Tropical Deforestation and Land Use: Special Issue of Land Economics 77:2 (May 2001) 0299237338, 9780299237332

Table of contents :
The Role of Roads and Protected Areas in North Thailand by Maureen Cropper Jyotsna Puri and Charles Griffiths ..............172
Empirical Evidence from Darien Panama by Gerald C Nelson Virginia Harris and Steven W Stone ..............187
Deforestation and Fuelwood Collection in South Asia by Gunnar Kohlin and Peter J Parks ..............206
Comparing the Impacts of Macroeconomic Shocks Land Tenure and Technological Change by Andrea Cattaneo ..............219
Land Use Impacts of Agricultural Intensification and Fuelwood Taxation in Uganda by Bernards Bashaasha David S Kraybill and Douglas D Southga.................241

Citation preview

Tropical Deforestation and Land Use

Edward B. Barbier and Joanne C. Burgess

G U E S T E D I TO R S

A SPECIAL ISSUE FROM

Land Economics May 2001, Volume 77, Number 2

The University of Wisconsin Press

SPECIAL ISSUE

Tropical Deforestation and Land Use Edward B. Barbier and Joanne C. Burgess Guest Editors

LAND MAY 2001

ECONOMICS VOL. 77, NO. 2

Special Issue: Tropical Deforestation and Land Use ARTICLES

Edward B. Barbier

The Economics of Tropical Deforestation and Land Use: An Introduction to the Special Issue . . . . . . . . . . . . . . . . . . . . . . . . 155

Maureen Cropper, Jyotsna Puri, and Charles Griffith

Predicting the Location of Deforestation: The Role of Roads and Protected Areas in North Thailand . . . . . . . . . . . . . . . . . . . . . . 172

Gerald C. Nelson, Virginia Harris, and Steven W. Stone

Deforestation, Land Use, and Property Rights: Empirical Evidence from Darie´n, Panama . . . . . . . . . . . . . . . . . . . . . . . . . 187

Gunnar Ko¨hlin and Peter J. Parks

Spatial Variability and Disincentives to Harvest: Deforestation and Fuelwood Collection in South Asia . . . . . . . . . . . . . . . . . . . 206

Andrea Cattaneo

Deforestation in the Brazilian Amazon: Comparing the Impacts of Macroeconomic Shocks, Land Tenure, and Technological Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

Bernard Bashaasha, David S. Kraybill, and Douglas D. Southgate

Land Use Impacts of Agricultural Intensificatio and Fuelwood Taxation in Uganda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

Ian Coxhead, Agnes Rola, and Kwansoo Kim

How Do National Markets and Price Policies Affect Land Use at the Forest Margin? Evidence from the Philippines . . . . . . . . . 250

Gerald E. Shively

Agricultural Change, Rural Labor Markets, and Forest Clearing: An Illustrative Case from the Philippines . . . . . . . . . . . . . . . . . 268

Arild Angelsen

Playing Games in the Forest: State-Local Conflict of Land Appropriation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

Gregory S. Amacher, Richard J. Brazee, and Meindert Witvliet

Royalty Systems, Government Revenues, and Forest Condition: An Application from Malaysia . . . . . . . . . . . . . . . . . . . . . . . . . 300

The Economics of Tropical Deforestation and Land Use: An Introduction to the Special Issue Edward B. Barbier ABSTRACT. This paper both introduces the special issue on the economics of tropical deforestation and land use and conducts a synthesis crosscountry analysis of tropical agricultural land expansion. Agricultural development is the main factor determining land expansion, but institutional factors have an important in¯ uence. Income effects vary from region to region, and do not always display an EKC relationship. The case studies comprising this special issue provide further case study insights into tropical deforestation and land use, through spatial analysis of locational factors, CGE modeling of policy scenarios, assessing external market impacts on land clearing, and modelling state interventions and taxation. (JEL Q2, Q23)

