparagus exports started with the help of USAID in the 1980s. Paprika exports started due to one entrepreneur hearing about it from a friend in Chile. Fresh artichoke exports started as a private initiative to coordinate efforts to discover a profitable growing technique after several uncoordinated attempts failed (eventually, only canned artichokes were exported).

African Export Successes

Fig. 8.6

8.5.1

323

Coffee exports unit value per kilogram

Moving up the Quality Ladder for “Traditional” Low-Quality Export Products

Introduction of Fully Washed Coffee in Rwanda Coffee is a traditional cash crop in all of East Africa. As such, it was usually of poor quality. However, we have witnessed a recent trend in producing high-quality (fully washed) coffee, for which labor-intensive processing is needed. Figures 8.6 and 8.7 show the upward trend in worldwide coffee prices and, in particular, in Rwanda, Ethiopia, and Uganda. However, we also see substantial differences between unit values across countries and across qualities of coffee. From figure 8.7 it is evident that higher quality fetches higher prices. The average price for fully washed Rwandan coffee is higher than the average price for ordinary coffee, while coffee sold by the exporting firm RWASHOSCCO and by the Maraba co-op fetches even higher prices per kilogram.17 The increase in coffee-export revenues for Rwanda, evident in figure 8.8, is not driven by an increase in volumes. Export quantities have fluctuated with no trend since 2002. The increase in revenue is driven by a shift toward fully washed coffee, which by 2009 accounts for 23 percent of exports and 32 percent of revenue (see figure 17. We thank Jean-Claude Kayisinga of the SPREAD project in Kigali for providing the detailed data for Rwanda coffee exporting.

Fig. 8.7

Rwanda coffee price comparison

Fig. 8.8

Coffee exports from Rwanda (millions of dollars)

African Export Successes

325

Fig. 8.9 Income and production shares of fully washed coffee exports from Rwanda, percent of total coffee exports

8.9). This, together with the global increase in prices, explains the increase in revenue. These seemingly small differences in prices are compounded by large and growing quantities of specialty coffee, and they also make a huge difference for the farmers. In our visit to the Maraba village, we saw new construction, and even a brand new bank branch. How was this achieved? The United States Agency for International Development (USAID)funded Project to Enhance Agriculture in Rwanda through Linkages (PEARL) and its descendant, the Sustaining Partnerships to Enhance Rural Enterprise and Agribusiness Development (SPREAD) project, in collaboration with Texas A&M University (Norman Borlaug Institute for International Agriculture), introduced fully washed coffee techniques to Rwanda in 2000. The first co-op to export fully washed coffee (Maraba) was founded by PEARL in 2001 and the first shipment was exported in 2002. Following this, many co-ops adopted the technology. Again, we note that this successful aid intervention is the exception, rather than the rule. And as the next quality upgrading case demonstrates, the private sector can also develop quality coffee exports and even develop this further into final products (roasted and airtight packed, rather than green coffee beans), so external intervention is not a necessary condition. Founded in 2005, RWASHOSCCO is an exporting firm that is owned by co-ops that export only fully washed coffee; RWASHOSCCO received funding from USADF. Another exporting firm that does the same is Misozi, founded in 2007 with help from the International Fund for Agri-

326

William Easterly and Ariell Reshef

Fig. 8.10

Single-sourced coffee from Bufcafe in New York, café Gimmee! Coffee

cultural Development (IFAD). Both are owned by coffee-growing co-ops. The Maraba co-op produces only fully washed coffee, which is exported as single-source/traceable coffee from Rwanda. The neighboring, privately owned Bufcafe washing station does the same. These are fully washed coffees that are bought for a premium, mostly by gourmet cafés and roasters in the United States (for example, Gimmee! Coffee in New York; see figure 8.10). Other examples are: Intelligentsia, which has cafés in Chicago and Los Angeles; Third Rail Coffee located in New York; and the Whole Foods supermarket chain—all of which sell the coffee under the name of the co-op, hence the term single-source/traceable. The owner of Intelligentsia travels to visit the co-ops from which he buys coffee to maintain personal relationships with growers, to maintain quality, and to advise. Good African Coffee from Uganda Another case of quality upgrading is the privately owned company Good African Coffee, based in Kampala, Uganda. But in this case not only is the coffee fully washed, it is roasted and packed and exported as a final product

African Export Successes

327

directly to supermarkets in the United Kingdom, thus capturing the entire value chain. As such, the marketing effort includes design and careful airtight packaging. Another distinction from the Rwandan case is that Good African Coffee is the brainchild of one African entrepreneur, Andrew Rugasira, who is also the sole owner. Rugasira founded Good African Coffee (GAC) in 2003 to produce, roast, and export quality coffee to Western markets. The goal is to capture as much of the value chain as possible. To ensure supply of quality coffee beans, GAC formed farmer co-ops for growing coffee in Western Uganda. They taught them how to process high-quality fully washed coffee beans and funded capital equipment. Today more than 14,000 farmers supply coffee to GAC through these coops.18 Coffee was roasted for local consumption since the 1970s by the Ugandan coffee board. Good African Coffee is not the first to roast locally, but is the first in Africa to export high-quality coffee. It is the only African-owned brand to export to the United Kingdom. Good African Coffee first sold roasted and ground coffee in South Africa in 2004, using a plant there. In 2005, GAC started to sell to the supermarket chain Waitrose. The roasting and grinding facility moved to Dublin and GAC pulled out of South Africa. In order to satisfy local tastes, in 2006 GAC launched their freeze-dried instant coffee, which together with their roast and ground coffees was listed in the British supermarket chain Sainsbury’s. Freeze-dried instant coffee is also sold to Tesco. In July 2009, GAC set up a roasting and packaging facility in Kampala to do all the processing in Africa. In July 2010, GAC started selling roasted and ground coffee to the British supermarket Tesco. From November 2010 GAC products will be available for purchase in the United States via the Internet. Note that one factor that may have played a role in inducing quality upgrading in Rwanda and Uganda are high transport costs. Both are landlocked countries with poor-quality ground transport both at home and in the countries with neighboring ports, while air freight is of course more expensive. Increasing the export value per unit weight by upgrading quality may have been a response to this transport cost problem. 8.5.2

Comparative Advantage

Comparative advantage manifested itself in the interviews as well, particularly in the following products: coffee from Rwanda and Uganda, flowers and cuttings from Uganda, and fish from Uganda and Tanzania. These are all exports that rely on natural endowments, but also on idiosyncratic features, which we detail below, case by case. The soil in Rwanda and western Uganda, as well as their relatively high altitudes, is particularly good for 18. Since many of the co-ops are located near national parks, USAID helped in educating the farmers on conservation. Rugasira stresses that the involvement of USAID was limited to this activity. He is a vociferous opponent of aid and has expressed his views (“trade, not aid”) in writing and speech.

328

William Easterly and Ariell Reshef

growing Bourbon Arabica coffees. Likewise, flower exports from Uganda also rely on suitable soil and high altitudes. Nile perch was abundant in Lake Victoria long before it was being exported chilled. But there are other endowments on which entrepreneurs in Africa draw on. For instance, the firm Gahaya Links from Rwanda exports woven baskets (and more recently jewelry) that are based on traditional Rwandan designs and techniques. In this case, it is a cultural endowment that is unique to Rwanda that helps explain the increase in handicraft exports from there. The case of Gahaya Links also exhibits interesting idiosyncratic patterns, on which we elaborate below. Labor cost advantages also play an important role in export success in some of the industries we studied. Fully washed coffee, handicrafts (Gahaya Links), and flowers are all labor-intensive activities. According to the wellknown Heckscher-Ohlin forces, low labor costs create a comparative advantage in these industries. According to the coffee importer and roaster, Intelligentsia, the quality of coffee achieved in East Africa cannot be achieved any more in Central America because labor costs there have increased. Fully washed coffee is still exported from Central America, but the quality of East African coffee is higher due to the fact that they can employ more labor due to low wages there. (See figures 8.11, 8.12, and 8.13.)

Fig. 8.11

Fish fillet exports from Uganda

Fig. 8.12

Fish fillet exports from Tanzania

Fig. 8.13

Handicraft exports from Rwanda (excluding antiques)

330

William Easterly and Ariell Reshef

8.5.3

Trade Liberalization and Trade Preferences

In a number of cases trade liberalization was the key factor behind the growth in exports. These are exports of beer and soft drinks from Rwanda, plastics from Tanzania, and oil and soap from Tanzania. Tariff reductions and free trade zones (East African Community, South African Development Community) are critical for the competitiveness of these export activities; without duty-free access to destination markets, they would not be economically viable.19 In the case of both plastics firms we interviewed in Tanzania, trade liberalization also changed importers into exporters. These firms were importers (not only plastics) before serving the local market, and then become exporters. By serving the domestic market with many imported final goods, importers learn about local demand: which products are popular, the price structure, and what are the costs of shipping these products from abroad. This way they discover products that can be produced more cheaply locally for the local market. After starting plastics production for the domestic market and gaining some scale, these firms started exporting, following trade liberalization. Both plastics exporters report that once they have a presence in one market with one product, they expand into other products. Contacts and distributers in destinations markets inform these decisions. Trade preferences are also quite important for a number of exports: coffee, tea, and fish are all imported duty free to the European Union and the United States (see figure 8.14). 8.5.4

State-of-the-Art Technology

Investment in state-of-the-art technology is an important factor in the success of entrepreneurs exporting tea from Rwanda, roasted coffee from Uganda (Good African Coffee), plastics from Tanzania, and oil and soap from Tanzania. In all cases, entrepreneurs invested in highly productive and reliable machinery for two main reasons. The first reason is that this technology is more productive and, in some cases, more flexible. The second reason is that due to lack of local technicians, they must invest in the most reliable machinery. In the case of the beer producer Bralirwa, state-of-theart technology and management best practices are dictated by its majority shareholder, Heineken. One of the oil and soap manufacturers (Bidco Oil and Soap) is a subsidiary of a Kenyan firm. Its technology was transferred from the mother firm. Entrepreneurs imported state-of-the-art machinery from South Korea, Germany, Belgium, and China.

19. We do not comment on whether these constitute trade diversion versus trade creation.

African Export Successes

Fig. 8.14

8.4.5

331

Beer and soft drink exports from Rwanda

Foreign Ownership

The case of the Bralirwa brewery is a clear case of foreign direct investment by Heineken. They succeed in exporting due to implementation of management practices and technology from the mother company, without using its brand. Likewise, Sorwathe, the tea exporter, is owned (80 percent) by Tea Importers Inc., a private tea-trading firm incorporated in Westport, CT, which also buys the lion’s share of Sorwathe’s output. Tea Importers makes sure that the technology used is state of the art, as well as changing the product mix toward higher-quality products (orthodox and green tea, organic certification, etc.). 8.4.6

Ethnic Networks

The impact of informal Indian ethnic networks is particularly evident in the plastics exporters that we studied. For both plastic exporters and one oil and soap exporter the decision to manufacture that particular good was influenced by information obtained from these networks. In particular, technology transfers assisted the entrepreneurs to start their businesses. Successfully serving the local market lead to exporting later on. The entrepreneur who was the first to export fish from Lake Victoria started exporting prawns

332

William Easterly and Ariell Reshef

Fig. 8.15 Exports of plastic table, kitchen, household, toilet articles from Tanzania

from Tanzania due to a connection of a fellow Indian in London. This first connection for exporting prawns opened the door to the seafood exporting business and eventually lead to the multimillion fish export industry.20 (See figure 8.15.) 8.4.7

Personal Foreign Experience

An interesting pattern that emerges from interviewing successful exporting entrepreneurs in Africa is that many of them had life experiences outside of their home country, or outside of Africa. In many instances, these experiences are not related to business. These experiences provided exposure to the world and a particular mind-set and lead to applying the correct “export business model” (as opposed to a domestic mode of operation). This means understanding the importance of meeting deadlines and being organized, meeting customers’ demands and accommodating their tastes, keeping contact with suppliers, and developing a reputation for reliability and keeping it. We do not claim that this is a causal link; after all, the more20. Cadot et al. (2011) report that personal contacts, such as relatives, friends, intermediaries, and suppliers not only provided most of the first-time contacts for new exporters, they also were the most prevalent means for contacting other buyers, buyers contacting the exporter, and for introduction of new product. These were followed by research online and trade fairs.

African Export Successes

333

able entrepreneurs are more likely to spend time abroad in the first place. But this exposure may be more significant for exporting than for serving the domestic market. For example, before emigrating from Canada to Rwanda, Gilbert Gatali, the managing director of RWASHOSCCO, was working as a counselor for Rwandese youth in Canada. Joy Ndunguste, one of the founders of handicraft exporter Gahaya Links, lived in Washington, DC before returning to Rwanda to start her company with her sister. These experiences have contributed to their ability to communicate with importers in Western countries, and to their understanding of how business is done in the West. The founders of Lake Bounty grew up and studied in India fisheries (agribusiness) before moving to Uganda and exporting fish from Lake Victoria. The founders of Bidco Oil and Soap came from Kenya and founded their subsidiary company in Tanzania. The founders of Jambo Plastics were educated in the United Kingdom. A case in point is the difference between Jambo Plastics and Cello Plastics, both exporters from Dar es Salaam. While Jambo actively seeks out new markets and conducts studies, Cello wait for importers (and other domestic buyers) to contact them. Cello does not have any marketing staff and relies on buyers to come to their plant in Dar es Salaam and place orders there. They are amenable to accepting orders by phone or e-mail, but do not maintain a website. Related to this, their life experiences did not include much exposure outside of Tanzania. Jambo Plastics is more export intensive and exports twice as much value.21 Artopoulos, Friel, and Hallak (2010) argue that life experience in the West is particularly important for pioneers who export there. We find evidence that is consistent with this view. The founder of Good African Coffee, Andrew Rugasira, studied law and economics in London before returning to Uganda. Rugasira is pioneering roasted-coffee exporting from East Africa (other countries, for example Ethiopia, have excellent locally roasted coffees but not much is exported). Harko Bhagat, the founder of the first fish exporter from Lake Victoria (Vicfish in Tanzania), studied chemical engineering in Canada. Both entrepreneurs became aware of culture and tastes of customers in the West, as well as a different way of doing business there by spending significant and critical periods of their lives studying in the West. Rugasira exports to the market where he studied, and where there are postcolonial ties (Uganda was a British colony). But Bhagat exports mostly to the European market, whereas his tertiary education was in Canada. This may signal something general about experience abroad, more than what personal contacts and networks might suggest.

21. Data from the Bank of Tanzania.

334

William Easterly and Ariell Reshef

8.4.8

Idiosyncratic Determinants of Success

In this section we outline some of the determinants of success that were documented in particular cases and that do not pertain to others. Singular Success of Aid: Exporting Quality Coffee from Rwanda The case of exporting fully washed coffee from Rwanda is a particularly successful example of international aid intervention.22 We explain why it is exceptional and why it is singular. After the 1994 genocide, the coffee industry was devastated. A slow recovery started from that point. The PEARL (starting in 2000) and SPREAD projects were funded by USAID. Initially, these projects were aimed at general capacity building, by financing twenty students’ education in agribusiness in the United States. Following this, district mayors requested from PEARL an evaluation of which economic activities could be enhanced in order to improve incomes of rural villagers. The decision to focus on specialty coffee followed an evaluation of what will work best, given local knowledge and conditions. The PEARL/ SPREAD projects helped form co-operatives, introduced the techniques for fully washed coffee, and got in touch with potential buyers, some of which were also employed as advisors. The aid money was used for initial capital to buy and build coffee-washing stations, as well as training co-op members in washing technique and teaching coffee-sorting principles. Perhaps the most important reason for the success of the PEARL and SPREAD projects is that they stress partnership and trust building with and between local stakeholders, exporters, buyers, and the government (the [PPP], that is, private-public partnership model). They do not impose their will, but try to empower farmer cooperatives and exporters. Today, SPREAD is gradually transferring its activities to the Rwanda Coffee Board, which was not promoting specialty coffee in the past—but now it is. They also promote leadership within the industry and hope to eventually not be involved. Another critical determinant was the flexibility of USAID, which allowed the evolution of PEARL from general capacity building into focusing on specialty-coffee exporting. This continues today with the (smaller) involvement of USAID in funding SPREAD. There are, even in this partnership model, problems with up-scaling, making efficiency improvements, quick response to market conditions and opportunities, and flexibility.23 22. A counter example has been the so far limited success in exports of Ethiopian shoes, despite extensive aid and government efforts to promote such exports. One obstacle seems to be the poor quality of local hides. In a previous trip to Ethiopia one shoe exporter we met in the countryside pointed to a cowherd beating a cow, leaving scars on the hide, saying “that’s our problem!” The market for hides in Ethiopia is underdeveloped, with shoe exporters buying hides complaining that the market is not discriminating enough about quality to establish separate prices for low- and high-quality hides. 23. In a previous research trip to Ghana we studied the case of a World Bank project to promote exports of pineapples. In stark contrast to the PEARL/SPREAD projects, presidents

African Export Successes

335

We interviewed executives from two specialty coffee exporting firms: RWASHOSCCO and Misozi Coffee Ltd. Both firms are owned by the co-ops that supply them with coffee, but their setup costs and some of the working capital came from aid organizations (USADF, USAID, and IFAD), either directly or indirectly, as soft loans. The firm RWASHOSCCO received funding from USADF to facilitate its expansion and control of the supply chain to include a dry processing plant and roasting facility. The IFAD organization supports coffee-growing cooperatives in Rwanda, whose produce is exported by Misozi Coffee. The IFAD stepped in to resuscitate cash crops by rehabilitating existing farms, distributing seedlings, helping to form co-ops and farmer associations, helping co-ops build coffee-washing stations (CWS), and providing soft loans. Before the establishment cooperative-owned exporters there used to be nine middlemen (!) between farmer and buyer. The co-op structure, ownership of CWSs and of exporting firms, is to shorten the value chain. The IFAD eventually also brought in representatives from Twin Trading Co. (a large coffee-trading company) to help teach how to wash coffee and control quality. When the Twin Trading project was being phased out, Misozi was established. The partnership model is the main reason for success of the intervention. But introducing fully washed coffee is relatively easy where coffee cultivation is widespread. Moreover, exporting fully washed coffee is a viable business because labor costs are low. Although many benefits currently accrue to the farmers and exporters involved, it does not seem to be a long- or even medium-term strategy for growth. For example, fully washed coffee from Central America is of lower quality because higher labor costs prevent using the extremely labor-intensive technology that is employed in Africa. Thus, if wages increase (as one would hope they do), then the quality of the coffee exported from Rwanda may suffer. Overcoming a Plethora of Obstacles: Exporting Handicrafts from Rwanda The case of Gahaya Links combines many of the determinants of success to an extent that stands out. The ability of the founders to overcome so many obstacles that plague most handicrafts enterprises in the developing world merits a closer examination. Gahaya Links was founded in 2003 (and registered in 2004) by Janet Nkubana and Joy Ndunguste (who are sisters) with financial and logistical help from USADF. It is a privately owned handicraft exporting firm. The sisters decided to be pioneers based on their perceived potential for the product and their desire to help women in Rwanda after the genocide. They of cooperatives were discontent with the way the World Bank imposed an exporting firm on them. This exporting firm turned out to be inefficient and did not completely fulfill its obligations.

336

William Easterly and Ariell Reshef

do not manufacture anything directly: co-ops do. The firm is founded only for exporting. Support from USADF was instrumental from the foundation and on, until this day. In 2005 USAID funded Gahaya’s participation in trade fairs in the United States. This led to establishing critical contacts with FairWinds Trading Inc., a marketing and trade company importing African crafts, and with Macy’s. The Rwandan government has also helped achieving exposure by showcasing products in its embassy in the United States. Since then, they have been successfully selling handwoven baskets with unique designs in the United States. Baskets were never exported from Rwanda before. The baskets that are exported are of higher quality than those found in Rwanda. The product is based on traditional Rwandese designs, but is modified to satisfy tastes in the United States. The breakthrough came in 2007 with a $300,000 contract from Macy’s. This was not only lucrative, it created visibility, and as such was instrumental in opening new markets. Gahaya is currently expanding into exporting handicraft jewelry and fabrics, all of which are based on traditional Rwandese designs. This is in addition to satisfying growing demand for their flagship products, woven baskets. They are currently in the process of purchasing a warehouse in the United States to help satisfy demand. In addition, they have founded a subsidiary in the United States to help with marketing. Gahaya currently works with 5,000 weavers organized in fifty-two co-ops. The firm has a training center in Kigali, where co-op members learn new designs and techniques, and how to maintain consistent high quality. As mentioned above, Gahaya taps into a Rwandan resource: traditional basket-weaving techniques and designs. However, they are not the only incidence of such reliance on “ethnic” resources. Several factors combined to make this firm a success story: quality improvements and design adjustments to satisfy tastes in the United States, low cost of labor, international aid involvement and government support, superior technology (training center), and Joy’s personal experience in working in the United States. All these combined to help Gahaya overcome the typical problems that other handicraft firms face when trying to sell internationally. The USADF and USAID funding, together with the founders experience, helped overcome technological problems: challenges of using the Internet, setting up modern accounting and payments systems, and so forth. Joy and Janet speak English fluently, so language barriers are not a problem. The designs are modified and quality upgraded to meet tastes in the United States. The training center keeps quality control. Their products are particularly durable, and carefully packed in the center of Kigali, so shipping problems are minimized. Since weaving was already a basic technique used by many women in Rwanda, the human capital investment is minimized (although techniques and designs are modified), and combines traditional skills with modern business practices.