I. INTRODUCTION Although global forest loss has occurred for centuries, rapid rates of tropical deforestation have only become an international concern in the last twenty-® ve years or so. As is the usual case with such phenomena, it was not economists but natural scientists who initially called the world’ s attention to the potential consequences of tropical forest destruction, whether it be biodiversity loss, climate change, or the loss of traditional livelihoods of indigenous peoples.1 Economists began studying such problems in the mid1980s, and since then, there has been steady progress in the economic analysis of tropical forest loss. The ``® rst wave’ ’ of studies focused on the economic causes of tropical deforestation and tended to be dominated by statistical analyses across tropical countries, or for selective countries and regions. More recently, a ``second wave’ ’ has focused on modeling and analyzing the economic behavior of Land Economics · May 2001 · 77 (2): 155± 171 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

agricultural households, timber concessionaires, and other agents within tropical countries who affect deforestation through their land use decisions. In some cases, the modeling has centered on a single representative agent of a sector (e.g., an agricultural household or a timber concessionaire). In others, the use of general equilibrium models has facilitated the integration of several key economic sectors (e.g., commercial agriculture, subsistence agriculture, and forestry) to analyze both their individual and collective impacts on deforestation. The following special issue consists mainly of ``second wave’ ’ studies of tropical forest land use and deforestation. The articles were chosen with several criteria in mind. First, they hopefully indicate the breadth of new economic modeling approaches currently being applied in developing countries. Second, these models are applied empirically to a range of economic and institutional policy in¯ uences on tropical land use decisions. As a result, the articles combine innovative conceptual approaches with original emThe author is John S Bugas Distinguished Professor of Natural Resource and Environmental Economics, Department of Economics and Finance, University of Wyoming. He is grateful to Joanne Burgess for assisting him on data preparation for the analyses contained in this paper, for providing useful comments, and for co-editing this special issue with me. Jo Burgess and I would like to thank Daniel Bromley for his enthusiastic support and advice in preparing this issue, Clare Meadley for her assistance with coordinating the manuscript submission and review process, and Carol Olsen for managing the ® nal stages of preparing this issue. We would also like to acknowledge our debt to the many anonymous reviewers who agreed to referee the submitted papers and who provided excellent and timely reviews that proved extremely helpful to both the authors and us. Finally, we would like to thank the authors whose papers appear in this special issue for joining us in this collective effort, and for producing such excellent studies on our tight deadlines. 1 One of the ® rst to popularize the scienti® c concerns over tropical deforestation was Myers (1979).

156

Land Economics

pirical analyses to the problem of tropical deforestation and land use. Third, the articles were also selected to include case studies from the three main tropical regions of the world: Africa, Asia, and Latin America. The countries covered in this special issue are Brazil (Cattaneo), India (KoÈhlin and Parks), Malaysia (Amacher, Brazee, and Witvliet), Panama (Nelson, Harris, and Stone), the Philippines (Coxhead, Rola, and Kim and Shively), Thailand (Cropper, Puri, and Grif® ths) and Uganda (Bashaasha, Kraybill, and Southgate). Last, but certainly not least, it is hoped that the papers in this issue would prove to be inspirational to those interested in pursuing further research in this topic, whether this involves conducting follow-up work to the country studies reported here or pursuing entirely new case studies in other tropical forest countries and regions. The purpose of this introductory article is twofold. First, it has the standard objective of introducing the collection of papers that comprise the special issue. However, an additional aim is to provide a bridge between the two current ``waves’ ’ of research into the economic analysis of tropical forest land use. Improving the statistical analysis of the factors determining tropical forest loss is still an important area of research. More importantly, insights gained from cross-country analysis of tropical deforestation may still inform case study analyses of the economic behavior determining land use decisions within speci® c tropical forest countries and regions. Similarly, new economic modeling approaches and case studies can improve our understanding of the underlying causes and the key processes determining deforestation. The outline of this introduction is as follows. The next section provides a brief summary of global tropical forest land use trends. This is followed by an overview of crosscountry analyses of tropical deforestation, highlighting the main factors and causes identi® ed by such studies. Four key economic approaches to cross-country analysis are then discussed, and from this a synthesis analysis is proposed and applied to a new cross-country data set. The implications and

May 2001

® ndings of the analysis are then used as a basis for introducing the articles that comprise this special issue. II. OVERVIEW OF DEFORESTATION AND LAND USE TRENDS The 1990 Global Forest Resource Assessment (FAO 1993) indicated that the annual deforestation rate across tropical countries over 1981± 1990 was approximately 0.8%, or 15.4 million hectares (ha) per annum (see Table 1). Although the highest rate of deforestation occurs in Asia (1.2%), the area of tropical forests cleared on average each year in Latin America, 7.4 million ha, is almost as much as the forest area cleared in Asia and Africa put together. The largest amount of deforestation is currently occurring in tropical South America (6.4 million ha), followed by Insular South East Asia (1.9 million ha), but the highest rates of deforestation are being experienced in Continental South East Asia (1.6% annually) and Central America and Mexico (1.5% annually).2 Table 2 shows trends in land area and use for different regions of the world since 1980. Over the last ® fteen years in most tropical areas dominated by developing economies, the decline in forest and woodlands is mainly the result of land conversion, in particular agricultural expansion. The loss of permanent pasture may be the result of both the serious degradation problems posed by over-grazing, and also the conversion of pasture land to cropland. Table 3 suggests that the land expansion occurring in tropical regions could be related to structural features of the agricultural sectors of developing economies, such as low irrigation and fertilizer use as well as poor crop yields. Increasing agricultural productivity and input use re¯ ect greater agricultural intensi® cation and development, which in turn mean less pressure is put on conversion of forests and other marginal lands for use in agriculture (Barbier 1997). 2 Although in recent years reforestation through plantation establishment has increased in some tropical countries, this trend has not been suf® cient to offset the loss in overall forest areas.

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

157

TABLE 1 Global Tropical Deforestation Trends, 1980± 1990 Annual Deforestation 1981± 90

Forest Cover

Number of Countries

Land Area (million ha)

1980 (million ha)

1990 (million ha)

Africa West Sahelian Africa East Sahelian Africa West Africa Central Africa Trop. Southern Africa Insulfar Africa

40 6 9 8 6 10 1

2,236.1 528.0 489.7 203.8 398.3 558.1 58.2

568.6 43.7 71.4 61.5 215.5 159.3 17.1

527.6 40.8 65.5 55.6 204.1 145.9 15.8

4.1 0.3 0.6 0.6 1.1 1.3 0.1

0.7 0.7 0.9 1.0 0.5 0.9 0.8

Asia and Paci® c South Asia Continental S.E. Asia Insular S.E. Asia Paci® c

17 6 5 5 1

892.1 412.2 190.2 244.4 45.3

349.6 69.4 88.4 154.7 37.1

310.6 63.9 75.2 135.4 36.0

3.9 0.6 1.3 1.9 0.1

1.2 0.8 1.6 1.3 0.3

33

1,650.1

992.2

918.1

7.4

0.8

7 19 7

239.6 69.0 1,341.6

79.2 48.3 864.6

68.1 47.1 802.9

1.1 0.1 6.2

1.5 0.3 0.7

90

4,778.3

1,910.4

1,756.3

15.4

0.8

Region

Latin America and Caribbean C. America and Mexico Caribbean Trop. South America Total

Million ha

% per annum

Source: FAO (1993).

TABLE 2 Global Trends in Land Area and Use, 1980± 1995 Land Use (million ha)

Africa Asia Europe North and Central America Oceania South America U.S.S.R. (former) World Source: FAO (1997). NA 5 Not available.

Land Area 1995 (million ha)

Population Density 1995 (per `000 ha)

2,964 2,678 473

Cropland

1995

Change since 1980

243 1,284 1,541

193 472 135

2,137 849 1,753 2,195

213 33 181 134

13,048

436

Permanent Pasture

Forest and Woodlands

1994

Change since 1980

1994

Change since 1980

10.3% 3.5% 24.3%

884 792 79

21.1% 14.1% 28.1%

713 537 157

22.0% 22.7% 1.0%

277 53 121 226

1.1% 8.2% 19.8% NA

366 429 495 355

2.2% 25.3% 4.2% NA

824 200 932 810

2.1% 20.3% 0.5% NA

1,476

3.4%

3,399

3.5%

4,172

22.9%

Land Economics

158

May 2001

TABLE 3 Global Trends in Agricultural P roductivity and Input Use, 1979± 1995 Cropland (million ha) 1995 Africa Asia Europe North and Central America Oceania South America U.S.S.R. (former) World

Cereal Yields (kg/ha)

Irrigated Land as a % of Cropland

Fertilizer Use (kg) per ha of Cropland

Cropland per capita 1995

1980

1995

1979± 81

1995

1979± 81

1995

193 472 135

0.27 0.14 0.19

1,124 2,072 3,655

1,128 3,060 4,316

6 29 10

6 35 12

18 67 225

18 144 156

277 53 121 226

0.61 1.87 0.38 0.77

3,260 1,089 1,710 NA

3,918 1,886 2,606 1,301

10 4 7 8

11 5 8 9

91 37 45 80

89 46 54 19

1,476

0.26

2,160

2,752

15

17

81

89

Source: FAO (1997). NA 5 Not available.