African Export Successes

337

The Passion of an African Entrepreneur: Good African Coffee from Uganda Although we have described the success of Good African Coffee as a case in which quality upgrading was the key determinant, there is another critical factor: the strong entrepreneurial spirit, commitment, and passion of the founder of Good African Coffee (GAC), Andrew Rugasira. Only after fourteen (!) trips to the United Kingdom and many more meetings with distributors over two-and-a-half years, a contract was signed in 2005 and GAC started to sell to the supermarket chain Waitrose. Contracts with British supermarket chains Sainsbury’s and Tesco followed. Rugasira explains that one of the major obstacles he faced is prejudice against the black African entrepreneur. Convincing buyers that roasting and packaging in Kampala is a safe mode of operation (versus the facility in Dublin) proved to be a challenge when opening the plant there. Coffee is eligible for duty-free access, but this is not the point. Rugasira claims that nontariff barriers matter more. The difficulty of obtaining a business visa as an African entrepreneur and negative prejudice toward the black African entrepreneur in the West are much more problematic. This perception is very detrimental in penetrating markets. Rugasira had to work very hard (and still does) to convince buyers in the West that he is personally reliable, and that he can supply products reliably. It is Rugasira’s passion that keeps him involved in coffee exporting, rather than moving into other more lucrative domestic activities. Luck: Nile Perch Exporting from Lake Victoria Sometimes an entrepreneur discovers a Big Hit by chance. This was the case of Nile perch exporting from Lake Victoria. This industry was started by Harko Bhagat in Tanzania. Bhagat received his BS degree in chemical engineering in Canada before returning to Tanzania. Before starting to work in the seafood industry, he worked for a publishing company in Dar es Salaam. At some point a businessman he knew (not family, an acquaintance) in London asked Bhagat whether he could supply prawns from Tanzania, where they are abundant and labor is cheap. This encouraged Bhagat to start his own business. This was a fairly safe bet, since he had a significant client, and he soon realized that there are large margins in exporting prawns. And this is how he entered business: by chance. After some time exporting seafood, Bhagat learned (word of mouth) in 1992–1993 that there is a shortage of white fish in European and US markets. Following some research, he realized that this is potentially a huge market. Fishing was always done in Lake Victoria, so the potential to harvest fish in the lake was there. After securing a customer in Europe, he founded Vicfish Ltd. and built his own fish-processing plant (five tons per day), using his own capital (although that initial buyer eventually failed to buy). Once other importers of fish in Europe heard about the high quality and competitive

338

William Easterly and Ariell Reshef

price of the product, the business took off quickly and others started their own fish-processing plants. Today Vicfish has a 100-ton-per-day capacity. Initially the exports were frozen fish. The jump in business came following harmonization with European fish-processing plants in 1996–1997: this allowed them to export fresh chilled fish. It took some effort by Bhagat, as head of the fish processors association, to convince other producers of the importance of the harmonization and to make the necessary investments (he recalls complaints of lack of proper infrastructure).24 Eventually, the harmonization took place and this gave the industry its big push. Cash flow went up because for frozen fish the turnover time is ninety days, whereas for chilled fish it is less than a week. The product is sold and packed so that it can go directly to the shelf in supermarkets, as well as to restaurants. Cost Shocks Can Reverse a Big Hit: Cut Flower Exports from Uganda The case of cut flower exports from Uganda shows the sensitivity to cost shocks. Roses were grown (at high altitudes) and then cut and bundled and flown to Amsterdam. This was a booming business until oil prices increased and made most operations in Uganda nonviable after 2003. Incidentally, the same industry in Ethiopia is doing just fine and their boom started in 2003. The reason is twofold: first, the flower bulbs from Ethiopia are larger (because Ethiopia grows them at a higher altitude than Uganda), so their value is commensurately higher. Second, and perhaps more importantly, the government provides subsidies and foreign aid supports the industry in Ethiopia. In Uganda there are no such subsidies or aid for the flower industry.25 Demand for flowers in Europe did not decline due to the cost shock, only the suppliers changed. Thus, cost shocks can reverse a hit; but if you have government subsidies and foreign aid, then one can turn others’ reversal into one’s own hit. The Ruparelia Group, based in Kampala, has one cut-flowers exporting company. But their cut-flowers business completely collapsed, as well as others in the industry in Uganda; this is evident in figure 8.16A (see figure 8.16B for exports in Ethiopia). Interestingly, exports of cuttings (potted plants) and live plants continue to boom (not an activity of the Ruparelia Group), despite the increase in transportation costs. Additional Determinants of Success Here we briefly list a few factors that were instrumental in the success some of the remaining firms. Entrepreneurship of the founders of the chilled fish 24. Bhagat founded the Lake Victoria Fish Processors Association in Tanzania and has recently returned to head it. Through a deal with the governments of Tanzania, Uganda, and Kenya, violators are sanctioned. This is the only example of this kind of self-police, worldwide. 25. The Uganda government has deregulated industry over the last fifteen years, so it is not involved in subsidizing any industry there, except for soft loans to farmers in the most wretched conditions.

Fig. 8.16A

Exports of flowers, cuttings, and live plants (Uganda)

Fig. 8.16B

Exports of flowers, cuttings, and live plants (Ethiopia)

340

William Easterly and Ariell Reshef

exporter Lake Bounty led to its creation: the founders seized the opportunity to capitalize on their previous employer’s failure to respond to market conditions. Product innovation is critical for the success of the plastics exporters Jambo Plastics and Cello Industries in Tanzania. Both firms constantly introduce new products, based on information from buyers and on market research. A government loan (financed by Japan) helped start Murzah Oil Mills Ltd. in Tanzania. On the other hand, Bidco Oil and Soap Ltd. belongs to a group of companies that started in Kenya. 8.6

Conclusions

In this chapter we demonstrate (subject to concerns about data quality) that (a) exports are characterized by Big Hits, but (b) the Hits do not stay the same from one period to the next, and (c) this change is not explained by anything obvious like global commodity prices. The stylized facts that we establish do not reflect the traditional view that sees African commodity exports as a passive endowment, with changes driven mostly by global commodity prices. In our case studies, we find that new exports emerge due to quality upgrading, finding new areas of comparative advantage, regional trade liberalization (which makes exporting some products viable), managing to understand what is demanded in United States and European Union markets, personal connections and personal experiences that expose entrepreneurs to new technologies, and knowledge of markets. However, there are many idiosyncratic factors at work in each success also. Some of our successes occur in areas that are usually unsuccessful. Luck, entrepreneurial drive, and unexpected cost shocks play a role as well. Moreover, even the more conventional paths to success described above probably play at most a loose general role in guiding the entrepreneur. The stylized facts and the case studies match in providing a picture of export success as a very uncertain voyage of discovery. This picture of African exports could suggest the advantages of a flexible and decentralized system for continually making these discoveries, while sometimes also succeeding in perpetuating the success of old exports. A system that might fit the bill is private entrepreneurs operating in a relatively free market, just as much in Africa as in the rest of the world.

References Arkolakis, Costas. 2008. “Market Penetration Costs and the New Consumers Margin in International Trade.” NBER Working Paper no. 142014, Cambridge, MA.

African Export Successes

341

Arkolakis, Costas, and Marc-Andreas Muendler. 2008. “The Extensive Margin of Exporting Goods: A Firm-Level Analysis.” Working paper, Princeton University. http://www.princeton.edu/~ies/IESWorkshop/ArkolakisMuendlerPaper.pdf. Artopoulos, Alejandro, Daniel Friel, and Juan Carlos Hallak. 2010. “Challenges of Exporting Differentiated Products to Developed Countries: The Case of SMEdominated Sectors in a Semi-Industrialized Country.” IDB Working Paper no. 48, Inter-American Development Bank. Baldwin, Richard, and James Harrigan. 2011. “Zeros, Quality, and Space: Trade Theory and Trade Evidence.” American Economic Journal: Microeconomics 3 (2): 60–88. Bernard, Andrew, and Bradford Jensen. 1999. “Exceptional Exporter Performance: Cause, Effects, or Both?” Journal of International Economics 47 (1): 1–26. Bernard, Andrew, Bradford Jensen, Stephen Redding, and Peter Schott. 2007. “Firms in International Trade.” Journal of Economic Perspectives 21 (3): 105–30. Besedes, Tibor, and Thomas J. Prusa. 2007. “The Role of Extensive and Intensive Margins and Export Growth.” NBER Working Paper no. 13628, Cambridge, MA. Cadot, Olivier, Leonardo Iacovone, Ferdinand Rauch, and Denisse Pierola. 2011. “Success and Failure of African Exporters.” Policy Research Working Paper no. WPS 5657, Washington, DC. World Bank. Clerides, Sofronis K., Saul Lach, and James R. Tybout. 1998. “Is Learning by Exporting Important? Micro-Dynamic Evidence from Colombia, Mexico, and Morocco.” Quarterly Journal of Economics 113 (3): 903–47. Eaton, Jonathan, Samuel Kortum, and Francis Kramarz. 2008. “An Anatomy of International Trade: Evidence from French Firms.” NBER Working Paper no. 14610, Cambridge, MA. Eaton, Jonathan, Marcela Eslava, Maurice Kugler, and James Tybout. 2007. “Export Dynamics in Colombia: Firm-Level Evidence.” NBER Working paper no. 13531, Cambridge, MA. Easterly, William. 2003. “Can Foreign Aid Buy Growth?” Journal of Economic Perspectives 17 (3): 23–48. Easterly, William, and Ariell Reshef. 2009. “Big Hits in Manufacturing Exports and Development.” Working Paper, Forum for Research on Empirical International Trade. http://www.freit.org/WorkingPapers/Papers/TradePatterns/FREIT073 .pdf. Freund, Caroline, and Martha Denisse Pierola. 2010. “Export Entrepreneurs: Evidence from Peru.” Policy Research Working Paper no. 5407, Washington, DC, World Bank. Freund, Caroline, and Nadia Rocha. 2010. “What Constrains Africa’s Exports?” Policy Research Working Paper no. 5184, Washington, DC, World Bank. Hausmann, Ricardo, and Dani Rodrik. 2006. “Doomed to Choose: Industrial Policy as Predicament.” Paper prepared for the first Blue Sky seminar organized by the Center for International Development at Harvard University on September 9, 2006. https://www.researchgate.net/publication/239584546_Doomed_to _Choose_Industrial_Policy_as_Predicament. Melitz, Marc. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.” Econometrica 71:1695–725. Rajan, Raguhuram G., and Arvind Subramanian. 2005. “Aid and Growth: What Does the Cross-Country Evidence Really Show?” IMF Working Paper no. WP/ 05/127, International Monetary Fund. http://www.imf.org/external/pubs/ft/wp /2005/wp05127.pdf.

342

William Easterly and Ariell Reshef

Rauch, James, E., and Joel Watson. 2003. “Starting Small in an Unfamiliar Environment.” International Journal of Industrial Organization 21:1021–42. Roberts, Mark J., and James R. Tybout. 1997. “What Make Exports Boom?” Directions in Development Monograph Series, Washington, DC, World Bank. http:// documents.worldbank.org/curated/en/1997/01/695019/makes-exports-boom. Tybout, James R. 2000. “Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?” Journal of Economic Literature 38 (1): 11–44.

9

AGOA Rules The Intended and Unintended Consequences of Special Fabric Provisions Lawrence Edwards and Robert Z. Lawrence

9.1

Introduction

The export performance of the small, landlocked nation of Lesotho is an African success story that demonstrates both the power and limitations of trade preferences. In 2004, just three years after Lesotho became eligible for preferences under the Africa Growth and Opportunities Act (AGOA), the clothing exports to the United States from one of Africa’s poorest landlocked nations had trebled to reach $460 million and provide employment for over 50,000 workers (Bennet 2006). The performance of Lesotho and several other preference recipients was particularly striking because it seemed to contradict the pessimistic verdict many had reached about Africa’s capacity to become a globally competitive exporter of manufactured products even when granted preferential market access.1 Lawrence Edwards is a professor in the School of Economics at the University of Cape Town. Robert Z. Lawrence is Albert L. Williams Professor of International Trade and Investment at the Kennedy School of Government at Harvard University and a research associate of the National Bureau of Economic Research. We are grateful to Mark Bennet for his invaluable assistance on our visit to Lesotho; to the various people we interviewed including Nkopane Monyane, Jennifer Chen, Grace Lin, Mpho Madia, Fiona Lee, and those from the Lesotho National Development Corporation; to Mike Morris for helpful conversations; to Pandey Bibek and Jenny O’ Connell for research assistance; and for funding to the NBER Africa project. We also thank the South African National Research Foundation, Economic Research Southern Africa, and the Center for International Development at Harvard for hosting our visits to work together on this project. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13364.ack. 1. Several studies have been devoted to explaining this poor performance, and most conclude that the problems lie with the African countries themselves, rather than on the access given their products in foreign markets. A host of inhibiting factors have been identified (Ng and Yeats 1996; Wang and Winters 1998).

343

344

Lawrence Edwards and Robert Z. Lawrence

On May 12, 2010, a ceremony was held on Capitol Hill in Washington, DC, to celebrate the tenth anniversary of AGOA. In his remarks at the gathering, United States Trade Representative Ron Kirk credited AGOA with “a substantial increase in two-way US-Africa trade since 2000, with African countries now exporting to the United States a more diverse range of value-added products.” Kirk also asserted that the trade program “powerfully demonstrates the link between trade and economic development.”2 In this chapter we will provide some evidence that supports Kirk’s positive verdict: AGOA has stimulated exports of manufactured products, especially clothing, but we will also suggest that the ultimate impact on economic development has been quite disappointing. We will argue that both the success and limitations are the predictable consequences of the manner in which the preferences have been constructed. We will show that although these preferences encourage exports, they simultaneously create disincentives for local value addition that may limit the program’s development benefits. 9.2

Background

As indicated by Mr. Kirk’s remarks, trade preferences are of interest not only because they might provide one-time benefits in the form of higher incomes and increased employment, but also because trade is associated with more dynamic benefits that lead to faster growth. Economic growth is an ever-expanding process in which actors not only replicate what they were doing on a greater scale, but continuously develop new capabilities that allow them to produce increasingly sophisticated goods and services (Hausmann, Hwang, and Rodrik 2007). More developed countries typically produce higher unit-value products and wider ranges of products than their less developed counterparts (Schott 2004). These products often face less elastic demands and provide higher profit margins than more standardized, commodity-like products. If they can “learn by doing” by using trade preferences, it is hoped that firms that start by exporting a few simple products can upgrade their product sophistication and diversify into other products and markets, and ultimately become competitors that no longer need preferential treatment.3 In addition, it is hoped that there are benefits to the rest of the economy. Other domestic firms could gain too through backward and forward linkages as exporters demand inputs and services and become increasingly embedded in the local economy. During the industrial revolution this form of development was evident in the textile industry, which was an important driver of industrialization. 2. http://www.america .gov/ st/ business-english/ 2010/ May/ 20100513122443SztiwomoD0 .8958856.html. 3. According to Hwang (2007) there is unconditional convergence at the six-digit level. If countries start to produce low-unit-value goods within a product category, they will eventually experience significant increases in their unit values. The claim is that this will happen more or less automatically, without any special supportive policies in place.

AGOA Rules

345

Japan, and later Korea, Hong Kong, and other dynamic Asians also all cut their teeth as exporters of clothing, continuously upgrading and diversifying (Gereffi 1999). Motivated in part by such considerations, the European Union (EU) and the United States both implemented multilateral generalized special preferences (GSP) programs in the 1970s. In addition, they both have regionally focused preferential programs.4 Yet, the notion that developed country markets are open to manufactured exports from least developed economies as a result of these concessions can be challenged. It is difficult for underdeveloped countries to produce complete complex products, but they are often quite capable of providing simple assembly operations. Some of the preferences given through programs are thus a sham because they include rules of origin that require more local production than these poor countries can provide. These rules are generally justified as necessary to prevent the trade deflection that could occur if products are imported from third countries and then, with little additional value added, claimed as originating from preference recipients—a practice sometimes known as “screwdriver plants.” This is a legitimate concern, but the rules are more constraining than strictly necessary, and they inhibit poor countries from specializing in the narrow slices of global production chains in which their comparative advantage is likely to lie. In the case of preference programs in apparel, these rules are particularly stringent, generally requiring at least two (in the case of the EU) or even (in the case of the United States) three transformation processes (e.g., yarn, fabric, assembly) in the preference-receiving or granting countries to qualify for duty-free entry. (For an excellent account, see Ahmad [2007]) These rules are especially problematic because fabric production is a highly capital- and technology-intensive activity that is beyond the capabilities of most very poor countries. The rules of the US AGOA program are, however, an important exception, indeed perhaps the exception that proves the rule. The AGOA not only gave all sub-Saharan countries extensive duty-free, quota-free access to the United States (table 9.1),5 but its rules of origin also contained an 4. The EU granted African, Caribbean, and Pacific (ACP) countries special preferences, first under the Lomé Conventions (starting in 1976) and later through the Contonou Agreement (2000). More recently the EU has concluded Economic Partnership Agreements (EPAs) with groups of ACP countries. The US has granted special preferences under the Caribbean Basin Initiative (CBI), the Andean Promotion Act, and AGOA. Preferences for least developed countries (LDCs) have received special attention. In 2001, the European Union introduced an “Everything but Arms” (EBA) program that provides LDC exports duty-free, quota-free access. In the Doha Round negotiations, the United States agreed to give duty-free access to LDCs in 97 percent of its tariff lines. 5. In May 2000, the US Congress passed AGOA. The Act granted duty-free access for 4,600 GSP tariff-line items plus another 1,800 tariff-line items not on the original GSP. This meant that, aside from some apparel and agricultural products, AGOA beneficiaries could export almost any product to the United States duty free. The AGOA preferences for garments required that they are made of 85 percent US-made yarn and fabric or from fabrics and yarns made in other AGOA beneficiary countries.

346 Table 9.1

Lawrence Edwards and Robert Z. Lawrence Summary of apparel rules of origin under AGOA

Description of the rules of origin requirements

Conditions of access

1. Apparel made from US yarns or fabric 2. Apparel assembled from regional fabric from United States or African yarn 3. Apparel assembled in a lesser-developed country using foreign fabric or yarn 4. Certain cashmere and merino wool sweaters 5. Apparel made of yarns and fabrics not produced in commercial quantities in the United States 6. Eligible hand loomed, handmade, or folklore articles and ethnic printed fabrics

Unrestricted Subject to tariff rate quota cap (currently 6.43675 percent to 2015) Unrestricted for four years, but extended to 2012 (cap of 3.5 percent of US imports) Unrestricted for selected products Unrestricted

Unrestricted for selected products from Dec. 2006 under AGOA IV

Note: Unrestricted implies duty-free and quota-free treatment.

unusual waiver for wearing apparel that was granted to “lesser developed beneficiary countries” (LDBCs). Subject to fairly generous market-share caps that have not been binding, the waiver allowed these LDBC countries to use third-country fabrics or yarn and still export clothing under the AGOA preferences.6 Instead of requiring individual items to meet specific transformation rules, such as minimum value-added requirements or the use of domestic fabric, the United States set up a simple inspection program that verified that genuine production activities were taking place.7 Although the special LDBC rule was originally scheduled to expire after three years, it was extended in 2004 for another three years and in 2007 for a further five. Countries not defined as “lesser developed” such as South Africa and Mauritius did receive AGOA preferences, but they were required to meet GSP rules of origin that for clothing required the use of US or regional yarns or fabric. Because the different treatment for higher-income countries provides a useful control group, AGOA provides an ideal opportunity to explore the role of different types of rules of origin in preferential arrangements, and the experience demonstrates how important they can be: US imports of clothing from AGOA countries (SITC 84- Apparel and Clothing Accessories) increased from $730 million in 2000 to $1755 million in 2004. This growth was dominated by US imports of clothing from the least developed 6. Most of the countries that were eligible for the waiver are classified as least developed by the United Nations. Botswana and Namibia did not meet the requirements for the special rule as their GDP per capita exceed the minimum of US$1,500 in 1998. However, they were designated as LDC countries under amendments to the AGOA Act in 2002 (AGOA II) and 2004 (AGOA IV). Mauritius was temporarily granted the third-country fabric derogation from October 2004–September 2005 under the Miscellaneous Tariff Bill of 2004 (known as AGOA III). More recently, Mauritius qualified for the third-country fabric derogation in November 2008 for a period of four years. 7. The AGOA privileges also require protecting US intellectual property rights, observing labor rights, proving access to US trade and investment, and implementing rule of law. Apparel exports require adopting an effective visa system to prevent transshipment.

AGOA Rules

Fig. 9.1

347

US apparel imports from AGOA countries according to import program

Source: USITC.

African countries, which increased by 400 percent, almost all of which took advantage of the lesser-developed country provision (see figure 9.1). The largest growth in exports between 2000 and 2004 came from Lesotho (up from $140 million to $456 million), and over the same period very significant increases also occurred in Kenya (up from $43 million to $270 million), Madagascar ($110 million to $323 million), Swaziland ($32 million to $179 million), and Namibia (0 to $79 million) (figure 9.2). By contrast, in 2004 US imports of clothing from South Africa and Mauritius, the two largest African clothing exporters when AGOA was passed, were actually 18 million dollars lower than they had been in 2000 (figure 9.2). The AGOA also stimulated entry into new clothing markets. Table 9.2 reports the number of HTS ten-digit apparel products produced by AGOA countries. Overall, AGOA countries export limited ranges of apparel products. South Africa, Mauritius, and Madagascar had the widest range of products (over 130 each) prior to the implementation of AGOA in 2000. The AGOA preferences increased product penetration. Many countries experienced exceptional increases in the total number of lines from 2000 to 2004 (see Kenya from 45 to 155, Swaziland from 47 to 139, and Lesotho from 60 to 118).8 In most countries, however, these trends reversed after 2005, but still remained above 2000 levels. The AGOA countries have experienced setbacks, however, first when the constraints on their (mainly Asian) competitors were lifted with the expiration of the Multi-Fiber Arrangement (MFA) in 2005, and second with the slump in the United States because of the global financial crisis.9 As a result, 8. The largest contractions in Lesotho occurred in firms producing knitted garments; those producing woven garments (e.g., denim) did better. (See Bennet 2006.) 9. In July 2007 Lesotho Clothing and Applied Workers Union estimated employment at 44,000 compared to 55,000 in 2004.