III. OVERVIEW OF CROSSCOUNTRY ANALYSES: KEY FACTORS AND PROBLEMS As noted above, an important area of economic research into tropical deforestation consists of cross-country, regional, and selective country-level statistical analyses of the factors determining declining forest cover. Several surveys have illustrated and synthesized the important ® ndings of this growing literature (Brown and Pearce 1994; Kaimowitz and Angelsen 1998; van Kooten, Sedjo, and Bulte 1999). These surveys suggest that the following key factors have an important in¯ uence on tropical deforestation both within and across countries: income; population growth/density; agricultural prices/returns; agricultural yields; agricultural exports/export share; logging prices/returns/production; roads and road building; scale factors (size of forest stock, land area, etc.); and institutional factors (political stability, property rights, rule of law, etc.). However, the literature has also pointed out a number of problems confronting crosscountry analysis of deforestation. First, the United Nations Food and Agricultural Orga-

nization (FAO), which has been the international agency responsible for compiling forest area data across all countries, based its 1990 global forest resource assessment on population growth projections in order to overcome an inadequate forest data base for some countries and regions. This means that the FAO country forest cover data are inappropriate for cross-country analyses of deforestation that use demographic factors as explanatory variables. As the list above indicates, the latter are considered important variables explaining global deforestation. This in turn means that cross-country analyses that employ the FAO forest cover data since 1990 are unreliable.3 In addition, time series data on some of the factors listed above, especially agricultural and logging returns or roads and road investments, are simply unavailable across many tropical countries. Thus these factors are more readily incorporated into deforestation analyses for single countries than across tropical regions or countries. The data sets for many other variables across countries also tend to be incomplete. For example, it may be possible to obtain export unit values 3 The inappropriateness of using the FAO 1990 assessment based country forest cover estimates in crosscountry deforestation analyses has been pointed out by Barbier and Burgess (1997), Cropper and Grif® ths (1994) and Deacon (1999).

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

for logs for a tropical country, but it may be more dif® cult to get an average domestic stumpage value for that country. Agricultural prices for speci® c crops may be available, but a reliable food or cereal price index for a country may not exist. Data on key institutional factors generally exist for a relatively small number of developing countries, and these factors tend to be averages over long periods of time. Finally, some approaches to cross-country analyses have a tendency to be ad hoc. The danger with analyzing deforestation across countries is that too much emphasis can be placed on trying to discover factors that explain trends in changes in forest cover rather than on examining a plausible hypothesis as to why certain economic factors might be correlated with deforestation. However, not all studies suffer from an ad hoc approach to cross-country analysis. In recent years, there have been a number of studies that have attempted to develop a speci® c model or approach to explaining deforestation, and then have tested the resulting hypothesis. The following section identi® es four such approaches in the literature. IV. FOUR APPROACHES TO CROSSCOUNTRY ANALYSES Environmental Kuznets Curve (EKC) Analyses

The environmental Kuznets curve hypothesizes that an environmental ``bad’ ’ ® rst increases, but eventually falls, as the per capita income of a country rises. Although the EKC model has generally been applied to pollution problems, there have been a number of recent studies that have also examined whether this hypothesis also holds for global deforestation (e.g., Antle and Heidebrink 1995; Cropper and Grif® ths 1994; Koop and Tole 1999; Panayotou 1995; Sha® k 1994). The basic EKC model for deforestation is usually F it 2 Fit21 5 F(Yit, Y2it; zit) 5 a1Yit 2 a2Y2it 1 zitb 1 eit ,

[1]

where Fit 2 Fit2 1 is the change in the forest stock over the previous period (which is neg-

159

ative if deforestation is occurring), Yit is per capita income and zit represents a vector of other explanatory variables, such as population density or growth and other macroeconomic variables.4 The application of such an EKC model to explain deforestation trends across countries has produced mixed results. When the model is tested for both temperate and tropical countries, it is inconclusive (Antle and Heidbrink 1995; Panayotou 1995; Sha® k 1994). When applied to just tropical countries, the inverted-U relationship tends not to hold for all countries but may apply to speci® c regions. For example, Cropper and Grif® ths (1994) ® nd some evidence that the EKC model is relevant to Latin America and Africa. However, for each of these regions the turning pointÐ the per capita income level at which the deforestation rate is zero and is about to declineÐ is generally two to four times higher than the average per capita income for that region. In addition, the EKC relationship is likely to vary considerably from country to country. Through employing a random coef® cients panel analysis, Koop and Tole (1999) were unable to reject the hypothesis that countryspeci® c coef® cients are likely to vary compared to cross-country averages (i.e., in Equation [1] a1 5 a1i, a 2 5 a2i for all i 5 1, . . . , N countries in the sample). This in turn implies that each tropical forest country or region is likely to have its own unique EKC relationship, which explains why obtaining a single relationship across all countries may be dif® cult or implausible. Competing Land Use Models

Some empirical analyses have taken as their starting point the hypothesis that forest 4 Strictly speaking, deforestation is de® ned as (minus) the percentage change in forested area, or (Fit21 2 Fit)/Fit21. However, deforestation is clearly related to the change in forest stock variable, Ft 2 Ft21, in equation [1]. In fact, various cross-country analyses have tended to use either speci® cation as the dependent variable to represent forest loss. To simplify notation, Ft 2 Ft21 is used in equation [1] and subsequent equations as a short-hand expression for deforestation.

160

Land Economics

loss in tropical countries is the result of competing land use, in particular between maintaining the natural forest and agriculture (e.g., Barbier and Burgess 1997; Ehui and Hertel 1989). As indicated in Table 2, the evidence across tropical regions is that substantial conversion of forest and woodlands to agriculture is occurring. From an economic standpoint, such conversion implies that potential timber and environmental bene® ts from forest land are irreversibly lost. Therefore, competing land use models usually include some measure of the ``price’ ’ or opportunity cost of agricultural conversion and deforestation in terms of the foregone bene® ts of timber production and environmental bene® ts from forest land Fit 2 Fit2 1 5 AD(vit; zit), ¶AD/¶vit , 0,

[2]

where vit is the opportunity cost or ``price’ ’ of agricultural conversion, AD is the demand for converting forest land to agriculture, and as before zit represents exogenous economic factors (e.g., income per capita, population density). A cross-country analysis of [2] was conducted by Barbier and Burgess (1997) for tropical countries for the ® ve-year change in forested area over 1980± 1985. The results indicated that increased population density increases forest clearance, whereas rising income per capita and agricultural yields reduce the demand for forest conversion. The latter effects suggest that as countries develop economically and the productivity of their existing agricultural lands improve there is less pressure for deforestation. However, there were some problems with the analysis that illustrate the general dif® culty of applying a competing land use model across tropical forest countries. First, the authors had to use a ``proxy’ ’ for vit, as preferred measures of the ``opportunity cost’ ’ of conversion (e.g., land values, timber rents) are not available across countries. Although several proxies were employed, including export unit values for timber, only roundwood production per capita proved to be signi® cant. This variable turned out to be positively related to the ® ve year change in

May 2001

forest area, which the authors concluded was not surprising, given that over the 1980± 1985 period much agricultural conversion occurred in tropical forests that were ® rst ``opened up’ ’ by timber operations. The competing land use model is not necessarily inconsistent with a possible EKC relationship between deforestation and income. To test for the latter hypothesis, Burgess (2000) re-analyzed the original data set used by Barbier and Burgess (1997), but did not ® nd any evidence of a signi® cant EKC relationship for tropical countries as a group over the 1980± 1985 period. Forest Land Conversion Models

Many country-level studies of tropical deforestation have focused on the forest land conversion decision of agricultural households (e.g., Barbier 2000; Barbier and Burgess 1996; Chomitz and Gray 1996; Cropper, Grif® ths, and Mani 1999; LoÂpez 1997; Nelson and Hellerstein 1996; Panayotou and Sungsuwan 1994). Such approaches model the derived demand for converted land by rural smallholders, and assume that the households either use available labor to convert their own land or purchase it from a market. This in turn allows the determinants of the equilibrium level of converted land to be speci® ed. In such models, the aggregate equilibrium level of cleared land across all households is usually hypothesized to be a function of output and input prices and other factors affecting aggregate conversion ADit 5 AD(pit, w Lit, w it; xit, zit), ¶AD ¶AD , 0, . 0. ¶w Lit ¶xit

¶AD . 0, ¶pit [3]

where p is the price of agricultural output, wL is rural wage (labor is a key component in land clearing), w is a vector of other inputs, x are factors in¯ uencing the ``accessibility’ ’ of forest areas (e.g. roads, infrastructure, distance to major towns and cities), and as before zit represents other economic explanatory variables.