348

Fig. 9.2

Lawrence Edwards and Robert Z. Lawrence

US imports of clothing and textiles for selected AGOA recipients

Source: USITC.

US imports declined, although for the least developed AGOA countries still remained three times as large as in 2000. By contrast, despite AGOA, imports from South Africa and Mauritius combined were decimated, and in 2008 were only a third of their 2000 levels. Several research papers have confirmed what is obvious to the naked eye—that the lesser-developed country provisions have played a key role in the outcomes. Using a variety of methodologies, empirical estimates confirm that preferences under AGOA are a significant determinant of apparel exports: Mattoo, Roy, and Subramanian (2003) stressed the role of rules of origin in limiting the overall benefits from AGOA to all recipients. Collier and Venables (2007) find that the AGOA apparel provision had a positive and significant effect. Frazer and Van Biesebroeck (2010) find that the AGOA had a “large and robust effect that grew over time” and estimate that overall AGOA apparel exports increased by 53 percent with stronger impacts on products with high initial levels of protection. Portugal-Perez (2008) reports an impact of 96 percent for twenty-two countries eligible for the third-country fabric provision, and 303 percent for the top seven beneficiaries.10 In addition to higher export volumes, there is also evidence that AGOA exporters enjoyed higher prices and captured some of the tariff rents created by the preferences (Olarreaga and Özden 2005). Apparently, whatever Africa’s handicaps, they have not prevented substantial responses: 10. Other studies include Brenton and Ikezuki (2005), Gibbon (2003), Seyoum (2007), Nouve (2005), Rolfe and Woodward (2005), and FIAS (2006).

Table 9.2

Products traded (out of approx. 1,500 possible products), sorted by 2004

Eligibility Apparel eligible Apparel eligible, LDC special rule

Nonapparel eligible

Country Mauritius South Africa Benin Botswana Burkina Cameroon Cape Verde Chad Ethiopia Ghana Kenya Lesotho Madagascar Malawi Mali Mozambique Namibia Niger Nigeria Rwanda Senegal Sierra Leone Swaziland Tanzania Uganda Zambia Angola Burundi Comoros Congo (Brazzaville) Congo (Kinshasa) Djibouti Gabon Gambia Guinea Guinea-Bissau Liberia Sao Tome and Principe Seychelles Togo All AGOA countries

Possible products

1996

2000

2004

2008

165 136 2 14 8 10 2 0 9 38 55 41 38 2 10 3 0 4 61 0 31 2 21 4 0 1 0 1 1 0 3 0 1 6 5 0 2 1 0 13

139 267 2 24 9 7 4 0 4 52 45 60 175 22 10 0 1 4 47 0 20 28 47 6 0 1 0 1 0 0 4 0 1 11 12 0 3 1 2 4

135 318 4 57 9 14 14 1 41 63 155 118 236 45 12 7 40 7 39 2 10 45 139 24 9 4 0 0 1 3 1 0 3 7 13 0 2 0 3 3

139 177 0 18 4 18 5 0 79 48 117 84 259 25 11 0 2 5 33 5 16 54 86 16 4 4 0 0 0 0 3 0 0 9 12 3 3 0 6 4

323

439

537

465

1,548

1,533

1,525

1,515

Notes: We use the Pierce and Schott (2009) concordance program to construct a HS ten-digit, time-consistent classification for the full period.

350

Lawrence Edwards and Robert Z. Lawrence

indeed, there is no evidence of differential effects in taking advantage of AGOA based on measures of corruption or institutional quality (Frazer and Van Biesebroeck 2010). Despite the impressive growth in volumes, there is also some disquieting evidence in AGOA’s performance that relates to the issue of dynamic benefits. Decompositions of output growth reported in table 9.3 reveal that the export of new product lines (the extensive margin) contributed only 30 percent of total AGOA import growth from LDC special rule countries between 2000 and 2004, and 42 percent of the decline from 2004–2008. Strikingly, only 8 percent of the growth in Lesotho’s apparel exports took the form of new products. The share of product lines accounted for by the top four and top ten HS ten-digit products is around 60 and 80 percent and has remained fairly constant throughout the period. In addition, production is predominantly CMT (cut-make-trim) with little value addition and there is little evidence of dynamic spillovers to other sectors of the economies (Lall 2005). These trends are exemplified by the development of Lesotho’s clothing industry in response to the AGOA preferences. Therefore, before presenting and testing a theory that can explain these outcomes, we describe briefly the history of Lesotho’s clothing industry. 9.2.1

Lesotho

As the largest apparel exporter to the United States, Lesotho is of particular interest. Whereas some countries such as Namibia, Malawi, and Botswana became clothing exporters for the first time after AGOA, the response of Lesotho actually built on a longer historical experience in which trade preferences and policies also played an important part. The industry was launched in the 1980s when Taiwanese manufacturers, originally based in South Africa, moved to Lesotho in order to avoid trade sanctions imposed by the United States and Europe on what was then the apartheid regime. More investors were attracted in the late 1980s, after the European Union signed the Lomé Convention, which granted special preferences to the African, Caribbean, and Pacific (ACP) countries that had formerly been colonies. While the clothing preferences in Lomé had a double transformation rule, Lesotho was granted a temporary derogation from the requirement allowing it to use third-country fabrics that the investors took advantage of. When the derogation expired in the mid-1990s, exports to Europe plunged and they have never recovered (figure 9.3). This experience provided the first demonstration of the importance of the role of these special preferences in the viability of Lesotho’s exports of clothing. Clothing exports to the United States were subject to tariffs but were also constrained by quota restrictions under the MFA. As these became increasingly binding on others, Lesotho’s foreign-owned firms shifted to exporting to the United States to take advantage of its unfilled quotas. AGOA countries’ concentration of exports in

Apparel eligible, LDC special rule

Table 9.3

Benin Botswana Burkina Faso Cameroon Cape Verde Chad Ethiopia Ghana Kenya Lesotho Madagascar Malawi Mali Mozambique Namibia Niger Nigeria Rwanda Senegal Sierra Leone Swaziland Tanzania Uganda Zambia

0.00 0.24 –0.02 0.00 0.33 0.00 0.00 0.02 0.68 0.92 0.78 0.51 1.19 0.00 0.00 –0.82 1.90 0.00 0.73 0.40 0.57 0.24 0.00 0.00

Contribution intensive growth 1.00 0.76 1.02 1.00 0.67 1.00 1.00 0.98 0.32 0.08 0.22 0.49 –0.19 1.00 1.00 1.82 –0.90 1.00 0.27 0.60 0.43 0.76 1.00 1.00

Contribution extensive growth

2000–2004

–0.42

–0.30 0.59 0.54 1.80

3.69 0.11 –0.07

4.00 1.18 0.59 0.34 0.31 0.38 –0.17

0.33 0.27 0.28 0.22 0.36

Average annual growth (US$) 0.00 –1.56 0.00 –0.27 0.00 0.00 0.52 0.95 0.81 0.87 0.91 0.32 0.90 0.00 0.07 0.44 –0.10 0.00 0.65 0.19 0.63 –0.25 –0.04 0.00

Contribution intensive growth

Decomposition of growth in US apparel imports: Extensive and intensive growth

1.00 2.56 1.00 1.27 1.00 1.00 0.48 0.05 0.19 0.13 0.09 0.68 0.10 1.00 0.93 0.56 1.10 1.00 0.35 0.81 0.37 1.25 1.04 1.00

Contribution extensive growth –1.00 –0.06 –0.49 0.21 –0.73 –1.00 0.30 –0.41 –0.03 –0.07 –0.04 –0.17 0.37 –1.00 –0.94 0.18 –0.08 0.92 0.16 –0.39 –0.09 –0.12 –0.44 –0.52

Average annual growth (US$)

2004–2008

0.00 0.01 0.01 0.01 0.01 0.01 0.02 0.02 0.18 0.44 0.62 0.64 0.64 0.64 0.68 0.68 0.68 0.68 0.68 0.68 0.79 0.79 0.79 0.79 (continued)

Cumulative imports 2004

0.68

All AGOA 0.42 –0.01 0.04

0.25

–0.34 –1.00 –0.64 –0.18

2.56 –0.14 –0.42

–0.19

–0.02 0.00 –1.00

Average annual growth (US$)

1.00 1.00

0.00 0.00 0.58 0.87 –0.18

0.42 0.13 1.18

0.30

1.00 1.00 1.00 1.00 0.10 –0.01 1.00 1.00

0.00 0.00 0.00 0.00 0.90 1.01 0.00 0.00

0.70

0.06 0.20

Contribution extensive growth

0.94 0.80

Contribution intensive growth

–0.07 –0.25 –0.21

–0.10

0.97 0.47

0.12

–1.00 1.45 –1.00 –1.00 0.32 0.11

–0.18 –0.40

Average annual growth (US$)

2004–2008

0.92 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Cumulative imports 2004

Note: Mauritius is treated as not eligible to export under LDC special rule, despite being granted temporary LDC status from October 2004–September 2005 under the Miscellaneous Tariff Bill of 2004 (known as AGOA III).

0.31 –0.05 3.03

0.32

1.00 1.00 0.05 1.00

0.00 0.00 0.95 0.00 0.69 1.05 –2.03

0.33 18.67 1.00 1.00 1.00 1.00 1.00 1.32 1.02

Contribution extensive growth

0.67 –17.67 0.00 0.00 0.00 0.00 0.00 –0.32 –0.02

Contribution intensive growth

2000–2004

Mauritius South Africa Burundi Comoros Congo (DROC) Congo (ROC) Gabon Gambia Guinea Guinea-Bissau Liberia Sao Tome & Principe Seychelles Togo

(continued)

LDC special rule eligible Other apparel eligible Other AGOA

Apparel eligible Nonapparel eligible

Table 9.3

AGOA Rules

353

products where quota constraints on Chinese exports were binding is clearly revealed in figure 9.4. Thus, even prior to the passage of AGOA, firms based in Lesotho, most of which were subsidiaries of Asian multinationals, were exporting to the United States. Indeed, after 1999, 99 percent of all Lesotho’s apparel exports went to the United States with only 0.8 percent going to South Africa and just 0.2 percent to the EU.

Fig. 9.3

Apparel exports to the EU 15, selected AGOA countries

Notes: Own calculations using data from Eurostat (http://epp.eurostat.ec.europa.eu/newxt web).

Fig. 9.4 AGOA apparel exports to United States according to Chinese quota fill rates Notes: Quota fill rates are obtained from OTEXA (http://otexa.ita.doc.gov/). Quotas on product lines are assumed binding if the 2003 Chinese fill rate is greater than or equal to 90 percent.

354

Lawrence Edwards and Robert Z. Lawrence

The small share of Lesotho’s exports going to South Africa also indicates the important role played by fabric rules of origin. The US most favored nation (MFN) tariff on clothing is around 17 percent, while the Southern African Customs Union (SACU) tariff is about 40 percent. Thus, garments exported to South Africa from Lesotho (which is within the customs union) have a much larger margin of preference. Yet Lesotho is far more competitive in the United States than in SACU.11 The reason is that to sell in South Africa, Lesotho has to pay SACU tariffs or SACU prices for fabric. By contrast, under AGOA it obtains these duty free.12 The AGOA has been in effect for a decade, but there is little evidence that much of Lesotho’s industry could survive without preferences or that it has diversified horizontally into new products and markets or vertically into greater domestic value addition. Factories in Lesotho continue to concentrate on just a narrow range of garments: the most basic low-unit-value categories of knitted tee-shorts, slacks, blouses, and blue jeans. The slice of the production chain they participate in is narrow and does not seem to be expanding. Most apparel manufacturing in Lesotho is CMT (cut-maketrim). The firms, almost entirely foreign owned, typically provide assembly, packaging, and shipping services and depend on their Asian headquarters to generate orders, design the clothes, and send them the fabric they need. This can be seen by comparing the industry wage bill for 50,000 workers (approximately $1,000 per worker), that is, $50 million in 2004 with total US exports valued at $456 million. Most of the value is thus added to other parts of the chain. Almost none of the managers are locals and the buyers of fabric and the marketers of the garments and the key strategic corporate decisions are all made thousands of miles away in Asia. The local production process is characterized by highly routine steps used to produce very large volumes. Just one buyer—US retailer The GAP— accounts for almost 40 percent of overall output. The combination of the large scale on which they operate and the large orders by concentrated buyers makes it difficult for small firms to enter the market. In addition, to move up the value chain and to produce differentiated products in smaller batches 11. Indeed, according to Sandrey et al. (2005), Lesotho cannot even compete in Lesotho! “Examination of the local clothing retail outlets reveals a predominance of both Chinese and South African garments.” 12. To be sure, factors besides favorable rules of origin have contributed to Lesotho’s performance (Maloney 2006). These include fluctuations in the Rand to which its currency is tied (favorable between 2000 and 2002) and other policies to assist exporters by the Lesotho government. In addition, Lesotho has benefited from a favorable international image as a non-sweatshop producer (Seidman 2009). It has also been promoted by Bono in his campaign against AIDS. The Lesotho National Development Corporation (LNDC) has played an active role, offering favorable rents for factory shells. The government also provided generous tax treatment—reduced from 15 to 0 in 2006—and sought to maintain industrial peace with a Directorate of Dispute Prevention and Resolution. The government has used the Duty-CreditCertificate Scheme of the South African Customs Union that gives apparel firms between 10 and 25 percent of the free on board (FOB) value of their exports in certificates that allow them to import textiles or apparel duty free.

AGOA Rules

355

requires more skilled workers. This is part of the explanation for Lesotho’s inability to do well in the relatively small South African market in which demand is more varied. One firm in Lesotho has built a denim plant.13 But with this exception, all fabrics are imported. Lesotho and other AGOA countries, even South Africa, therefore, lack the domestic textile industry that would allow them to meet the regular clothing rules of origin in US preference programs. Lesotho’s workers have relatively low productivity levels and their skills do not appear to have increased over time.14 Lall ascribes the lack of improvement in part to the Labor Code Rule that prohibits the use of piece rate. He noted “Despite over a decade of clothing assembly, productivity in Lesotho still lags way behind major competitors. With similar wages, therefore, its competitiveness cannot outlast trade privileges unless productivity improves sufficiently to match competitors” (Lall 2005, 1016). The relatively low quality of Lesotho’s (and other AGOA) apparel exports is also revealed in the comparative price of its exports. Table 9.4 presents the average unit values of the top fifteen apparel products at HS ten-digit level exported by Lesotho to the United States in 2004. These unit values are compared against the average unit value of other lesser-developed AGOA countries and the 10th, 25th, median, 75th, and 90th percentile unit values of the 226 countries in the sample. In all but one case, when it is just below the 25th percentile, the unit values of Lesotho’s apparel exports fall between the 25th and 50th percentile range. What is also striking is the range of unit values even within these highly disaggregated product lines (see Schott 2004). For example, the 90th percentile unit value of a dozen women’s or girls’ cotton pullovers (Lesotho’s top apparel export) in 2004 was 280 dollars versus 31 dollars for Lesotho exports. The combination of a productivity disadvantage and almost no domestic textile industry makes the industry’s survival totally dependent on its preferences. Each time the expiration of the special rule has drawn near, therefore, studies have issued credible and dire warnings about the industry’s ability to survive without them (Salm et al. 2002; Bennet 2006). This experience makes it clear that trade need not automatically lead to growth and the manner in which trade is stimulated could well prove consequential for the amount and nature of the growth it stimulates. In particular it suggests that trade that is stimulated by preferences might well have different effects than trade that occurs for other reasons. Why this disappointment? Both Lall (2005) and Collier and Venables 13. In 2004, the industry faced a major challenge with the potential expiration of the special rule. Partly anticipating the expiration of the special rule in 2004, the Nien Hsing Group of Taiwan invested over $100 million to build the Formosa Mill, a state-of-the-art denim fabric mill. 14. Lall (2005) estimated that while Lesotho’s wages were similar to Asian levels, its productivity was typically only 50 percent of East Asian levels. According to Morris and Sedowski (2006), worker productivity has not increased over a ten-year period. (See also Morris 2006.)

 

37

70 35 60 50 70 52 69 63

31 36 90 68 87 71 36

Lesotho

Notes: The mean price for other AGOA is the exponent of the mean log price.

All products exported by Lesotho

6204624040 6110303050 6110202040 6104632011 6203424045 6203424060 6203424050 6204624055

Description

59

66 52 56 46 49 53 53 49

27 36 73 66 71 58 46

Other lesserdeveloped AGOA

22

43 30 41 28 26 25 27 34

26 24 48 42 52 39 34

10th

33

59 41 60 43 47 39 48 52

32 34 63 54 76 65 43

25th

60

73 68 77 60 67 55 74 67

46 60 100 68 92 93 57

Median

Percentiles

122

100 163 170 198 93 84 148 113

99 110 191 98 199 232 120

75th

Price (US$ per dozen) of top fifteen Lesotho products in terms of export value, ranked largest to smallest, 2004

Women’s/Girl’s other pullovers of cotton Men’s/Boys’ other pullovers of cotton Men’s trouser breeches, cotton, blue denim Boys’ trouser/breeches, cotton, blue denim Women’s trousers/breeches, cotton, blue denim Women’s trousers/breeches other cotton, not knit Women’s/girl’s other sweaters, man-made fibers Girls’ trousers, cotton, blue denim not imp. playsuit parts, not knit Men’s/boys’ other sweaters, man-made fibers, knit Men’s/boys’ sweatshirts of cotton Women’s trousers and breeches, synthetic fibers Boys’ trouser, etc., not cotton Boys’ shorts, cotton not playsuit parts, not knit Men’s shorts of cotton Women’s shorts of cotton

6110202075 6110202065 6203424010 6203424035 6204624010 6204624020 6110303055

HS

Table 9.4

299

470 287 306 409 243 207 288 337

280 222 465 269 449 384 406

90th

52 56 58 61 63 65 67 69

14 26 32 37 42 45 49

Cumulative trade share Lesotho (%)

AGOA Rules

357

(2007) suggest it may be that these AGOA countries are simply too underdeveloped for the exports to ignite the process. Collier and Venables argue it reflects a lack of complementary inputs that are required to exploit scale economies. They suggest that preferences are only likely to work if countries already have “the skills and infrastructure to be near the threshold of global manufacturing competitiveness” (1328). Lall also suggests that part of the explanation could lie with having foreign factory owners—most of whom are Taiwanese—that are not closely integrated into the local community. Ironically, this might suggest that these kinds of preferences should be given to the more advanced developing countries like South Africa rather than the least developed countries that have received them. In this chapter, however, we will explore a different explanation that has been overlooked in the literature. We will argue that both the positive and negative responses to AGOA are no accident. Indeed, they are the consequences that economic theory would lead us to expect, given the form in which the preferences have been granted. As we will show using the theory of effective protection, preferences combined with the third-country fabric rule can have powerful financial effects. They could easily be the equivalent of a subsidy to production that is two or three times higher than the 17 percent preference margin granted by AGOA through MFN tariff relief on clothing. This allows AGOA producers to offset cost disadvantages due to the lower productivity of their workers and greater distance from suppliers and markets and helps explain why the initial responses to AGOA (and the availability of unused MFA quotas) were so powerful. On the other hand, in theory the preferences also have two deleterious effects. First, they steer firms mainly toward the simplest products in which clothing producers add little value. Thus the preferences tax skills acquisition and discourage firms from moving up the value chain. Second, the preferences (and the MFA) discourage backward linkages because they induce exporters to use relatively expensive fabrics rather than the cheaper fabrics that are more likely to be produced in poor countries. In sum, trade preferences “work.” They can stimulate trade, raise incomes in developing countries, and boost employment. But whether they actually lead to development conceived of as a cumulative growth process is much less certain. In addition, changes in other trade policies at first helped and then hindered AGOA’s performance. On the one hand, the MFA initially provided an especially favorable environment for AGOA countries to produce lowunit-value products because it not only constrained their Asian competitors, but also induced these exporters to shift toward higher-quality products. On the other hand, when the MFA was removed, constrained countries such as China moved strongly into precisely the markets in which AGOA countries had specialized. Although AGOA helped the least developed African countries withstand this shock, they were nonetheless adversely affected.

358

Lawrence Edwards and Robert Z. Lawrence

This chapter proceeds now in three sections. In the first we discuss the economic theory of the effects these regimes are likely to have. In the second, we conduct several empirical tests of the theory, and in the final section we present our conclusions. 9.3

Theory

The overview of Lesotho’s export performance identified the influence of two trade policies: (a) the effect of MFA quotas and their removal, and (b) the effect of AGOA tariff preferences and rules of origin. In this section we draw on economic theory to investigate the impact of these policies. The focus is on incentives they create for the production and export of particular types of clothing products. We are particularly interested in the impact on product characteristics such as quality, fabric use, and value addition in recipient countries. We will show that the regime governing clothing trade can be expected to have a profound impact on clothing production choices in countries like Lesotho. In particular, we will demonstrate that the MFA not only provided a subsidy to Lesotho’s clothing exports but also created incentives for it to specialize in low-quality and low-valued-added products. The AGOA program provided an even more powerful incentive to expand exports of low-value-added clothing products, but it had an additional effect. The third-country fabric provision encouraged further specialization in clothing products with high fabric-cost shares. Some of the arguments we will use are not new. The body of literature on how trade policies influence product characteristics is well established in the case of quotas (Falvey 1979; Krishna 1987; Feenstra 1988) and transport costs (Alchian and Allen 1964; Hummels and Skiba 2004). The central result in this literature is that quotas and unit transport costs lead to quality upgrading, while tariffs do not. However, the literature generally assumes integrated production, and less studied are the effects of quotas and tariffs on the quality and value addition when products contain imported intermediate inputs. This analysis therefore explores how tariff preferences and their associated rules of origin lead to changes in the quality of goods produced and exported. We focus on clothing and fabric inputs, although some of the results would be applicable to other products that are manufactured with imported intermediate inputs. Most apparel firms located in Lesotho sell products to the United States through “full-package” intermediaries located in East Asia. These fullpackage suppliers compete with others for orders in the United States and Europe (Lall 2005). They then contract these out to their associated apparel producers, either through competitive bidding or through some allocation rule. Lesotho, for example, will export all products for which its production costs are lower than its competitor suppliers.

AGOA Rules

359

We are therefore going to develop a simple model that captures this intrafirm allocation and thus the relative shares of two countries, think Lesotho and China, in the US apparel imports. We model clothing as an array of products containing varying quantities of labor and fabric inputs. We determine patterns of value addition and specialization in a Ricardian framework that captures the effects of differences in fabric intensity. We then explore the effects of changes in market-access policies on production choices in both countries. In particular, we highlight how these policy changes affect both the volumes (intensive margin) and the types of products (the extensive margin) each of the countries will export. 9.3.1

Model

We assume apparel products are differentiated by type of product and the content of fabric.15 Each apparel product z is associated with a point on an interval [0,1] and is assembled using labor and fabric according to a constant returns to scale Leontief production function:16 y(z) = min [ L(z) a(z),F (z) u(z) ] a(z) is the labor used per unit output, L(z) is the quantity of labor, F(z) is the quantity of fabric, and θ(z) is the unit fabric requirement in square meters (the input-output coefficient). Using the cost function dual to the production function, the unit cost c(z) of clothing (assuming no transport costs) is given as: (1)

c(z) = a(z)w + u(z)PF (z)

where w is the wage and PF(z) is the price per square meter of fabric associated with product z. We also assume firms are competitive, so equilibrium profits are zero and the free on board price equals costs, that is, (2)

p(z) = c(z) = a(z)w + u(z)PF (z).

Products are therefore differentiated according to their unit labor requirements, as well as unit fabric costs, which are affected by the quantity and price of fabric used. For example, we would expect more complex apparel products (e.g., suits) to require more labor than simple products (e.g., T-shirts). Although we do not model quality specifically, we would also 15. Mattoo, Roy, and Subramanian (2003) develop an alternative model with decreasing returns and infinite demand to show how both tariff preferences and waivers of rules of origin increase exports of existing products. They do not deal with the impact on product quality, nor export of new varieties. 16. Portugal-Perez (2008) assumes a similar production function. A clear limitation of this model is that it does not take into account capital (sewing machines, fabric cutters, irons, washing and drying machines) used in the production of apparel. However, in a world where this type of capital is internationally mobile, it is the nontraded factors that become the primary determinant of a country’s comparative advantage (Wood and Mayer 2001).

360

Lawrence Edwards and Robert Z. Lawrence

expect higher-quality apparel to require more labor services and higherpriced fabric than lower-quality products. 9.3.2

The Allocation Decision

The allocation process depends on the relative cost of production across different locations. Lesotho will export all products for which its costs are less than or equal to those from China. This allocation condition can be specified as: (3)

a(z)w + u(z)PF (z) ≤ a ∗ (z)w∗ + u(z)PF ∗ (z),

where transport costs are assumed to be zero and * denotes foreign competitor (China). Under free trade where fabric is internationally traded and there are no differences in unit fabric costs (PF(z) = PF*(z)), we can respecify the relationship in terms of Lesotho’s relative unit labor cost (RULC): RULC(z) =

(4)

wa(z) ≤ 1. w∗a ∗ (z)

Here, home (Lesotho) exports all apparel products for which its unit labor costs are lower than its foreign competitors. Alternatively, China exports all apparel products for which its relative wages are less than or equal to its relative productivity. This result implies that comparative advantage when intermediate inputs can be obtained at world prices is entirely dependent on the relative effective price of the nontraded factor. This outcome is equivalent to that of the Dornbusch, Fischer, and Samuelson (DFS) (1977) Ricardian model with a continuum of goods. The fabric content of apparel, therefore, has no influence on what is produced by the home country. We will see later that this no longer holds once we introduce quotas and preferential trade barriers. 9.3.3

Consumption and Equilibrium

To close the model, we assume that there is no US apparel production, that countries only export apparel to the United States, and that US consumers have identical and homothetic preferences. Utility is Cobb-Douglas for a numeraire good (nonclothing products), but is a constant elasticity of substitution (CES) function in the quantities of the differentiated clothing products. The utility function is specified as:17 (5)

u=

( ∫ C (z)dz ) 1

r

0

ar

C01−a

0 < r < 1,

where C(z) denotes total US consumption of apparel products z and C0 is consumption of all other goods. The US consumers spend a constant frac17. See Dixit and Norman (1980, 282).

AGOA Rules

361

tion α of their income on apparel products with the remainder spent on the numeraire good. In addition, the differentiated apparel products are substitutes with a constant elasticity of substitution given by σ = 1/(1 – ρ) > 1.18 Since u is a separable utility function, the optimal choice of apparel products can be obtained by maximizing the CES component of the utility function subject to expenditure being less than or equal to αI where I is US income. Optimal demand by US consumers for each product z is given by ⎛ p(z) ⎞ C(z) = ⎜ ⎝ P ⎟⎠

(6)

−s

aI , P

where the price index of the CES quantity index is given by P = 1 (1−s) 1−s 1 dz . A rise in the relative price of a particular product will 0 p(z) therefore result in a disproportionate reduction (as σ > 1) in the relative consumption of that product.19 Assuming a sufficiently large number of products, the elasticity of demand for each product will be given by σ. The model so far differs importantly from the Dixit and Stiglitz (1977) and Krugman (1979, 1980) monopolistic competition models in that we assume perfect competition and constant returns to scale. The full range of apparel products will therefore be produced and prices will equal marginal cost in equilibrium.20 In addition, the value of exports of any product decline in response to a rise in relative prices, with a greater change the more substitutable are the differentiated products:

(∫

)

⎛ p(z) ⎞ p(z)C(z) = ⎜ ⎝ P ⎟⎠

1−s

aI .

The final condition for equilibrium is that labor demand equals labor supply (L), or alternatively that labor income equals the wage bill in the clothing sector. Assume that apparel products are indexed according to diminishing Chinese relative unit-labor requirements, (a*(z)/a(z)). With z denoting a hypothetical dividing line between Chinese exports (0,z ) and Lesotho exports (z ,1), the home and foreign labor market clearing condition are respectively represented as: (7)

wL =

z

∫0 wa(z)C(z) dz

18. Note that this ensures that the differentiated goods are closer substitutes among themselves than are the differentiated goods and the numeraire good. We do not modify the CES function to allow for quality as in Hummels and Klenow (2005). A quality index acts as a demand shifter, leading to higher consumption at every given price. 19. To see this take the ratio of (6) for product 1 to product 2 to obtain: C(z1)/C(z2) = (p(z1)/p(z2))–σ. 20. While we could follow DFS (1977) and use a Cobb-Douglas utility function for US consumers, this has the disadvantageous outcome that the value of US imports of each variety does not change. Growth in foreign exports to the United States can only be achieved through growth along the extensive margin. This outcome is inconsistent with the empirical evidence.

362

Lawrence Edwards and Robert Z. Lawrence

and (8)

w∗L∗ =

1

∫z w a (z)C(z) dz. ∗ ∗

Taking the ratio of these two conditions gives: (9)

⎛ z ⎞ w ⎜ ∫0 wa(z)C(z) dz ⎟ ⎛ L∗ ⎞ = . w∗ ⎜ 1 w∗a ∗ (z)C(z) dz ⎟ ⎜⎝ L ⎟⎠ ⎝ ∫z ⎠

This schedule is upward sloping on z. A rise in the range of products exported by Lesotho at constant relative wages increases the demand for labor in Lesotho and reduces the demand for labor in the competitor country. This raises the relative wage in Lesotho required to equate demand and supply of labor. Equilibrium is achieved through reductions in relative US consumption of Lesotho exports in response to higher prices. If relative wages are fixed, as may be expected in Lesotho where unemployment is very high, then the adjustment to equilibrium will be through changes in the relative employment of labor in the apparel sector (L*/L falls). In what follows, we impose the fixed-wage assumption to avoid unnecessary complexity associated with the marginal effect of relative wage changes on the range of products exported. Together, equations (2), (3), (6), (7), (8), and the price index solve for Chinese and Lesotho wages, the geographic specialization of apparel exports, the price index P, and US consumption and prices across the full spectrum of products. 9.3.4

Quotas and Product Choice for Exporting Firms

The MFA was important in the markets in which Lesotho and other clothing producers operated, and its application and elimination had major effects (Harrigan and Barrows 2009). Quotas on clothing imports into developed economies were widely applied under the MFA, with imports from China particularly constrained (Brambilla, Khandelwal, and Schott 2010). The MFA, therefore, led to a geographical dispersion of clothing production as producers relocated to countries where there were unused quotas. Lesotho (and other AGOA countries) was a beneficiary of this relocation of production as its US quotas were not filled,21 but the effects on clothing products were not all the same. As we will argue, quotas under the MFA induced the export of low value-added, fabric-intensive, and low-priced (lowquality) clothing products in developing countries such as Lesotho. It is well established in the literature that under competitive conditions a quota is equivalent to a specific tariff (Falvey 1979). The result also holds 21. For data on quota fill rates see the US Office for Textiles and Apparel (OTEXA) (http:// otexa.ita.doc.gov/). Brambilla, Khandelwal, and Schott (2010) provide a review of the fill rates for various countries since the 1980s.

AGOA Rules

363

in cases of imperfect competition (Feenstra 1988, 2004).22 While the quota restricts the total volume of sales, its effect differs across products produced by the firm. Firms adjust exports of different products to ensure that they earn the same quota premium from each good exported (Feenstra 2004). The effect is that exports of low-priced (low-quality) products are the most adversely affected. We find similar effects in our model. Assume apparel quotas are imposed on imports from China. The specific tariff effect of the quota (denoted as s) alters the allocation condition (equation [3]) that determines the range of apparel products exported by Lesotho. The condition becomes: (10)

a(z)w + u(z)PF (z) ≤ a ∗ (z)w∗ + u(z)PF ∗ (z) + s.

Assuming, for exposition purposes, that both Lesotho and China have access to fabric at world prices (PF *(z) = PF(z)), this equation can be respecified in terms of Lesotho’s relative unit labor cost (RULC):23 (11)

s wa(z) . ≤ 1+ ∗ a (z)w∗ w∗a ∗ (z)

Further, if we let λ*(z) denote the share of fabric in foreign costs (λ*(z) = θ(z)PF *(z)/c*(z)), and therefore 1–λ*(z) as unit labor costs as a share of total costs (= w*a*(z)/c*(z)), we can simplify the allocation condition even more to: (12)

wa(z) s ≤ 1+ . (1 − l∗ (z))c ∗ (z) w∗a ∗ (z)

The effect of the quota is a modified allocation condition in which the righthand side of the DFS equation (4) is raised by the term s/(1 – λ*)c*. This term is positive and rises if, ceterus paribus, s increases, costs fall, or the share of fabric in production rises. We can consider four implications of quotas under the MFA for apparel exports from Lesotho using this relationship.24 First, the effect of a quota is equivalent to a specific subsidy on exports from non-quota-constrained countries such as Lesotho. This enables Lesotho’s producers to export 22. See Krishna (1987) for an imperfect competition model where firms jointly select the quantity and the quality of the products they export in response to a quota. Feenstra (1988, 2004) also show how quotas lead to an upgrading of the characteristics within each variety produced. 23. In Lesotho, for example, import duties were rebated on imported fabric used in the production of apparel exports. We ignore the effects that transport cost differentials have on the relationship. Specific transport costs on output can be modeled in an equivalent way to the effect of specific tariffs and quotas. For example, relatively high specific transport costs on output for the competitor country have the equivalent effect on quality as our example for quotas. See Falvey (1979) and Hummels and Skiba (2004). 24. There are two additional considerations. Missing from this story is the fact that withinquota tariffs were also imposed under the MFA. As shown by Hummels and Skiba (2004), ad valorem tariffs lower the relative demand for high-quality goods in the presence of per unit

364

Lawrence Edwards and Robert Z. Lawrence

apparel products even if they do not have a comparative advantage in the production of that product, that is, where their RULC exceeds 1 by up to s/(1 – λ*(z))c*(z). The implicit subsidy conferred by the tariff compensates the relatively inefficient apparel producers for their high relative unit-labor costs and helps explain why countries such as Lesotho exported apparel under the MFA despite productivity levels that were lower and wage levels that were comparable to those of Asian levels (Lall 2005). The second consideration is that the implicit subsidy of the quota for Lesotho (and other non-quota-constrained countries) is a greater percent of the overall value the lower is the price (c*(z)) of the product exported by China. This is the standard result for quotas obtained by Falvey (1979). Quota-constrained countries upgrade quality of exports by shifting to higher-priced varieties. What we show here is that the gap in the market is filled by non-quota-constrained countries that may not have been able to compete prior to the quotas, that is, where their RULC exceeded 1. The third consideration is novel to our model. Holding costs constant, the effective subsidy, that is, the subsidy as a proportion of value added, rises exponentially with the share of costs (of the efficient producer) attributed to fabric. Alternatively, the effective subsidy is greater the smaller the value added of the product. Take, for example, two apparel products each priced at US$10, but differing in terms of fabric intensity: Fabric costs make up 90 percent of the cost of product A and 1 percent of the cost of product B. Assume further that the specific tariff equivalent of a quota on imports from China is one dollar. The resulting effective export subsidy for Lesotho is just over 1 percent for product B, but is 100 percent for product A. In fact, the effective subsidy is 100 percent for any product in which China’s labor costs are equal to one dollar. The implication is that firms in Lesotho will be able to compete in exporting product B, even if their unit labor costs are 100 percent higher than those of the competitive quota constrained supplier (China). These considerations explain how the quotas enable an expansion in the range of products exported by Lesotho and other non-quota-constrained countries, that is, growth along the extensive margin. The impact of the quota is to shift the Chinese out of low-priced products. This implies that relative price increase of Chinese exports will be strongest in low-priced products. This leads to the fourth effect: Lesotho will also experience growth in the volume and value of existing exports (intensive margin) transport costs (or equivalently quotas). As tariffs rise, the shadow price of the quota constraint falls and dampens the effect (but not direction) of the quota on relative demand for high-quality products. The final consideration is that import quotas administered by the Office of Textiles and Apparels (OTEXA) are specified in terms of yardage of fabric equivalents and not quantity of goods. In this case, the quota is equivalent to a specific tariff on the price per square meter of fabric equivalence, that is, the allocation condition is: c(z)/θ ≤c*(z) /θ + s. The implication for Lesotho is that relative demand and relative prices shift in favor of exporting low-priced clothing varieties that are intensive in the use of cheap fabric.

AGOA Rules

365

as the relative price of their existing exports falls relative to the exports from quota-constrained countries (their export price falls relative to the average price index; see equation [6]). Therefore, in Lesotho we expect the strongest intensive margin growth in exports in existing low-priced products. In contrast, quota-constrained countries experience a decline in the range, value, and volume of their exports, particularly in low-priced products. In conclusion, we expect four effects of the MFA on Lesotho and other AGOA countries (and other non-quota-constrained countries): a rise in the export of both (a) existing and (b) new apparel products combined with specialization in (c) cheap low-quality products with (d) very little value addition. The removal of the MFA would have had the opposite effects. Previously, quota-constrained countries would shift production toward cheaper products with lower labor-value added. Unconstrained countries would thus be especially adversely affected in these shifts, both in terms of the range and value of their apparel exports. 9.3.5

Tariff Preferences and Product Choice

We now turn to an analysis of the effect of the tariff preferences granted under AGOA. Generally, theory suggests that in a competitive market ad valorem tariffs have no impact on value addition as they preserve relative prices faced by the firm and the consumer (Falvey 1979; Feenstra 1988).25 This changes once we introduce tariffs and tariff preferences in a model where products contain internationally traded intermediate inputs such as fabric. Once tariffs are introduced, what determines whether Lesotho exports product y(z) is whether the tariff-inclusive price of its good in the US market is less than or equal to its foreign competitor, China:26 (13)

(

)

c(z) (1 + t ) ≤ c ∗ (z) 1 + t ∗ .

Letting ϕ(z) denote the Lesotho fabric price relative to the Chinese fabric price, (PF(z) = PF *(z)), we can express the allocation condition (13) in terms of relative unit labor costs (RULC) and fabric-cost shares (λ) as follows: (14)

(

)

1 + t∗ ⎤ wa(z) l(z) ⎡ (1 + t ∗ ) ≤ + − w(z) ⎥. ⎢ ∗ ∗ w a (z) (1 + t ) 1 − l(z) ⎣ (1 + t) ⎦

The cut-off point defining what products will be exported by Lesotho is now a function of relative tariff rates faced, fabric intensity, and relative 25. Krishna (1987) presents an imperfect competition model where the firm’s choice of output and quality is influenced by ad valorem tariff rates. 26. To simplify the model we have assumed that the ad valorem tariff does not vary by variety. Apparel tariffs actually vary enormously according to the type of fabric used, and in some cases according to the quantity and amount of fabric used in production. Extending the model to allow for variation in tariffs across z does not alter the main insights of the theory.

366

Lawrence Edwards and Robert Z. Lawrence

fabric prices. To explore the implications for product choice under AGOA, three different scenarios are compared: (a) Case 1: Pre-AGOA with MFN tariffs and competitive input supplies; (b) Case 2: AGOA tariff preferences for LDC special rule beneficiaries; and (c) Case 3: AGOA tariff preferences for non-LDC special rule beneficiaries. Case 1: No preferences, MFN tariff rates (t = t*), and access to competitively priced inputs. In the first scenario, the United States imposes common MFN tariffs on apparel imports from Lesotho and China. For MFN trade there are no rules of origin requirements or restrictions on access to internationally priced fabric, so barring domestic restrictions on use of inputs, all countries have access to internationally priced fabric, that is, (ϕ(z) = 1). In this scenario, the product allocation condition (14) reduces to the standard RULC condition of the DFS model (equation [4]). Tariffs affect both countries equivalently and the unit fabric-cost components cancel each other out. The geographic location of production is determined entirely by relative unit-labor costs, with specialization according to comparative advantage. Fabric intensity has no bearing on what a country exports. Tariff protection in this scenario introduces no fabric-use bias. Case 2: Preferential access granted to home (t = 0, t* >0) and no rules of origin. The second scenario is set up to reflect the AGOA preferences granted to Lesotho and other LDBCs. These countries are granted a tariff preference into the United States (t = 0, t* > 0), but under the LDC special rule are also able to use internationally competitive third-country fabric in the production of apparel exports. Given our assumption of no-transport costs, fabric prices are therefore equal in Lesotho and China ((ϕ(z) = 1). The allocation condition in this case simplifies to: (15)

wa(z) l∗ (z) ≤ 1 + t∗ + t∗ . ∗ ∗ w a (z) 1 − l∗ (z)

(

)

( )

What can be observed from the relationship is that the effective preference is a function of both the tariff preference as well as the ratio of fabric cost shares to value-added costs shares (λ*/(1 – λ*)). We explore the implications of this in more detail. Take a scenario where apparel products contain no fabric, that is λ* = 0, and the term on the far right of equation [15] falls away. The tariff preference has a uniform impact on all apparel products and allows Lesotho to export products in which it is up to 1 + t* times less efficient at producing than China. For example, a tariff preference of 20 percent enables the home country to export new apparel products where its unit labor costs are up to 20 percent greater than their foreign competitors. In addition to the export of new products (i.e., growth along extensive margin), the tariff reductions under AGOA also raise US consumption of existing products exported by Lesotho (i.e., the intensive margin) through reductions in the relative US consumer price of these goods. The effect on

AGOA Rules

367

the volume and value of exports could be very large if products are highly substitutable. We would therefore expect to see growth in exports along both the intensive and extensive margin. Once fabric is introduced, the AGOA preferences alter relative incentives to export products of different unit-fabric contents. In particular, the tariff preference is greater for products with higher fabric-cost shares. This is revealed by the second term on the right-hand side, which is positive and increasing (exponentially) in λ*. As the fabric-cost share approaches 1, the effective preference granted to Lesotho converges on positive infinity. Diagrammatically, this relationship is represented in figure 9.5. Assume for simplicity’s sake, that Lesotho’s relative unit-labor costs are fixed at (aw/a*w*)1 on the vertical axes for all apparel products. We have also assumed that Lesotho’s RULC exceeds 1. On the horizontal access, apparel products are ordered according to rising fabric-cost shares (or diminishing labor-cost shares). In a competitive environment where Lesotho and China face the same US tariffs (see Case 1 in figure 9.5), Lesotho would not export any products as its RULC exceeds 1. With tariff preferences plus waivers from the rules of origin granted to Lesotho, the relevant comparison is between RULC and the solid line (equals the right-hand side of equation [15]) identified as Case 2 in the figure. Lesotho is still unable to export products with low fabric-cost shares. However, because the effective tariff preference rises with fabric-cost share, Lesotho is able to export all products in which the fabric-cost share exceeds λ1, despite having no comparative advantage in these products. In sum, the tariff preference affects Lesotho’s exports in three ways. First, it raises the relative unit-labor-cost threshold by (1 + t*), which is equivalent to what we would expect in a tariff-adjusted DFS model. Second, the threshold defining the cut-off-point is higher for fabric-intensive products. This arises because tariffs not only tax foreign unit-labor costs, but also tax the fabric content of the product. The total tariff equivalent preference per unitlabor cost is therefore an increasing function of the unit-fabric-cost share.27 Finally, by reducing the relative price of exports, the preferences increase the volume and value of existing imports from beneficiary countries. The implication for LDC AGOA beneficiaries is that they enter and specialize in the export of the most fabric-intensive apparel products. The AGOA preferences to LDC beneficiaries therefore compound the existing incentives to produce low value-added or fabric-intensive products brought about by the MFA. There is one important difference. The AGOA incentives are unre27. In a small country price-taking model, the tariff effects are greatest for fabric intensive products even among those goods where it has a comparative advantage (the intensive margin). The tariff preferences therefore create incentives for firms to expand production most in the low-value-added, fabric-intensive varieties of products they are already exporting. In addition, the preferences would encourage entry of the least efficient firms into the most fabric-intensive apparel products.

368

Lawrence Edwards and Robert Z. Lawrence

Fig. 9.5 The effect of tariff preferences on incentive to export fabricintensive products

lated to the price of the product, only fabric intensity, whereas the implicit subsidy for non-quota-constrained countries under the MFA is greatest for low-priced, fabric-intensive products. Case 3: Preferential access granted to home (t = 0, t* > 0), but rules of origin constraints on fabric inputs. This scenario reflects the situation for non-LDC special rule AGOA countries such as South Africa (and Mauritius for most of the post 2001 period). Apparel exports from these countries have preferential access into the US market, but production is subject to a two-stage transformation requirement. Apparel producers from these countries are therefore required to use domestic (or US)-produced fabric in the production of exports to the United States under AGOA preferences. If these countries produce fabric at internationally competitive prices, ϕ(z) = 1, then the outcome will be equivalent to Case 2. However, if local fabric is more expensive than foreign fabric, ϕ(z) > 1, the allocation condition is given by: (16)

wa(z) l∗ (z) ≤ 1 + t∗ + 1 + t ∗ − w(z) . ∗ ∗ w a (z) 1 − l∗ (z)

(

)

(

)

The relationship differs from equation [15] in that while the home country is granted a tariff preference, it has to utilize more expensive domestic fabric. The impact on clothing production relative to the pre-AGOA period is ambiguous and depends on the fabric-price disadvantage relative to the tariff preference. Take for instance a scenario (Case 3a) where the home relative fabric-price disadvantage is less than the tariff preference such that

AGOA Rules

369

(1 + t* – ϕ(z)) > 0. In this scenario, the effective preference rises with fabric intensity, but less so than in Case 2. In figure 9.5 this is depicted by the dashed line identified as Case 3a. The home country will export all products in which the fabric-cost share exceeds λ2. An alternative scenario (3b) is one where the fabric-price disadvantage is greater than the tariff preference such that (1 + t* – ϕ(z)) < 0. Here the effective preference declines as the fabric intensity of the product rises. At some level of fabric intensity, the fabric-price disadvantage will dominate the tariff preference effect and reduce the right-hand side of equation (15) to below 1. At this point, there is a disadvantage associated with exporting under the preferential access scheme as opposed to exporting under MFN rates (Case 1). Firms that are competitive in these products, that is RULC < 1, will then export under MFN rates. In figure 9.5, Case 3b depicts the declining effective preference (Case 3b), although in this example, the non-LDC AGOA beneficiaries will not export at all as the allocation condition [15] is not met for any product. Bar the case of competitive domestic fabric producers, our model predicts that LDC AGOA beneficiaries such as Lesotho will experience higher growth in export volumes (along both the extensive and intensive margin) to the United States than other AGOA beneficiaries. The effect will be particularly pronounced in fabric-intensive apparel products. 9.3.6

Other Effects

Other considerations relate to the development of a comparative advantage in the nascent industry. Our model raises a number of concerns in relation to this. First, the incentives steer firms to producing products with the lowest value addition conditional on price, rather than up the value chain. If these products are characterized by low positive-growth externalities, then the preference may trap firms into a lower-growth path than alternative preferences that incentive greater value addition. Second, our model does not deal with the opportunity cost of resources used in the production of apparel. If labor supply is not infinite, then the growth in the apparel industry will raise wages, which may actually drive out export firms in other sectors where the home country has a comparative advantage. This also holds for other scarce resources such as infrastructure, land, and water. Third, the specialization by firms in fabric-intensive products makes these exporters highly vulnerable to international price volatility (either through exchange rates or international prices), preference erosion through lower MFN tariff rates, and the ending of the waiver of the rules of origin. Changes in these variables result in an amplified impact on the effective subsidy provided by the AGOA preferences and the MFA quotas. Preference erosion could therefore provide an additional blow that would be seriously

370

Lawrence Edwards and Robert Z. Lawrence

underestimated if models fail to capture the contribution of the rule of origin preference. Finally, the preferences restrict backward linkages by discouraging the addition of value-added services from other sectors and inducing exporters to use expensive fabric that is less likely to be produced in poor countries. 9.4

Empirical Application: Testing Methods and Data

Our background review identified three distinct trade regimes facing AGOA recipients from the mid-1990s: (a) quotas under the MFA; (b) AGOA preferences, including the third-country fabric provision; and (c) the expiration of the MFA. Theory suggests that each of these trade regimes had different impacts on the incentives facing AGOA and non-AGOA countries In what follows we describe the testing approach we use and then apply this to highly disaggregated US import data. Our specific focus is on changes in the characteristics (value addition and fabric intensity) of AGOA apparel exports associated with the MFA and AGOA preferences. We ignore the effects on the value and range of imports, as this is already covered by existing empirical research.28 The empirical method we use is differencein-difference estimation applied to price equations. In essence, we identify changes in the fabric intensity of US apparel imports from AGOA recipients by analyzing changes in the relationship between apparel import prices and fabric input prices. We find support for our theoretical predictions. Under the MFA, AGOA recipients are found to be specialized in fabric-intensive clothing products with low value addition relative to quota-constrained (and other) countries. Our estimates suggest, however, that the implementation of AGOA led to no further increases in the overall fabric intensity of these exports. Lesserdeveloped beneficiaries predominantly expanded the output of the products they were already exporting as a result of their MFA preferences; that is, growth was primarily along the intensive margin. However, support for our hypothesis of rising fabric content in response to the AGOA preferences is found after the expiration of the MFA. China and other previously quota-constrained countries raised the fabric content of their exports after 2005 relative to other emerging economies, as predicted by our theory. More important for this study is that we also find a rise in the fabric content of lesser-developed AGOA apparel exports relative to the 28. We have estimated triple difference-in-difference equations similar to those of Frazer and Van Biesebroeck (2010) and do find a surge in apparel imports from lesser-developed beneficiaries relative to other AGOA recipients (and the rest of the world) in response to the thirdcountry fabric provision. The average growth in imports from 2001 through 2004 associated with the fabric provision is estimated to be up to 282 percent, with stronger effects in products facing high preference margins. We also find that the expiration of the MFA adversely affected exports from AGOA recipients, but the effect was mitigated for the least developed AGOA countries by the third-party fabric preferences provided under AGOA.

AGOA Rules

371

emerging country control group. The AGOA preferences therefore helped insulate recipients in those fabric-intensive products that China and other quota-constrained countries increasingly entered into after 2005. 9.4.1

Empirical Specification of the Price Equation

An important limitation of existing empirical studies on the effect of AGOA on import values (Collier and Venables 2007; Portugal-Perez 2008; Frazer and Van Biesebroeck 2010) is that import value data, even at the HS ten-digit level, is too aggregated to fully capture changes in product characteristics. By only looking at the value or range of HS ten-digit products exported by each country, existing studies may miss important changes occurring within each product line. Take, for example, figure 9.6 that plots US import unit values on exporter per capita GDP (both in logarithmic form) for women’s and girls’ cotton pullovers (Lesotho’s top apparel export) in 2004. The price of imports of this highly disaggregated HS ten-digit product ranges from under ten dollars to over one thousand dollars per square meter, equivalent with higherincome economies producing the more expensive (higher quality) varieties (as in Hummels and Klenow [2005] and Schott [2004]). The lesser-developed

Fig. 9.6 Unit values and level of development: Top apparel product exported by Lesotho in 2004 (women’s or girls’ other pullovers of cotton, knitted) Notes: Triangles are AGOA countries eligible to export apparel. Square blocks reflect the top quota-restricted countries from 1984 to 2004 as identified by Brambilla, Khandelwal, and Schott (2010).

372

Lawrence Edwards and Robert Z. Lawrence

AGOA recipients predominantly situate at the low-price, low-income per capita end of the spectrum.29 Our particular focus is on product prices. More specifically, we use difference-in-difference estimation to exploit the distinct breaks arising from the implementation of AGOA and the ending of the MFA and to identify whether price changes and changes in the fabric-intensity of apparel products are consistent with those predicted by our theory. Following Feenstra (2004), the US domestic price of an imported good i from country c is specified as a function of marginal costs (c*), the exchange rate (e), import tariffs (tar), aggregate domestic expenditure (I), and the price of substitute goods (q) as follows: (17)

∗ ln pict = a + b1 lncict + b2 lnect + b3 lnqict

+ b4 ln(1 + tarict ) + b5 I t + ´t

.

This is an unrestricted version of a price equation that imposes symmetric pass-through of the exchange rate and foreign costs (where β1 = β2), symmetric pass-through of the tariff and exchange rate (where β2 = β4), and homogeneity of degree 1 in its arguments (β1 (= β2 =β4) + β3 + β5 = 1). We are particularly interested in isolating changes in the fabric content of US apparel imports using this equation. This requires a more precise specification of the influence of fabric costs on unit costs c*. To simplify the analysis, we impose a unit-cost function derived from a constant return to scale Cobb-Douglas production function: (18)

∗ 1−a , = At pfita pvaict cict

pf is the price of the fabric used in the production of good i, pva is the valueadded price (made up labor and capital costs), and At measures total factor productivity. This specification imposes the restriction that the proportion of expenditure spent by the firm on fabric is constant and is given by α. Substituting (18) into (17) gives the following equation: (19)

ln pit = a + d1 ln pfit + d2 ln pvait + b2 lnet + b3 lnqit + b4 ln(1 + tarit ) + b5 I t + ´t

where d1 = b1a and d2 = b1 (1 − a). Given the assumptions imposed, the fabric content of the clothing product can be calculated as δ1/(δ1 + δ2) = β1α/(β1α + β1(1 – α)) = α. Fabric-intensive products would therefore be characterized by large coefficients on the fabric price (δ1) relative to the coefficient on the value-added price (δ2). There are two changes in response to the MFA and AGOA that we wish to identify: (a) changes in the price level, and (b) changes in the fabric intensity of US apparel imports. 29. There are exceptions. Apparel unit values of China, India, and Indonesia, who were among the top four quota-restricted countries under the MFA (Brambilla, Khandelwal, and Schott 2010) are higher than predicted. This is consistent with theoretical predictions of quality upgrading in response to quota restrictions.

AGOA Rules

373

To identify changes in the level of import prices from a region in response to a shock, say AGOA recipients after 2001, the above equation is modified to include an interaction between an AGOA dummy variable (DAg) and a dummy variable for the post-AGOA period (D01). The basic price equation in this example is then specified as: ln pict = u1D01 × DAgc + d1 ln pfit + d2 ln pvaict

(20)

. + b2 lnect + b3 lnqict + b4 ln(1 + tarict ) + cntry / prodct + lt + ´ict

where θ1 measures the marginal effect of the AGOA preferences (D01) on unit values of US imports from AGOA countries (DAgc) relative to all other countries in the sample (the control group). Country by product (cntry/ prod) fixed effects are included, so the regression uses the within-country by product variation of prices and the other variables over time to estimate the coefficients.30 Year fixed effects λt are also included to account for common shocks across all product varieties. To identify changes in the fabric intensity associated with the various trade regimes, we focus on changes in the coefficients on the fabric and value-added prices. For example, we would expect a shift by AGOA recipients to more fabric-intensive varieties within each ten-digit product line to be revealed by a rise in the coefficient on fabric prices and a decline in the coefficient on value-added prices. We use difference-in-difference estimation to identify changes in the relative fabric content of apparel imports from AGOA beneficiaries. The specification in the case of AGOA preferences is as follows: r1 : ln pict = u1D01 × DAgc

(21)

r2 :

+ (a1 + u2 D01) × DAgc × ln pfit

r3 :

+ (a2 + u3 D01) × DAgc × ln pvaict

r4 :

+ (d1 + u4 D01) × ln pfit

r5 :

+ (d2 + u5 D01) × ln pvaict

r6 :

+ b2 lnect + b3 lnqict + b4 ln(1 + tarict )

r7 :

+ cntry / prodic + lt + ´ict

.

The first row tells us the effect of AGOA on US import unit values of apparel products imported from AGOA recipients. Rows 2 and 3, however, are of most interest to us. The coefficients on the triple interaction terms (θ2 and θ3) 30. The standard most restrictive difference specification includes a dummy variable for AGOA countries (DAgc), but in equation (22) these have been replaced with country by product fixed effects (cntry/prod) to allow for country and product-level heterogeneity in the base-level of import prices.

374

Lawrence Edwards and Robert Z. Lawrence

measure the marginal impact of AGOA preferences on the fabric intensity of US imports from AGOA recipients (first difference) relative to changes in the fabric intensity of US imports from the control group (second difference). The latter effect is captured by the coefficients θ4 and θ5 in rows 4 and 5. For example, support for our hypothesis that AGOA preferences raise the fabric intensity of imports from recipient countries would be revealed by a positive coefficient on the AGOA country by post-2001 interaction with the fabric price (θ2) in row 2 and a negative coefficient on the AGOA country by post-2001 interaction with the value-added price (θ2) in row 3. Note that these coefficients reflect the post-2001 impact on fabric intensity in AGOA countries relative to the post-2001 impact on fabric intensity in the control group, which is captured by the coefficients θ4 in row 4 and θ5 in row 5.31 Equations (20) and (21) summarize the main approaches used in the following analysis. Further refinements to isolate the marginal effects of the MFA and AGOA preferences on lesser-developed AGOA beneficiary countries will be explained in the relevant empirical sections. 9.4.2 Data The empirical analysis draws on a panel of time-consistent ten-digit HTS import data for the United States from 1996 to 2008.32 The raw data contains approximately 1,202 product lines for clothing (HS 61, 62, and various subcodes of HS 64 and 65) covering 224 countries. Unlike the price equations specified above, the dependent variable is the log import price of clothing exclusive of tariffs, insurance, and freight costs. This does not affect the estimates, except that the pass-through of tariffs to US domestic prices of imports is calculated as 1 – δ7. Looking at the independent variables, we use the foreign industry value-added deflator (in foreign currency) for pva, the US dollar to foreign currency exchange rate for e, and US producer prices (at six-digit NAICS level) (usppi), and competitor clothing unit values (at ten-digit level) (Pcompete) for substitute products q. Applied tariff rates are defined at the four-digit HS level.33 In addition to these variables, real GDP per capita measured in purchasing power parity 31. We could also include product by year fixed effects and country by year fixed effects. In this case, only variables defined over product, country, and time will be retained. 32. The HTS classification changed frequently throughout the period as new product lines were introduced and old product lines were aggregated. We use the Pierce and Schott (2009) concordance program to construct a time-consistent classification for the full period. 33. We use the average tariff at the HS four-digit level to avoid erroneous correlations arising from the construction of the variables (tariff rate = duty/import value, and price = import value/import quantity). Using the average may also reduce biases associated with the potential endogeneity of product-level tariff rates. The trade data are obtained from Peter Schott who constructed the database using US Customs Service data. The US producer prices are obtained from the Bureau of Labor Statistics, fabric prices are constructed using UNComtrade data, and the exchange rates are obtained from the World Bank World Development Indicator database. Country-specific tariff rates at the four-digit HS level are constructed as the sum of duties collected over value of imports. Competitor clothing prices are calculated as the geometric average price of all other countries (using import values as weights).

AGOA Rules

375

Fig. 9.7 Fabric price indices (based on world exports and imports from UNComtrade) Notes: Based on Tornqvist price index constructed using HS six-digit unit values obtained from UnComtrade trade data. HS six-digit product lines for fabric (HS 50-silk; 51-wool, fine animal hair; 52-cotton; 54-man-made fiber; and 55-man-made staple).

[PPP] prices is included to capture the impact on prices of general productivity improvements in the economy and relative technological advantage in producing higher-quality goods (Hummels and Klenow 2005).34 For fabric prices, we calculate Tornqvist price indices for silk (HS50), wool and fine animal hair (HS51), cotton (HS 52), and man-made fiber and staple (HS 54 and HS55) using unit values derived from world trade data obtained from UNComtrade.35 The calculated fabric indices are presented in figure 9.7.36 Of interest is the relatively close association between the average US import unit value of wearing apparel (HS 61 and HS 62) and fabric prices, particularly man-made fabrics. The relevant fabric price (silk, cotton, man-made, wool, or weighted average of these) is allocated to each ten-digit HTS clothing product based on the dominant fabric used in producing the good.37 Unfortunately, we are 34. Although the industry value-added price is the net effect of productivity and nominal factor prices, the real GDP per capita also embodies productivity improvements in the services sector. 35. The following HS codes for synthetic fibres are also included in man-made products: 550110, 550120, 550130, 550190, 550200, 550310, 550320, 550330, 550340, 550390, 550410, 550490, 550610, 550620, 550630, and 550690. The average of the fabric prices were calculated using world exports and world imports. 36. The fabric prices correspond closely with the dominant agricultural commodity used to produce the fabric. For example, there is a close fit between cotton-based fabric and raw cotton prices and wool-based fabric and wool prices. 37. The allocation was done manually on the basis of the product description.

376

Lawrence Edwards and Robert Z. Lawrence

unable to construct weighted average fabric price indices for apparel products produced using different combinations of fabric types.38 We now separately apply the various equations to the AGOA and MFA trade regimes. 9.5

African Growth and Opportunity Act

Our expectation is that AGOA preferences stimulated US imports from beneficiary countries, with relatively high growth in imports of fabricintensive and low-value-added products. The effects are predicted to be particularly pronounced in LDC recipients eligible to use third-country fabric. Table 9.5 presents regression results for various specifications of the price equation. The first column presents benchmark estimates of the price relationship over the period 1996–2004 and is used to evaluate the consistency of the price equation with our theoretical priors. Overall, the price model produces results that are consistent with theory and other empirical evidence (see Feenstra 2004). The dollar price of US clothing imports rise with increases in foreign and US competitor’s prices. Import unit values rise with foreign GDP per capita reflecting a positive association between income and quality of exports as explained by Hummels and Klenow (2005). Applied tariffs reduce the fob price of apparel products with a coefficient of –0.60, which is very close to the effect of an equivalent depreciation of the dollar.39 Foreigners therefore absorb 60 percent of tariff increases or depreciation either through lower mark-ups (in case of imperfect competition) and/or reduced marginal costs (from upward-sloping supply curve). Further, rising foreign production costs result in higher US import prices. The US import prices are equally affected by increases in foreign fabric costs and value-added costs, implying a fabric-share coefficient of approximately 50 percent.40 Various diagnostic tests reveal that the aggregate model fails the homogeneity test and the hypothesis of symmetric pass-through of the tariff and exchange rate. However, far fewer instances of rejection are found in the disaggregated HS four-digit-level estimates. The disaggregated results and hypotheses tests are presented in table 9A.1 in the appendix. We are therefore reasonably satisfied with our basic price equation and proceed with our objective of identifying differences in the fabric content of AGOA apparel exports. The second column of results extends the base regression by including 38. See Goldberg and Knetter (1997) on how aggregate production cost indices can bias the exchange-rate pass-through downward. The value-added deflator is also more aggregated than is desired. 39. The estimated exchange rate pass-through coefficient of 0.6 falls between Feenstra’s (1988) estimates for trucks (0.63) and cars (0.71) and more general estimates based on aggregate import data (Marazzi et al. 2005; Gopinath and Rigobon 2008). 40. The coefficients on value added and fabric prices are insignificantly different from each other.

N F Fixed effects

Estimates are robust to heteroskedasticity. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

3 4

1 2

255,231 81.61 country/product year

255,231 90.9 country/product year

255,231 67.2 country/product year

–0.063 0.124*** –0.547*** 0.037*** 0.109 –0.684***

0.405*** –0.423*** 0.099*** –0.104***

0.323*** –0.389***

0.115*** –0.548*** 0.037*** 0.131*** –0.637***

0.316*** 0.213***

–0.114 0.113

All AGOA (3)

0.268*** 0.252***

All AGOA fabric intensity (2)

0.126*** –0.538*** 0.037*** 0.135 –0.600***

0.272*** 0.237***

All base (1)

Marginal impact of AGOA preferences on fabric intensity in beneficiary countries

Marginal impact of AGOA on fabric intensity LDC AGOA countries relative to other AGOA D01 × Dldc × ln(pf) D01 × Dldc × ln(pva) AGOA countries relative to control D01 × DAg × ln(pf) D01 × DAg × ln(pva) Other coefficients ln(pf) ln(pva) Dldc × ln(pf) Dldc × ln(pva) DAg × ln(pf) DAg × ln(pva) D01 × ln(pf) D01 × ln(pva) D01 × Dldc D01 × DAg ln(GDP/capita), PPP ln(e) ln(Pcompete) ln(US ppi) ln(1 + t)

Country sample

Table 9.5

255,231 54.8 country/product year

0.315*** 0.213*** –0.049 0.144 0.446** –0.537** 0.099*** –0.104*** 0.034 –0.092 0.122*** –0.546*** 0.037*** 0.11 –0.689***

–0.175 0.186

0.067 –0.083

All marginal LDC AGOA (4)

378

Lawrence Edwards and Robert Z. Lawrence

interactions between an AGOA dummy (DAg) and fabric costs and valueadded prices (see rows 9 and 10). The objective of this estimate is to identify the average fabric intensity of US imports from AGOA beneficiaries throughout the 1996 to 2004 period. The results indicate that AGOA countries produce relatively fabric-intensive clothing products with low value addition. The coefficient on the fabric price (DAg × ln(pf)) is positive and significant (0.323), while the coefficient on value-added prices (DAg × ln(pva)) is significant and negative (–0.389). Therefore, US unit values of apparel imports from AGOA beneficiaries are far more sensitive to fluctuations in fabric prices than apparel imports from the rest of the world. We infer from this result that AGOA beneficiary exports are relatively fabric intensive. This outcome is consistent with both the effect of the AGOA preferences and the MFA. To identify the effect of AGOA preferences on beneficiary exports, we use the specification in equation (21) where the time period dummy variable in the interactions refers to the 2001 to 2004 period.41 The relevant results are presented in rows 3 and 4 in column (3) of table 9.5. These are the coefficients on the difference-in-difference terms that measure the change in fabric intensity of US imports from AGOA beneficiaries after 2001 relative to the change in fabric intensity of imports from the rest of the world. Our expectations are that AGOA preferences raised the fabric intensity of imports from beneficiary countries. However, contrary to our theoretical predictions, we find no increase in the fabric intensity of apparel exports from 2001 to 2004 in response to the AGOA preferences. The coefficients on the interaction terms (D01 × DAg × ln(pf)) in row 3 and (D01 × DAg × ln(pva)) in row 4 are insignificantly different from zero. One reason may be that the above estimates are an average for both LDC AGOA and other AGOA countries. Our theory suggests that the effect of AGOA preferences on fabric intensity is particularly pronounced among LDC AGOA countries who are eligible for the third-country fabric provision. To isolate the marginal impact of the third-country fabric provision on fabric content, we include additional interactions of ln(pva) and ln(pf) on dummy variables for LDC AGOA countries (Dldc) over the full period and over the 2001–2004 period. Estimates of this relationship are presented in column (4). The coefficients on the LDC interaction terms in rows 2 and 3 are interpreted as the marginal impact of AGOA on fabric intensity in LDC special rule countries relative to the rest of AGOA beneficiaries. We still find no increases in the fabric content of apparel exports by lesserdeveloped AGOA countries relative to other AGOA countries or the rest of the 41. Not all countries became eligible to export apparel in 2001. D2001 therefore varies by country and time and equals 1 for all years from the time the country becomes eligible to export apparel products. The dummy variable is set equal to 1 for the initial year if eligibility occurred within the first six months of the year.

AGOA Rules

379

world from 2001 to 2004. None of the marginal effects for LDC special rule countries are significantly different from zero. Overall, the results suggest that the preferences under AGOA had very little impact on the within-product fabric content of apparel exports to the United States by recipient countries. The AGOA beneficiaries, including lesser-developed special rule countries, were already specialized in fabricintensive products prior to receiving AGOA preferences. The impact of AGOA was to make production of these products more attractive, and they responded by increasing exports of these products rather than of new fabricintensive products. This is consistent with the decomposition of growth analysis in table 9.3, which showed that the expansion of exports was overwhelmingly along the intensive margin. 9.6

Expiration of MFA

The ending of the MFA presents an additional policy “experiment” to test our theory as applied to AGOA beneficiaries. As noted, quotas under the MFA were removed on the January 1, 2005, although some quotas were reimposed in industrialized countries in response to the rapid growth in imports from China.42 In this section, we exploit this break to indentify whether import values, import unit values, and the fabric intensity of US apparel imports moved in accordance with our predictions. Theory predicts that firms in previously quota-restricted countries respond to the ending of quotas by downgrading the quality of their apparel exports. In our model, this would be revealed by relatively strong growth in imports of low-priced varieties from previously quota-restricted countries that include AGOA beneficiaries. Evidence in support of quality downgrading is found by Brambilla, Khandelwal, and Schott (2010) and Harrigan and Barrows (2009). A second hypothesis derived from our theory is that, conditional on price, quota-restricted countries responded to the ending of the MFA by increasing exports of fabric-intensive apparel products. In this section, we test these two hypotheses focusing on the response by quota-restricted countries relative to AGOA beneficiaries. Preliminary support for the effect of the MFA on product quality is provided in figure 9.8 that presents a measure of within-product price differences for selected countries relative to Lesotho. These are calculated by aggregating up the log ratio of export prices relative to Lesotho using Lesotho export values as weights. Higher values reflect the export of more expensive apparel varieties than Lesotho within each product line. During the MFA period, quota-constrained countries such as China, 42. We do not take into account the reimposition of quotas on selected Chinese apparel products from late 2005. As shown by Harrigan and Barrows (2009) these contained, but did not reverse, the import response to the end of the MFA.

380

Lawrence Edwards and Robert Z. Lawrence

Fig. 9.8 Import weighted average price relative to Lesotho (using Lesotho exports as weights) Note: The import weighted average price for country c is calculated as ADD EQUATION where wiLt is the share of product i in Lesotho’s apparel exports to the United States, PiLT is the price of Lesotho exports, and Pict is the price of the comparator country apparel exports.

Bangladesh, and India exported varieties within each HS ten-digit line that were up to twice as expensive as those from Lesotho. The expiration of the MFA, however, saw a dramatic decline in the relative price as these countries downgraded the quality of their apparel exports: see the relative price of Chinese apparel that fell from 1.95 times to 90 percent of those from Lesotho in one year. There was a slight rebound from 2006 as new quotas on Chinese apparel exports were imposed, but by 2008 relative prices had still fallen by over 55 percentage points from 2004. The composition of imports from quota-constrained countries also shifted toward the low-priced products exported by Lesotho. Figure 9.9 presents import weighted prices (per square meter equivalent) of apparel imports from each country calculated using the product-level median prices for the entire sample and period and time-varying import values by country as weights. Reductions in the average price reflect across-product shifts in the composition of apparel exports to the United States toward lower-priced products. The shift in composition is most noticeable for China, whose apparel exports were initially concentrated in relatively expensive ten-digit apparel products, but then fell in 2002 as quotas imposed under Phase I, II, and III of the MFA were eliminated in response to China’s entry into the World Trade Organization (WTO). A further shift toward low-priced products occurred in 2005 after the ending of Phase IV of MFA, and by 2008 the import weighted median price of Chinese apparel exports was very similar to those of Lesotho.

AGOA Rules

381

Fig. 9.9 Structural shifts in the composition of US apparel imports, import weighted US average, unit value ($) per SME Note: The import weighted average price for country c is calculated as ADD EQUATION where ADD EQUATION is the median price of product i over the entire period and mict is the share of i in country c’s apparel exports to the United States.

The trends in these diagrams provide some support for our hypotheses regarding the effect of quotas on product prices. We now apply the differencein-difference estimation to test for significant changes in the price and fabric content of apparel exports by AGOA recipients. 9.6.1

Quotas and Price Levels

The first objective of this section is to estimate if the expiration of the MFA reduced average US import unit values from quota-constrained countries who are predicted to have shifted apparel production toward lower-priced products. The equation used to identify these price effects is the differencein-difference specification of equation (20), except that we replace DAg with a dummy variable Dquotacntry for quota-constrained countries and D01 with a post-2005 dummy variable (D05). Table 9.6 presents the results. In line with theoretical predictions (and the price trends in figure 9.8 and figure 9.9), quota-constrained countries responded to the end of the MFA by reducing the quality of their apparel exports by shifting toward lower-priced varieties and products. The average unit value of US apparel imports from the top four most quota-constrained countries declined by 31.9 log points relative to other countries after 2005 (see row 1 of column [1] of table 9.6; see also Brambilla, Khandelwal, and Schott [2010]; Harrigan and Barrows [2009]).43 This arises from a combination of across-product shifts of imports toward lower-price products and within-product shifts toward lower-priced 43. The decline for the top thirty quota-constrained countries is lower at 13.9 percent.

–0.319***

Base price emerging (1)

–0.087*** 0.073** 0.219*** –0.112

–0.107*** 0.093*** 0.217*** –0.112

0.294*** –0.317***

–0.567*** 0.422***

AGOA emerging (3)

–0.580***

Base fabric emerging (2)

0.273** –0.260**

0.209*** –0.104

–0.098*** 0.084***

0.111

–0.573***

LDC AGOA emerging (4)

Marginal impact of the ending of the MFA on import unit values and fabric intensity in apparel-eligible AGOA beneficiaries

Impact of ending of MFA on US import prices 1 Quota constrained relative to control (D05 × Dquotacntry) 2 AGOA relative to control (D05 × DAg) 3 LDC AGOA relative to other AGOA (D05 × Dldc) Marginal impact of ending of MFA on fabric intensity Control group 4 D05 × ln(pf) 5 D05 × ln(pva) Quota-constrained group relative to control group 6 D05 × Dquotacntry x ln(pf) 7 D05 × Dquotacntry x ln(pva) AGOA countries relative to control group 8 D05 × DAg × ln(pf) 9 D05 × DAg × ln(pva) LDC AGOA countries relative to other AGOA 10 D05 × Dldc × ln(pf) 11 D05 × Dldc × ln(pva)

Country sample

Table 9.6

N F Fixed effects  

 

Notes: Estimates are robust to heteroskedasticity. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Other variables ln(pf) ln(pva) Dquotacntry × ln(pf) Dquotacntry × ln(pva) DAg × ln(pf) DAg × ln(pva) Dldc × ln(pf) Dldc × ln(pva) ln(GDP/capita), PPP ln(e) ln(Pcompete) ln(US ppi) ln(1 + t)

12 13 14 15 16 17 18 19 20 21 22 23 24

0.443*** 0.288*** 0.599*** –0.898***

0.578*** –0.886*** 0.038*** 0.230* –0.681*** 102,208 131 country/product year

0.619*** 0.381***

0.152*** –1.141*** 0.028** 0.216* –0.637*** 102,208 168 country/product year

102,208 108 country/product year

0.580*** –0.980*** 0.037*** 0.234* –0.661***

0.468*** 0.332*** 0.631*** –0.938*** 0.432*** –0.324***

102,208 106 country/product year

0.373*** –0.297*** 0.534*** –0.918*** 0.038*** 0.220* –0.701***

0.430*** 0.320*** 0.628*** –0.917***

384

Lawrence Edwards and Robert Z. Lawrence

varieties. The expiration of the MFA therefore adversely affected the competitiveness of non-quota-constrained countries such as Lesotho that produced low-priced products in response to the MFA. 9.6.2

Quotas and Fabric-Intensity

We now test for changes in the fabric intensity of apparel imports in response to the expiration of the MFA. Our theory predicts a rise in the fabric content of exports by previously quota-constrained countries relative to AGOA beneficiaries and other non-quota-constrained exporters. We commence with the simplest difference-in-difference specification to identify changes in the fabric intensity of quota-constrained countries in response to the end of the MFA. The specification is similar to that of equation (21), except, as above, a post-2005 dummy variable is used and we also include various interactions between Dquotacntry and value added and fabric prices covering the pre- and post-MFA period. We are interested in two effects: (a) the change in fabric intensity of exports of the control group (non-quota-constrained emerging economies) after January 2005, and (b) the change in fabric intensity of exports of the quota-constrained group relative to the control group. The first effect is given by the interactions between the post-MFA dummy (D05) and fabric and value-added prices in rows 4 and 5 of table 9.6. The second effect is given by the triple interaction between D05, Dquotacntry, and fabric- and value-added prices in rows 6 and 7. The results in rows 4 and 5 in column (2) indicate a decline in the fabric intensity of apparel exports to the United States from emerging economies after 2005. The coefficient on log fabric prices declines by 10.9 log points, while the coefficient on log value-added prices rises by 9.3 log points. This change is consistent with our theory that predicts shifts out of fabricintensive products by non-quota-constrained countries in response to the removal of quotas. Our estimates also reveal significant increases in the fabric intensity of US apparel imports from the most quota-restricted countries.44 This is revealed by the significant positive coefficient of 0.217 on the interaction term (D05 × Dquotacntry × ln(pf)) in rows 6 and 7 of column (3). Apparel exports from Bangladesh, India, China, and Indonesia therefore became more responsive to fabric price fluctuations after 2005 relative to all other emerging economies. We infer from this result that the fabric intensity of apparel exports to the United States from these previously quota-constrained countries has risen. The next two estimates focus on identifying the MFA effect on prices and fabric intensity for AGOA beneficiaries relative to other non-quotaconstrained emerging economies. We do this by separately including addi44. There is no significant difference from the control group for the top thirty most quotarestricted countries.

AGOA Rules

385

tional triple interactions for the AGOA group (see rows 8 and 9, column [3]) and the LDC AGOA group (see rows 10 and 11, column [4]). The estimates produce interesting results. The data suggest that the expiration of the MFA led to a rise in the fabric intensity of AGOA exports relative to other emerging economies. This is revealed by the rising responsiveness of US import prices from AGOA recipients to changes in fabric prices relative to the control group. If we focus on lesser-developed beneficiary countries (column [4]), we get a similar result. Clearly AGOA countries have responded differently to other non-quotaconstrained emerging economies. This is precisely what our theory predicts would happen under AGOA preferences. We found earlier that AGOA resulted in no changes in fabric intensity of exports by beneficiary countries. Our explanation was that these countries were already specialized in fabric-intensive low-value-added apparel products as a result of the incentives introduced by the MFA quotas. With the end of the MFA, China and other quota-constrained countries moved into the fabric-intensive products they were previously discouraged from exporting under the quotas. This led to increased competition in fabric-intensive products that non-quota-constrained countries specialized in under the MFA. The response by these countries was to reduce the fabric intensity of their apparel exports. The AGOA recipients, however, are an exception. Why? An explanation based on our theory is that AGOA preferences insulated the recipients in the most fabric-intensive products as the effective preferences in these products are the greatest. The effect of AGOA on fabric intensity is only revealed in our estimates once MFA is removed, as prior to this we had an identification problem as both AGOA and MFA encouraged specialization in fabric-intensive products. In sum, the MFA induced AGOA countries to specialize in low-valueadded, high-fabric-content apparel products. The AGOA preferences, and particularly the third-country fabric provision were expected, according to our theory, to compound this specialization in low-value-added, fabricintensive varieties and products. We do not find evidence of significant changes in the fabric content of apparel exports in response to the AGOA preferences. Rather, the apparel producers in AGOA recipient countries responded by increasing exports of existing products. The dependence of these exports on the tariff preferences and quota restrictions in competing countries made AGOA recipients and other nonconstrained emerging economies very vulnerable to the ending of the MFA. The elimination of quotas (quotas were reintroduced on Chinese exports in later 2005) induced China and other previously quota-restricted countries to downgrade product quality and increase exports of those products and varieties that AGOA countries were specialized in. However, the effect on fabric content of AGOA-recipient exports was insulated relative to other countries by the AGOA preferences that grant the greatest effective prefer-

386

Lawrence Edwards and Robert Z. Lawrence

ences in fabric-intensive products. The AGOA preferences helped mitigate the effects of the expiration of the MFA. 9.7

Conclusions

Lesotho and other lesser-developed beneficiary countries enjoyed rapid growth in their clothing exports to the United States as a result of the thirdcountry fabric provision of AGOA. Although adversely impacted by the expiration of the MFA and the recession in the United States, the clothing industries of these least developed African countries have clearly benefited from the provisions. But these economies have not enjoyed the more dynamic upgrading and spill-over benefits that might have been hoped for. Most of the export growth has come in the products that these countries were already producing. Success in the US clothing market has also not translated into success in other clothing markets or in success in exporting other labor-intensive products. The LDBCs have generally remained specialized in a small number of garment categories that are particularly favored by the preferences. These typically embody low value added in sewing and are relatively intensive in fabric. Although the AGOA program has operated for a decade, it is unlikely that most of the industry in these poor sub-Saharan countries could survive without the special rule. This experience provides important lessons. Trade preferences can have important effects on export success. First, they can offer powerful inducements to beneficiary exporters that are financed through foregone tariff revenues by developed countries rather than taxpayers in developing countries. Second, by providing a form of infant industry protection in export rather than domestic markets, they ensure that products have to meet the requirements of consumers in advanced economies. And third, since they are externally imposed, they do not give rise to domestic rent seeking. The positive response to AGOA’s special rule highlights the importance of providing exporters with access to inputs at world prices. Requiring exporters to use expensive inputs can seriously impede their competitiveness. This is clearly seen in the contrast between Lesotho’s prowess in the United States where it is allowed to use fabrics that are priced at world prices, with its weak performance in the EU and SACU where it is not. The positive response to AGOA highlights the restrictive nature of other rules of origin that have been imposed on least developed country exports. Allowing LDBCs to use imported fabrics provided powerful effective subsidies for clothing exports. This served to compensate producers in poor countries for the lower productivity of domestic workers and other institutional and infrastructural deficiencies. The fact that the program has operated smoothly without problems relating to trade deflection demonstrates the potential gains from modifying the restrictive rules that continue to limit the benefits to poor countries from programs such as the EBA program of the European Union. Such changes

AGOA Rules

387

would create more realistic possibilities that the least developed countries could participate in global production chains. It would be particularly welcome given the problems faced by these countries as a result of the expiration of the MFA. In the Doha Round, it is recognized that lower MFN tariffs will result in preference erosion. But typically studies have suggested that the effects would not be large.45 However, if the models that are used to estimate the impact of erosion fail to take the third-country fabric provision into account, they could seriously underestimate the impact on the effective protection provided to the lesser-developed AGOA recipients. The experience also shows, however, that trade preferences are not a panacea. The outcomes associated with the special rule conform to those suggested by theory. The special rule distorts decisions on value addition and fabric use in opposite directions. On the one hand, the incentives are most powerful in lower-quality products that require less value addition. This may limit the dynamic benefits that are hoped for from these preferences by discouraging skills development and other forms of quality upgrading. On the other hand, it encourages the use of more expensive fabrics. This makes it less likely that there will be backward linkages into domestic textile industries that are still at rudimentary stages of development. Preferences are thus an opportunity, but not a substitute for, more comprehensive industrial strategies that involve complementary domestic policies to improve private and governmental capabilities. This does not mean that these preferences are unimportant, but suggests they are unlikely to be sufficient. In addition, problems arise when most of the entrepreneurs taking advantage of the preferences are foreign, with many other crucial parts of the value chain being provided thousands of miles away. The experience analyzed in this chapter is a case study of the links between trade and growth— a topic that has been the subject of considerable empirical investigation. This example highlights the obvious, but often ignored consideration, that both trade and growth are quintessentially endogenous variables rather than policy instruments and suggests that the reasons for trade are likely to be important in the impact on growth. Even if on average trade and growth are associated, and even if on average trade may cause growth, the widely used proposition that trade leads to growth should not be used as an unconditional forecast. The precise reasons for trade, and the other domestic conditions and policies that are associated with it, are likely to play key roles in the growth impact. In the case of Lesotho and other AGOA countries, utilizing preferences may lead to more trade but are not a substitute for the more difficult challenges of developing more comprehensive development strategies. In sum, the slogan of “trade not aid” can be misleading. Trade preferences may help create the conditions for growth, but they are not sufficient. 45. For estimates of the impact of preference erosion, see IMF (2003), Olarreaga and Özden (2005), Hoekman and Prowse (2005), and Grynberg and Silva (2004).

0.329**

0.199

0.358***

0.406***

0.469***

0.231**

0.752**

0.439***

0.225***

0.574***

0.241

0.736***

1.015***

0.647***

–0.530***

–0.282***

–0.319***

0.071

–0.149

–0.384

–0.454***

–0.105*

–0.179***

–0.299

–0.442***

0.037

–0.375***

ln(pf)

–0.186

Coefficients ln(GDP worker)

0.279***

–0.034

0.286***

0.381***

0.298***

0.273***

0.370***

–0.13

0.255***

0.331***

0.358***

0.326***

0.420***

0.217***

ln(pva)

Price equation estimates by four-digit HS level

6101 Men’s or boys’ overcoats etc., knit or crochet 6102 Women’s or girls’ overcoats etc., knit or crochet 6103 Men’s or boys’ suits, ensembles, etc., knit or crochet 6104 Women’s or girls’ suits, ensembles, etc., knit or crochet 6105 Men’s or boys’ shirts, knitted or crocheted 6106 Women’s or girls’ blouses & shirts, knit or crochet 6107 Men’s or boys’ underpants, pjs, etc., knit or crochet 6108 Women’s or girls’ slips, pjs, etc., knit or crochet 6109 T-shirts, singlets, tank tops, etc., knit or crochet 6110 Sweaters, pullovers, vests, etc., knitted or crocheted 6111 Babies’ garments & accessories, knitted or crocheted 6112 Track suits, ski suits & swimwear, knit or crochet 6113 Garments, knit, etc., coated, etc., rubber, plastic, etc. 6114 Garments nesoi, knitted or crocheted

HS4 code Description

Table 9A.1

Appendix

–1.028***

–1.180***

–1.101***

–0.829***

–0.819***

–0.632***

–0.912***

–1.377***

–0.555***

–0.678***

–0.812***

–0.672***

–0.704***

–0.660***

ln(e)

0.016

0.076

0.079**

0.229***

0.056***

0.023

0.064*

0.262**

0.161***

0.150**

0.070***

0.062**

0.107**

0.062

ln(Pcompete)

–0.181

0.989

1.894***

1.329**

0.406**

0.780***

0.238

0.274

0.539

0.649*

1.444***

0.774***

–0.546

–0.439

–0.43

4.128*

–0.367

1.139

–0.423***

0.042

–1.884***

1.029

–0.557*

–0.293

–0.244

–1.265***

0.047

0.873**

ln(US ppi) ln(1 + t)

9,940

2,653

6,478

5,254

29,316

14,877

10,262

2,321

5,332

4,272

24,243

8,136

3,833

2,890

N

33.5

8.76

24.4

20.8

98.4

39.1

33.7

3.96

15.1

14.8

76.2

17.6

14.3

7.19

F

0.031

0.030

0.033

0.048

0.031

0.025

0.033

0.018

0.027

0.035

0.030

0.024

0.035

0.024

0.484

0.162

0.004

0.054

0.041

0.067

0.794

0.902

0.651

0.118

0.001

0.122

0.719

0.169

r2 HOD 1

0.004

0.036

0.067

0.101

0.003

0.000

0.003

0.008

0.114

0.001

0.011

0.714

0.048

0.072

δ1 + δ2 = β2

0.114

0.099

0.239

0.082

0.100

0.027

0.000

0.658

0.731

0.928

0.772

0.002

0.397

0.003

Erate = tariff

Hypothesis tests (p-value)

6115 Pantyhose, socks & other hosiery, knit or crochet 6116 Gloves, mittens and mitts, knitted or crocheted 6117 Made-up clothing access nesoi. parts, etc., knit, etc. 6201 Men’s or boys’ overcoats, cloaks, etc., not knit, etc. 6202 Women’s or girls’ overcoats, etc., not knit or crochet 6203 Men’s or boys’ suits, ensembles, etc., not knit, etc. 6204 Women’s or girls’ suits, ensembles, etc., not knit, etc. 6205 Men’s or boys’ shirts, not knitted or crocheted 6206 Women’s or girls’ blouses, shirts, etc., not knit, etc. 6207 Men’s or boys’ undershirts, etc., not knit or crochet 6208 Women’s or girls’ slips, etc., not knit or crochet 6209 Babies’ garments & accessories, not knit or crochet 6210 Garments of felt, etc., or fabric impregnated, etc. 6211 Track suits, ski suits & swimwear, not knit. etc. 6212 Bras, girdles, garters, etc., knitted, etc., or not 6213 Handkerchiefs 6214 Shawls, scarves, mufflers, mantillas, veils, etc.

0.151*

0.689***

0.367**

0.343***

0.410***

0.238***

0.447***

0.368***

0.453***

0.352

0.306***

0.602***

0.429***

0.386***

0.509*** 0.147

–0.04

–0.348**

–0.677***

–0.561***

0.032

–0.043

0.057

–0.009

0.113

0.028

–0.068

–0.095

–0.358**

–0.299***

–0.046

–0.117 –0.758**

–0.097

0.208*

0.390*** 0.053

0.289***

0.149**

0.238**

0.195***

0.263

0.205***

0.319***

0.368***

0.358***

0.354***

0.351***

0.392***

0.441***

0.609***

–0.099

–0.854*** –1.064***

–0.674***

–0.577***

–1.258***

–0.577***

–0.987***

–0.704***

–0.641***

–0.822***

–0.573***

–0.788***

–0.638***

–0.678***

–1.108***

–0.718***

0.156*

–0.049 –0.044

0.079***

–0.052

–0.084

0.116***

–0.071

0.052*

–0.035

0.044***

0.100***

0.013

0.068**

0.096***

–0.054

0.047

2.429

1.165* 11.178***

0.582***

–0.093

0.923

–0.463

0.628

0.989***

0.924***

0.797***

0.698***

–0.509

–0.403

3.312***

0.725

1.775**

–0.055

–1.159** 1.781

0.555**

–0.137

–0.923

–1.424***

–2.315

–0.687***

–0.718***

–0.504***

–0.579***

–0.438**

–0.225

–0.866

–1.525***

–0.564

3,625

5,872 1,049

28,191

7,067

3,786

7,748

2,337

7,965

6,467

50,694

22,945

14,450

12,265

5,824

5,314

5,535

2.66

21.8 4.38

54.1

5.03

19.9

15.5

4.27

38.3

19.2

245

72.3

57

34.8

15.7

21.5

18

0.007

0.032 0.046

0.018

0.007

0.054

0.021

0.024

0.042

0.029

0.044

0.030

0.035

0.026

0.023

0.032

0.030

0.246

0.101 0.001

0.144

0.315

0.329

0.064

0.893

0.037

0.087

0.019

0.026

0.294

0.089

0.000

0.132

0.036

0.448

0.382 0.001

0.937

0.978

0.002

0.077

0.105

0.085

0.168

0.510

0.199

0.183

0.050

0.304

0.779

0.379

0.162 (continued)

0.072 0.428

0.001

0.561

0.118

0.001

0.249

0.119

0.198

0.043

0.398

0.296

0.556

0.435

0.005

0.728

(continued)

–0.157

0.897***

0.102

–0.699

1.500*

0.656

2.189

0.126

0.545***

–0.027

–0.571**

–1.000**

0.359

–0.243

0.498

–0.099

–0.300***

ln(pf)

–0.817***

Coefficients ln(GDP worker)

0.278***

0.214

–0.012

–0.117

–0.629

0.713**

0.323**

0.232*

0.733***

ln(pva)

–0.817***

–0.595**

–2.288**

–0.75

–0.838

–0.06

–0.627***

–0.945***

–0.472***

ln(e)

0.151***

0.228**

–0.046

0.136*

–0.054

–0.012

–0.006

–0.033

0.126**

ln(Pcompete)

0.554**

0.754

1.255

3.266*

1.703

4.351**

8.474***

–1.059

0.823

–1.285**

–0.658*

–81.49***

–0.79

–2.434

1.158

–0.019

–0.737

0.318

ln(US ppi) ln(1 + t)

9,802

1,653

243

671

381

691

4,782

2,991

1,974

N

24.2

4.05

9.51

3.31

0.885

2.79

9.05

3.62

6.86

F

0.027

0.019

0.347

0.043

0.027

0.045

0.020

0.013

0.031

0.044

0.683

0.535

0.074

0.401

0.045

0.000

0.305

0.519

r2 HOD 1

0.867

0.062

0.657

0.382

0.893

0.900

0.034

0.110

0.247

δ1 + δ2 = β2

0.039

0.557

0.000

0.657

0.528

0.558

0.652

0.366

0.284

Erate = tariff

Hypothesis tests (p-value)

Note: Year fixed effects are not included as the fabric costs do not vary across products for some of the HS four-digit groups. Estimates are robust to heteroskedasticity. Overall, the coefficients are broadly consistent with expectations. In most sectors the coefficient on fabric prices is positive and ranges from 0.23 to just over 1. Similarly, the coefficients on value-added prices and the exchange rate are mostly of the correct sign. The estimates for headgear (HS 64) and footwear (HS 65) products are poor, but is likely that the fabric costs indices do not adequately reflect the inputs used in the production of these products. For example, HS 6406 covers parts of footwear; removable insoles, heel cushions, and similar articles; gaiters, leggings, and so forth. HS65 covers headgear products often comprising of felt, strips of any material, lace, and the like. These products also make up a very small proportion of AGAO country exports. Most estimates fail to reject the homogeneity and symmetric pass-through (both the tariffs and exchange rate and exchange rate and production costs) hypotheses. Each hypothesis is rejected at most twelve times (out of forty). ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

6215 Ties, bow ties & cravats, not knitted or crocheted 6216 Gloves, mittens and mitts, not knitted or crocheted 6217 Made-up clothing access nesoi, garment, etc., parts nesoi 6406 Parts of footwear: insoles, etc.: gaitors, etc., parts 6501 Hat forms/bodies, hoods, plateaux & manchons of felt 6502 Hat shapes, plaited or assembled strips any material 6503 Felt hats & other felt headgear from heading 6501 6504 Hats & other headgear, plaited/ assembled strips any material 6505 Hats & headgear, knit, etc., lace, etc., in pc, hair net

HS4 code Description

Table 9A.1

AGOA Rules

391

References Ahmad, Munir. 2007. “Impact of Origin Rules for Textiles and Clothing on Developing Countries.” ICTSD Issue Paper no. 3, International Centre for Trade and Sustainable Development. http://www.ictsd.org/downloads/2009/02/impact-of -origin-rules-for-textiles-and-clothing-on-developing-countries1.pdf. Alchian, Armen A., and William R. Allen. 1964. University Economics. Belmont, CA: Wadsworth. Bennet, Mark. 2006. “Lesotho’s Export Textiles & Garment Industry.” In The Future of the Textile and Clothing Industry in Sub-Saharan Africa, edited by Herbert Jauch and Rudolf Traub-Merz. Bonn: Friedrich-Ebert-Stiftung. Brambilla, Irene, Amit K. Khandelwal, and Peter K. Schott. 2010. “China’s Experience under the Multi-Fiber Arrangement (MFA) and the Agreement on Textiles and Clothing (ATC).” In China’s Growing Role in World Trade, edited by Robert C. Feenstra and Shang-Jin Wei. Chicago: University of Chicago Press. Brenton, Paul, and Takako Ikezuki. 2005. “The Initial and Potential Impact of Preferential Access to the US Market under the African Growth and Opportunity Act.” Policy Research Working Paper no. 3262, Washington, DC, World Bank. Collier, Paul, and Anthony J. Venables. 2007. “Rethinking Trade Preferences: How Africa Can Diversify Its Exports.” World Economy 30 (8): 1326–45. Dixit, Avinash K., and Victor Norman. 1980. Theory of International Trade. Cambridge: Cambridge University Press. Dixit, Avinash K., and Joseph E. Stiglitz. 1977. “Monopolistic Competition and Optimum Product Diversity.” American Economic Review 67 (3): 297–308. Dornbusch, Rudiger, Stanley Fischer, and Paul A. Samuelson. 1977. “Comparative Advantage, Trade and Payments in a Ricardian Model with a Continuum of Goods.” American Economic Review 65:297–308. Falvey, Rodney E. 1979. “The Composition of Trade within Import-Restricted Product Categories.” Journal of Political Economy 87 (5, Part 1): 1105–14. Feenstra, Robert C. 1988. “Quality Change under Trade Restraints in Japanese Autos.” Quarterly Journal of Economics 103 (1): 131–46. ———. 2004. Advanced International Trade: Theory and Evidence. Princeton, NJ: Princeton University Press. FIAS. 2006. “The Competitiveness of Regioal and Vertical Integration of Lesotho’s Garment Industry.” Washington, DC: IFC and World Bank. Frazer, Garth, and Johannes Van Biesbroeck. 2010. “Trade Growth under the African Growth and Opportunity Act.” Review of Economics and Statistics 92 (1): 128–44. Gereffi, G. 1999. “International Trade and Industrial Upgrading in the Clothing Commodity Chain.” Journal of International Economics 48:37–70. Gibbon, Peter. 2003. “The African Growth and Opportunity Act and the Global Commodity Chain for Clothing.” World Development 31 (1): 1809–27. Goldberg, Pinelopi K., and Michael M. Knetter. 1997. “Goods Prices and Exchange Rates: What Have We Learned?” Journal of Economic Literature 35:1243–72. Gopinath, Gita, and Roberto Rigobon. 2008. “Sticky Borders.” Quarterly Journal of Economics 123 (2): 531–75. Grynberg, Roman, and Sacha Silva. 2004. Preference-Dependent Economies and Multilateral Liberalization: Impacts and Options. London: Commonwealth Secretariat. Harrigan, James, and Geoffrey Barrows. 2009. “Testing the Theory of Trade Policy: Evidence from the Abrupt End of the Multifiber Arrangement.” Review of Economics and Statistics 91 (2): 282–94.

392

Lawrence Edwards and Robert Z. Lawrence

Hausmann, Ricardo, Jason Hwang, and Dani Rodrik. 2007. “What You Export Matters.” Journal of Economic Growth 12 (1): 1–25. Hoekman, Bernard, and Susan Prowse. 2005. “Policy Response to Preference Erosion: From Trade as Aid to Aid for Trade.” Paper presented at the international symposium “Preference Erosion: Impacts and Policy Responses,” Geneva, June 13–14. Hummels, David, and Peter J. Klenow. 2005. “The Variety and Quality of a Nation’s Exports.” American Economic Review 95 (3): 704–23. Hummels, David, and Alexandre Skiba. 2004. “Shipping the Good Apples out? An Empirical Confirmation of the Alchian-Allen Conjecture.” Journal of Political Economy 112 (6): 1384–402. Hwang, Jason Jaemin. 2007. “Patterns of Specialization and Economic Growth.” PhD diss., Harvard University. International Monetary Fund (IMF). 2003. “Financing Losses from Preference Erosion, Note on Issues raised by Developing Countries in the Doha Round.” Communication to the WTO from the International Monetary Fund, WT/TF/COH/ 14 February. Prepared by Arvind Subramanian. Krishna, Kala. 1987. “Tariffs versus Quotas with Endogenous Quality.” Journal of International Economics 23 (1–2): 97–112. Krugman, Paul R. 1979. “Increasing Returns, Monopolistic Competition, and International Trade.” Journal of International Economics 9 (4): 469–79. ———. 1980. “Scale Economies, Product Differentiation, and the Pattern of Trade.” American Economic Review 70 (5): 950–59. Lall, Sanjaya. 2005. “FDI, AGOA, and Manufactured Exports from a Land-Locked Least Developed African Economy: Lesotho.” Journal of Development Studies 41 (6): 998–1022. Maloney, Christopher. 2006. “All Dressed up with No Place to Go? Lesotho’s Rollercoaster Experience with Apparel.” Harvard Kennedy School MPA/ID Second Year Policy Analysis. Marazzi, M., N. Sheets, R. Vigfusson, J. Faust, J. Gagnon, J. Marquez, R. Martin, T. Reeve, and J. Rogers. 2005. “Exchange Rate Pass-Through to US Import Prices: Some New Evidence.” International Finance Discussion Papers no. 833. http:// www.federalreserve.gov/pubs/ifdp/2005/833/default.htm. Mattoo, Aadditya, Devesh Roy, and Arvind Subramanian. 2003. “The Africa Growth and Opportunity Act and Its Rules of Origin: Generosity Undermined?” World Economy 26 (6): 829–51. Morris, Mike. 2006. “China’s Dominance of Global Clothing and Textiles: Is Preferential Trade Access an Answer for Sub-Saharan Africa?” IDS Bulletin 37 (1): 89–97. Morris, Mike, and Leanne Sedowski. 2006. “Report on Government Responses to New Post-MFA Realities in Lesotho.” Report for Institute of Global Dialogue. http://www.cssr.uct.ac.za/sites/cssr.uct.ac.za/files/pubs/prism_lesotho_clothing rept.pdf. Ng, Francis, and Alexander Yeats. 1996. “Open Economies Work Better: Did Africa’s Protectionist Policies Cause Its Marginalisation in World Trade?” Policy Research Working Paper no. 1636, Washington, DC, World Bank. Nouve, Kofi. 2005. “Estimating the Effects of AGOA on African Exports Using a Dynamic Panel Analysis.” World Bank Working Paper, July, Washington, DC, World Bank. Olarreaga, Marcelo, and Çaglar Özden. 2005. “AGOA and Apparel: Who Captures the Tariff Rent in the Presence of Preferential Market Access?” World Economy 28 (1): 63–77.

AGOA Rules

393

Pierce, Justin R., and Peter K. Schott. 2009. “Concording US Harmonized System Categories over Time.” CES Working Paper no. 09-11, US Census Bureau Center for Economic Studies. Portugal-Perez, Albert. 2008. “The Costs of Rules of Origin in Apparel: African Preferential Exports to the United States and the European Union.” UNCTAD Policy Issues in International Trade and Commodities Study Series no. 39, United Nations Conference on Trade and Development. Rolfe, Robert J., and Douglas P. Woodward. 2005. “African Apparel Exports, AGOA, and the Trade Preference Illusion.” Global Economy Journal 5 (3): 1–28. Salm, Andrew, William J. Grant, Thuso J. Green, John R. Haycock, and John Raimondo. 2002. “Lesotho Garment Industry Subsector Study.” Report for the Government of Lesotho, and funded by the Department for International Development. http://www.sarpn.org/documents/d0001166/. Sandrey Ron, David Maleleka, Adelaide Matlanyane, and Dirk Ernst van Seventer. 2005. “Lesotho: Potential Export Diversification Study.” Working Paper. http://www.tralac.org/publications/article/4562-lesotho-potential-export -diversification-study.html. Schott, Peter K. 2004. “Across-Product versus within-product Specialization in International Trade.” Quarterly Journal of Economics 119 (2): 646–77. Seidman, Gay W. 2009. “Labouring under an Illusion? Lesotho’s ‘Sweat-free’ Label.” Third World Quarterly 30 (3): 581–98. Seyoum, Belay. 2007. “Export Performance of Developing Countries under the Africa Growth and Opportunity Act.” Journal of Economic Studies 34 (6): 515–33. Wang, Zhen Kun, and Alan Winters. 1998. “Africa’s Role in Multilateral Trade Negotiations: Past and Future.” Journal of African Economies 7:1–33. Wood, Adrian, and Jörg Mayer. 2001. “Africa’s Export Structure in a Comparative Perspective.” Cambridge Journal of Economics 5 (3): 369–94.

Contributors

Franklin Allen The Wharton Finance Department University of Pennsylvania 3620 Locust Walk Philadelphia, PA 19104

Taryn Dinkelman Department of Economics H. Box 6106 Dartmouth College Hanover, NH 03755

Dimitris Batzilis Department of Economics University of Chicago Chicago, IL 60637

Pascaline Dupas Department of Economics Stanford University 579 Serra Mall Stanford, CA 94305–6072

Elena Carletti Department of Finance Bocconi University Via Roentgen 1 20136 Milan Italy Lisa D. Cook Department of Economics Michigan State University 106 Old Botany East Lansing, MI 48824–1038 Robert Cull The World Bank 1818 H Street, NW Washington, DC 20433

William Easterly Department of Economics New York University 19 W. 4th Street, 6th floor New York, NY 10012 Lawrence Edwards Faculty of Commerce University of Cape Town Rondebosch, 7701 South Africa Sebastian Edwards UCLA Anderson Graduate School of Business 110 Westwood Plaza, Suite C508 Box 951481 Los Angeles, CA 90095–1481

395

396

Contributors

Sarah Green High-Level Task Force for the ICPD 125 Maiden Lane, 9th Floor New York, NY 10038–4730

Prime Nyamoya OGI Consulting Group PO Box 2085 Bujumbura, Burundi

Simon Johnson MIT Sloan School of Management 100 Main Street, E52–562 Cambridge, MA 02142

Emily Oster Department of Economics Brown University 64 Waterman Street Providence, RI 02912

Sebnem Kalemli-Ozcan Department of Economics University of Maryland Tydings Hall 4118D College Park, MD 20742–211 Anthony Keats Department of Economics Wesleyan University 238 Church Street Middletown, CT 06459 Robert Lawrence Kennedy School of Government Harvard University 79 JFK Street Cambridge, MA 02138 Isaac Mbiti Frank Batten School of Leadership and Public Policy University of Virginia P. O. Box 400893, Garrett Hall, 235 McCormick Rd Charlottesville, VA 22904–4893 Léonce Ndikumana Political Economy Research Institute (PERI) University of Massachusetts at Amherst Gordon Hall, 418 N. Pleasant Street Amherst, MA 01002 Janvier D. Nkurunziza Commodity Research and Analysis Section Special Unit on Commodities, UNCTAD Palais des Nations, 8–14 Avenue de la Paix CH-1211 Geneva, Switzerland

Jun Qian Shanghai Advanced Institute of Finance Shanghai Jiaotong University 211 West Huaihai Road Shanghai, P.R. China 200030 Ariell Reshef CNRS, 106-112 Boulevard de l’Hôpital 75647 Paris cedex 13, France Paris School of Economics 48 Boulevard Jourdan, 75014 Paris, France Jonathan Robinson Department of Economics University of California, Santa Cruz 457 Engineering 2 Santa Cruz, CA 95064 Lemma Senbet Robert H. Smith School of Business University of Maryland College Park, MD 20742 Bent E. Sørensen Department of Economics University of Houston 204 McElhinney Hall Houston, TX 77204 Rebecca Thornton Department of Economics University of Illinois at UrbanaChampaign 18 David Kinley Hall 1407 W. Gregory Dr. Urbana, IL 61801

Contributors Patricio Valenzuela Department of Industrial Engineering University of Chile Republic 701 Santiago, Chile David N. Weil Department of Economics Box B Brown University Providence, RI 02912

Deric Zanera Malawi National Statistical Office P.O. Box 333 Zomba, Malawi

397

Author Index

Abiad, A., 157 Adegbite, E. O., 163 Aghion, P., 259 Ahmad, M., 345 Aker, J. C., 217n1, 247, 255, 258, 264 Alchian, A. A., 358 Alesina, A., 39n19, 217n2 Alfaro, L., 183, 185n3 Allen, F., 17, 51, 58, 59 Allen, W. R., 358 Anderson, S., 79n19 Anyanzwa, J., 264 Aportela, F., 64 Arkolakis, C., 301n8, 304n9 Armendáriz, B., 66n7 Arnquist, S., 217 Artopoulos, A., 299, 322 Aryeetey, E., 14n1 Atkinson, J., 90 Balakrishnan, A., 217n1 Baland, J.-M., 79n19 Baldwin, R., 300n5 Banerjee, A., 64, 67, 88, 183, 184n2, 217n2 Baqir, R., 217n2 Barro, R. J., 111 Barrows, G., 362, 379, 379n42, 381 Bartelsman, E., 124, 183, 185n3 Barth, J. R., 157, 163 Beck, T., 14n1, 14n2, 14n3, 16n6, 18, 18n8, 19, 21n11, 25n13, 45n21, 52, 157,

159n3, 159n4, 167, 167n15, 175, 178, 184, 270 Bekaert, G., 45n20 Benartzi, S., 90 Bennet, M., 343, 355 Benson, T., 220 Beny, L., 178 Bernard, A., 300, 301n8 Besedes, T., 301 Besley, T., 79n19, 217n2 Bigsten, A., 124 Bikker, J., 178 Bower, J. L., 259 Brambilla, I., 362, 362n21, 371, 372n29, 379, 381 Brenton, P., 348n10 Brownbridge, M., 72 Bruhn, M., 64 Burgess, R., 64, 270 Buys, P., 215, 216 Cadot, O., 332n20 Cameron, A. C., 239 Caprio, G., Jr., 71, 157, 157n1, 163 Carroll, G., 90 Chaia, A., 52, 63, 63n2 Charlton, A., 183, 185n3 Chrétien, J. P., 132 Christensen, C. M., 259 Clarke, G. R. G., 14n3 Clerides, S. K., 300 Cletus, A., 284

399

400

Author Index

Coate, S., 79n19 Cohen, M., 65n6 Collier, P., 355, 371 Collins, D., 88, 288 Cook, L. D., 160n8, 160n9, 178 Crépon, B., 64, 67 Cull, R., 18n7, 19, 22, 66n7, 159n4 Daumont, R., 71, 72 Davoodi, H., 217n2 Deaton, A., 194 De Mel, S., 66n8, 67 Demirgüç-Kunt, A., 14n3, 16n6, 45n21, 52, 66n7, 67n9, 157, 167, 173, 175, 178, 184, 270 Dinkelman, T., 241 Dixit, A. K., 360n17, 361 Duflo, E., 66, 88, 90, 183, 288 Dupas, P., 64, 64n3, 66n8, 87 Easterly, W., 13, 21, 39n19, 217n2, 301, 322, 322n14 Eaton, J., 301n8, 304n9 Effron, L., 18n7, 19, 22 Eijkman, F., 251, 256, 278 Engberg, H. L., 284 Ezeoha, A., 161n10, 173 Faccio, M., 105 Fafchamps, M., 66n8 Falvey, R. E, 358, 362, 363n23, 364, 365 Feenstra, R. C., 358, 363, 363n22, 365, 372, 376n39 Feige, E. L., 284 Ferguson, M., 65n6 Ferreira, F., 64n3 Fisman, R., 105 Frazer, G., 348, 371 Freund, C., 321n13, 322n16 Friel, D., 322, 333, 341 Fuchs, M., 159n3 Gentzkow, M., 241 Gereffi, G., 345 Gibbon, P., 348n10 Gikunju, W., 259, 260 Godlonton, S., 222n7 Goldberg, P. K., 376n38 Gopinath, G., 376n39 Grynberg, R., 387n45 Gugerty, M. K., 79n19, 288

Habyarimana, J., 71 Hallack, J. C., 300, 322, 333 Haltiwanger, J., 124, 183, 185n3 Harrigan, J., 300n5, 362, 379, 379n42, 381 Harvey, C., 45n20 Hass, S., 248, 276 Hausman, J., 217n1 Hausmann, R., 301, 344 Hesse, H., 173 Heston, A., 194 Hoekman, B., 387n45 Honohan, P., 14n1, 63n1 Howitt, P., 259 Hsieh, C.-T., 104, 105n2, 124, 124n18, 183, 185, 185n4, 190 Hu, H., 178 Huizinga, H., 173, 175, 178 Hummels, D., 358, 361n18, 363n23, 363n24, 371, 376 Hwang, J. J., 344, 344n3 Ikezuki, T., 348n10 Isern, J., 160n8 Jack, W., 65, 67, 70n14, 217n1, 248, 251, 254, 270, 288 Jensen, B., 300 Jensen, R., 217n1, 241, 258 Jerome, A., 159n4 Johnson, S., 184, 185, 205 Johnston, D., Jr., 67 Kabbucho, K., 253, 259, 260, 263 Kamil, H., 15n4 Kanczuk, F., 183, 185n3 Karacaovali, B., 67n9 Karlan, D., 64, 90 Kasekende, L., 174n22 Kast, F., 90 Kaufmann, D., 22, 23 Kayisinga, J.-C., 323n17 Kendall, J., 52, 63, 251, 256, 278 Khandelwal, A. K., 362, 362n21, 371, 372n29, 379, 381 Khwaja, A., 104, 105, 124, 124n19, 217n2 Kimenyi, M. S., 217n2, 282, 283 King, R. G., 157, 167, 167n15 Klenow, P., 104, 105n2, 124, 124n18, 183, 185, 185n4, 190, 361n18, 371, 376 Klingebiel, D., 71, 157n1 Klonner, S., 217n1

Author Index

401

Kneiding, C., 66n7 Knetter, M. M., 376n38 Kortum, S., 304n9 Kraay, A., 22, 23 Kramarz, F., 304n9 Kremer, M., 66, 66n8, 90, 288 Krishna, K., 358, 363n22, 365n25 Krudy, J. P., 217n1 Krugman, P. R., 361 Kumar, A., 64n5

Moll, B., 183 Morawczynski, O., 248, 250, 253, 254, 259, 268, 269, 270, 276 Morduch, J., 66n7, 67 Morris, M., 355n14 Muendler, M.-A., 301n8 Mukuri, M., 17 Mukwana, P., 253, 259, 260, 263 Mwega, F., 278 Mylenko, N., 52, 63

Lach, S., 300 Laeven, L., 67n9 Lall, S., 350, 355, 355n14, 358, 364 Laurent, R. D., 284 Lederman, D., 21 Le Gall, F., 71, 72 Lemarchand, R., 106, 106n3 Leroux, F., 71, 72 Leuz, C., 105 Levine, R., 13, 14n2, 14n3, 18, 19, 21, 39n19, 45n21, 157, 163, 167, 167n15, 173, 175, 178, 217n2 Lewis, P., 159n2, 159n4 Lim, C., 126 Loayza, N., 14n2, 18, 22, 167n15 Loury, G., 79n19 Love, I., 64 Lundblad, C., 45n20

Nagarajan, G., 248, 276 Ndikumana, L., 104, 106, 106n3, 107, 108 Ndung’u, N. S., 282, 283 Ng, F., 343n1 Ngaruko, F., 103, 104, 105, 106, 106n3, 107, 112, 116, 124 Ngowi, R., 217n1 Nguyen Tien Hung, G., 65n6 Ng’weno, A., 248 Nissanke, M., 14n1 Njiraini, J., 264 Nkeshimana, L., 132 Nkurunziza, J. D., 103, 104, 105, 106, 106n3, 107, 112, 116, 124, 126 Nolen, P., 217n1 Norman, J. E., 360n17 Nouve, K., 348n10 Ntibazonkiza, R., 106, 106n3 Nyamoya, P., 115, 132

Maksimovic, V., 45n21, 184 Maloney, C., 354n12 Maloney, W. F., 21 Manjoo, F., 223n8 Marazzi, M., 376n39 Martinez Peria, M. S., 52, 174n20 Mas, I., 68, 248, 251, 256, 270, 278 Mastruzzi, M., 22, 23 Mattoo, A., 348, 359n15 Mauro, P., 217n2 Mayer, J., 359n16 Mbiti, I., 16n5, 70n14, 76, 217n1, 247, 255, 264, 278 McGreal, C., 217n1 McKenzie, D., 66n8, 67 McMillan, J., 184, 185, 205 Meier, S., 90 Melitz, M., 300 Meyer, R., 122 Mian, A., 104, 105, 124, 124n19 Mody, A., 157, 174n20

Oberholzer-Gee, F., 105 Ogunleye, A. A., 161n11 Olarreaga, M., 348, 387n45 Onaolapo, A. A., 173 Opiyo, D., 68n10 Oster, E., 241 Otchere, I., 14n1, 51 Özden, Ç., 348, 387n45 Pande, R., 64, 270 Peria, S. M., 270 Pickens, M., 248, 268, 269, 270, 276 Pierce, J. R., 349 Pierola, M. D., 322n16 Plyler, M., 248, 276 Pomeranz, D., 90 Ponce, A., 52, 63 Porteous, D., 270 Portugal-Perez, A., 348, 359n16, 371 Prina, S., 64

402

Author Index

Pritchett, L., 34 Prowse, S., 387n45 Prusa, T. J., 301 Pulver, C., 288 Radcliffe, D., 68 Rai, K., 15n4 Rajan, R., 14n2, 22, 157, 322 Ranciere, R., 22 Rauch, J. E., 299n4 Renelt, D., 21 Reshef, A., 301 Restuccia, D., 183 Rigobon, R., 376n39 Roberts, M. J., 299 Robinson, J., 64, 64n3, 66, 66n8, 87, 90, 288 Robinson, M. S., 122 Rocha, N., 321n13 Rodrik, D., 301, 344 Rogerson, R., 183 Rolfe, R. J., 348n10 Rosenberg, R., 66n7 Rosenzweig, M., 64n3 Rotman, S., 270 Roy, D., 348, 359n15 Rugwabiza, L., 126 Ruiz, C., 64 Sachs, J. D., 16, 20 Salm, A., 355 Sander, C., 253, 259, 260, 263 Sandrey, R., 354n11 Scarpetta, S., 124, 183, 185n3 Schady, N., 64n3 Schimmelpfennig, S., 322n15 Schott, P. K., 344, 349, 355, 362, 362n21, 371, 372n29, 379, 381 Schumpeter, J. A., 259 Sedowski, L., 355n14 Seidman, G. W., 354 Senbet, L., 14n1, 18n7, 51 Sensarma, R., 124n17 Seyoum, B., 348n10 Shapiro, J., 241 Shleifer, A., 105 Silva, S., 387n45 Skiba, A., 358, 363n23, 363n24 Slemrod, J., 251 Soludo, C., 160n7

Somanathan, R., 217n2 Somoye, R. O. C., 173, 179 Sorge, M., 18n7 Spindt, P. A., 284 Stein, H., 159n2, 159n4 Stiglitz, J. E., 361 Stuart, G., 65n6 Stulz, R. M., 173 Subramanian, A., 322, 348, 359n15 Suri, T., 65, 66, 67, 70n14, 217n1, 248, 251, 254, 270, 288 Szabo, A., 194 Tanzi, V., 217n2 Thaler, R., 90 Thornton, R., 222n7 Thyagarajan, S., 64n4 Trivedi, P. K., 239 Tybout, J. R., 299, 300 Vail, L., 220n4 Van Biesebroeck, J., 348, 371 Vaughn, P., 248, 250, 254 Venables, A. J., 355, 371 Venkatesan, J., 64n4 Vishny, R. W., 105 Wang, Z. K., 343n1 Warner, A., 16, 20 Warutere, P., 71 Watson, J., 299n4 Weil, D. N., 16n5, 70n14, 76, 278 Winters, A., 343n1 Wolpin, K., 64n3 Wood, A., 359n16 Woodruff, C., 66n8, 67, 184, 185, 205 Woodward, D. P., 348n10 Wurgler, J., 104, 124 Xu, L. C., 14n3 Yeats, A., 343n1 Yeh, E., 64n3 Zeller, M., 122 Zervos, S., 157 Zingales, L., 14n2, 22, 157 Zinman, J., 64 Zou, H.-F., 14n3

Subject Index

Note: Page numbers followed by “f ” or “t” refer to figures or tables, respectively. Africa Growth and Opportunities Act (AGOA): about, 343–44; background, 344–50; case of Lesotho, 350–58; empirical results, 376–79; model of, 359– 70; testing methods and data for, 370– 76; theory of, 358–59 Allocation, credit, in Burundi, 124–29. See also Capital misallocation ATMs, 65 BANCOBU (Banque Commerciale du Burundi), 134, 134n28, 134n29 Banking scandals, 70–72 Banking sector development, factors explaining, 16. See also Financial development; Financial development gap Banking services: basic, 63–64; expansion of, 64–65; quality issues of, 65. See also Financial development Banks on wheels, 65 Banque de Commerce et de Développement (BCB), 139 Banque Nationale pour le Développement Economique (BNDE), 117–19 Banque Populaire du Burundi (BPB), 139–41 Barclays Bank (United Kingdom), 136 BCB (Banque de Crédit de Bujumbura), 134, 134n28, 134n29, 135 Bhagat, Harko, 333, 337–38

Bidco Oil and Soap, 333, 340 Big Hits: changes in, 304–11; pathways to, 320–40; successes in exports and, 301–4. See also Exports Burundi: allocation of credit to economic sectors of, 124–29; banking profitability in, 133–36; challenges of access to credit in, 141–47; characteristics of nonbank financial institutions in, 120t; commercial banking sector of, 117, 118t; comparison with other EAC countries of gross capital formation in, 112f; credit concentration in, 141; credit from banking sector in, 122–23; credit to government from banking sector in, 124; development banks in, 117–20; effect of poor governance on financial sector of, 109–14; financial liberalization experiment in, 114–17; findings of study for, 104; growth in postindependence, 103–4; highlights of financial liberation in, 151–52; history of political instability in, 106–9; indicators of investment culture in, 153t; microfinance in, 121–22, 122t; political factors and economic performance in, 104–6; reasons for collapse of financial institutions in, 137–41, 137n32; term structure of bank credit in, 129–32; trends of macroeconomic variables in,

403

404

Subject Index

Burundi (continued) 110f; undercapitalization of financial banks in, 136–37 Caisse de Mobilisation et de Financement (CAMOFI), 137–38 Caisse d’Épargne du Burundi (CADEBU), 115, 137 Capital misallocation: access to finance and firm-level evidence, 205–10; construction of measures for, 189–93; countrylevel data for, 192; country-level institutions and, 198–205; data used for study of, 187–88; extent of, 192–98; introduction to, 183–86; investment climate and, 198–205; observations from Ghana, 186–87; questions on obstacles in study, 188–89 Cash loop, defined, 281 Cello Plastics, 333, 340 Cellular phones: coverage of, 215, 216f; in Malawi, 220–24; technologies for, 215. See also Malawi Credit: allocation of, in Burundi, 124–29; challenges of access to, in Burundi, 141–42; concentration, in Burundi, 122–23; experiment design for, in Western Kenya, 75–76; to government from banking sector, in Burundi, 124; products, in Western Kenya, 70; term structure of bank, in Burundi, 129–32 Current account balance, 22 Development banks, in Burundi, 117–20 Diamond Trust Bank (DTB), 135 Economic performance, effects of political economy factors on, 104–5 E-float, 70 E-money loop, M-Pesa and, 281–83, 283f Equity Bank (Kenya), 17; expansion of, 54, 54t; and financial access, 51–59; impact of, 58–59; performance of, 56, 57f. See also Kenya Ethiopia: analysis of leather and hide exports, 315–18; data for shoes, 318, 319f, 319t; top ten exports, 2001 and 2008, 306t Exports: changes in big hits in, 304–11; changes in shares of, and prices, 311– 13; comparative advantage and, 327–

29; decomposition of growth in, 310t; ethnic networks and, 331–32; foreign ownership and, 331; idiosyncratic determinants of success, 334–40; introduction for, 297–301; measurement error concerns and, 313–19; pathways to big hits, 321–27; personal foreign experience and, 332–33; state-of-the-art technology and, 330; successes, 301–4; trade liberalization and, 330 FINABANK of Kenya, 137 Financial development: additional tests for, 39–45; benchmarking, 18–19; benchmarking results for determinants of, 25–26; calibrating gap in, between Africa and other developing countries, 15–16; differences in determinants of African, 31–39; factors explaining, 16–17; firm-level tests and, 45–51; predicted vs. actual African, 26–31; regression model of, 19–25; robustness checks for, 39–45; in sub-Saharan Africa, 13–15. See also Banking services Financial development gap, 25–31; calibrating, 15; in sub-Saharan Africa, 14. See also Banking sector development Financial development indicator variable, 19–21 Financial liberalization experiment, in Burundi, 114–17 Financial market discipline, 140 Financial scandals, in Kenya, 70–72 Fonds de Promotion de l’Habitat Urban (Fund for the Promotion of Urban Housing [FPHU]), 119 Gahaya Links (exporting firm), 328, 333, 335–36 Gatali, Gilbert, 333 Ghana, top ten exports, 1996 and 2008, 305t Good African Coffee (GAC), 326–27, 333, 337 Governance, effect of poor, on financial sector in Burundi, 104, 109–14 Institutional development, as variable, 22, 25 Inter Bank Burundi (IBB), 134, 134n28 Jambo Plastics, 333, 340

Subject Index Kenya: access to banking services in, 57–58, 58t; banking system in, 53–56; Equity Bank and financial access in, 51–59; indicators of investment culture in, 153t; pyramid schemes in, 71; reasons for selection of, for analysis of financial development, 17. See also Equity Bank (Kenya); M-Pesa; Western Kenya Kenya Commercial Bank (KCB), 135 Kirk, Ron, 344 Lake Bounty (fish exporter), 333, 338–40 Lesotho, AGOA and, 343, 350–58 Liberalization experiment, financial, in Burundi, 114–17 Macroeconomic stability, indicators of, 21–22 Malawi: background description of, 219– 20; cell phone access and performance in, 217–19; cellular phone costs and accessing, 222–25; cellular telephones in, 220–24; data sets used for study of, 224–29; licensing process for cellular networks in, 221–22; overview of cellular providers in, 220–21; results for correlates of network performance, 241–43; results for patterns of rollout and correlates of coverage, 229–41. See also Cellular phones Manufacturing, as variable, 22 Marginal product of capital (MPK), 183 Meridien Bank Burundi (MBB), 138 Microfinance institutions (MFIs), 17; in Burundi, 121–22, 122t; in Nigeria, 169–73 Microloans, 64 Misallocation. See Capital misallocation Misozi Coffee Ltd. (exporting firm), 325– 26, 335 M-Kesho, 251n1 Mobile banking. See M-Pesa Mobile money, 64–65, 69–70. See also M–Pesa Mobile phones. See Cellular phones; Malawi Money supply, implications of M-Pesa for measuring, 283–85 M-Pesa, 68, 70; basic structure of, 249–51; characteristics of users of, 264–68, 265t, 266t; cross-tabulations of frequency of use, 267t; distribution of

405

withdrawals and deposits, 256–58, 256t; economic impacts of, microlevel evidence, 268–77; e-money loop and, 281–83; frequency of withdrawals, 257f; impact of, on money transfer companies, 258–64; implications for measuring money supply of, 283–85; as money transfer system, survey results, 252–54; overview of, 247–49; pricing and, 251–52; storing value and, 285–88; uses of, 254–55, 255f; velocity of, 277– 81. See also Kenya Multi-Fiber Arrangement (MFA), 347, 357, 358; expiration of, 379–86 Murzah Oil Mills Ltd., 340 Natural resources, as variable, 20–21 Ndunguste, Joy, 333, 335 Nigeria: alternative financial indicators for, 167, 168f; characteristics of banked/ unbanked in, 169t; data used for study of banking system in, 162; graphical evidence data used for, 162–67; microfinance institutions in, 169–73; need for reforms in, 160–61; recent banking reform efforts in, 158–60; reform and changes in financial indicators for, 173–79; reform outcomes in, 161–62 Nkubana, Janet, 335 Offshore centers, as variable, 21 Per capita income, as variable, 21 Political instability, in Burundi, 104; history of, 106–9 Population, as variable, 20 Prices, changes in export shares and, 311–14 Productivity and Investment Climate Survey, 188 Project to Enhance Agriculture in Rwanda through Linkages (PEARL), 325, 334 Pyramid schemes, in Kenya, 71 Real growth, 21–22 Resource curse, 20 Resources, natural, as variable, 20–21 Rotating saving and credit association (ROSCA), 276 Rugasira, Andrew, 327, 327n18, 333 Rural banking (Kenya): about, 63–68; credit products, 70; financial institutions and,

406

Subject Index

Rural banking (Kenya) (continued ) 68–69; savings products and, 69–70. See also Western Kenya Rwanda, 106; analysis of leather and hide exports, 314, 315–18; coffee exports, 319, 320f, 323–26; exporting handicrafts from, 335–36; indicators of investment culture in, 153t; success of aid in, 334–35 RWASHOSCCO (exporting firm), 323–26, 333, 335 Savings and Credit Cooperatives (SACCOs), 68, 71 Screwdriver plants, 345 Secondary/primary enrollment, as variable, 22–25 Shoes, Ethiopian, 318, 319f, 319t Sub-Saharan Africa: basic banking services in, 63–64; financial development gap in, 14; financial sectors of, 13–14 Sustaining Partnerships to Enhance Rural Enterprise and Agribusiness Development (SPREAD), 325, 334 Tanzania: banking scandals of, 72; indicators of investment culture in, 153t; top ten exports, 1998 and 2007, 308t

Uganda: banking scandals of, 71; cut flower exports, 338; Good African Coffee (GAC), 326–27, 333, 337; indicators of investment culture in, 153t; top ten exports, 1995 and 2008, 307t Union Bank of Nigeria (UBA), 135–36 Velocity, of M-Pesa, 277–81 Western Kenya: credit experiment design for, 75–76; credit products in, 70; data sources study, 76; experimental design for study of, 74–76; factors contributing to low borrowing rates in, 90–94; factors contributing to low rates of formal banking in, 81–90; financial institutions in, 68–69; history of financial scandals in, 70–72; overview of banking options in, 65–68; reasons for low borrowing rates in, 90–94; reasons for low levels of formal banking in, 81– 90; sample size and statistics for study of, 72–74, 73t, 74t; savings experiment design for, 74–75; savings products in, 69–70; snapshots of households and their money in, 76–81. See also Kenya; M-Pesa