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

161

Studies for representative countries in Asia, Africa, and Latin America have tended to con® rm the hypothesis suggested in eq. [3] that agricultural conversion is positively related to agricultural output prices and decreases with rural wage rates (Barbier 2000; Barbier and Burgess 1996; LoÂpez 1997; Panayotou and Sungsuwan 1994). Cropper, Grif® ths, and Mani (1999) also provide evidence from Thailand that the accessibility of forest areas, in this case measure by distance from Bangkok, increases forest land conversion to agriculture. Chomitz and Gray (1996) and Nelson and Hellerstein (1996) include location-speci® c input and output prices to investigate the impact of roads on the agricultural land conversion decision. Although the land conversion model appears to work well for speci® c tropical forest countries, it is dif® cult to obtain time series data on agricultural input and output prices (especially rural wage rates) for many tropical countries. Cross-country data on important ``x’ ’ variables, such as rural road expansion and road building investments, are also hard to ® nd. This means that applying the model to a cross-country panel data set is very problematic. To date, cross-country analyses of agricultural land conversion have tended to leave out prices and ``x’ ’ factors. For example, Southgate (1994) used annual population growth, agricultural export growth, crop yield growth, and a land constraint dummy to explain annual agricultural land growth across Latin America over 1982± 1987. He found that population and agricultural export growth were positively related to land expansion, whereas yield growth and the land constraint were negatively related. Although the results clearly suggest that structural agricultural, economic and geographic factors are signi® cant in explaining land conversion, data constraints meant that the analysis was unable to test the key relationships of the complete forest land conversion model [3].

deforestation of institutional factors, such as land use con¯ ict, security of ownership or property rights, political stability, and the ``rule of law’ ’ (e.g., Alston, Libecap, and Mueller 1999, 2000; Deacon 1994, 1999; Godoy et al. 1998). The main hypothesis tested is that such institutional factors are important factors explaining deforestation

Institutional Models

5 The measure of ownership risk in the study was an index derived by Bohn and Deacon (1997). The index was formed from an estimated investment function that relates investment rates in a cross-country panel to macroeconomic variables and political attributes, including measures of government instability and regime type.

In recent years, a variety of empirical analyses at both the country and cross-country level have explored the impact on tropical

Fi 2 Fit2 1 5 F(q it; zit)

[4]

where q it is a vector of institutional factors and zit is a vector of other economic explanatory variables. Deacon (1999) has applied such an approach to explain the 1980± 1985 change in forest cover as a fraction of land area across all countries, omitting those countries with forests less than 5% of land area or with more than 50% of their forests classi® ed as ``open.’ ’ The main institutional variable, ownership security, proved to be signi® cant and positive in all models, suggesting that greater security reduces forest loss.5 Excluding agricultural yield variables from the models increased the ownership security coef® cient by 25± 30%, which indicates that the latter effect operates partly through agricultural yields. Although such models have demonstrated the importance of institutional factors in determining deforestation, an important question is how much weight should be given to such factors compared to explanatory variables identi® ed by other approaches to crosscountry analyses of forest loss? Nevertheless, by excluding institutional indices, other models applied to cross-country analyses of deforestation may have omitted a potentially important explanatory variable. However, cross-country data sets on institutional factors exist for only a small sub-set of tropical developing countries. In addition, those indices on political stability, corruption, ownership security, and other institu-

162

Land Economics

tional factors that are available for tropical countries tend not to vary much over time, or are constructed as averages over long time periods. Thus the inclusion of institutional indices means the use of a time-invariant variable in a panel analysis explaining forest loss in only a representative sample of tropical countries. V. TOWARDS A SYNTHESIS MODEL Although each of the above four approaches encounter speci® c dif® culties when applied to cross-country analyses of tropical land use conversion, each model also has produced its own unique insight into the possible factors explaining this land use change. An interesting issue is whether it is possible to construct a synthesis model based on the above approaches. The following model discusses brie¯ y one possible synthesis model and its application to the cross-country analysis of tropical land use change. However, in constructing such a synthesis model, the following points need to be kept in mind: First, given the problems with recent (i.e., post-1985) FAO forest stock data highlighted above, the synthesis model should concentrate on explaining agricultural land expansion, Ait 2 Ait2 1, rather than deforestation across tropical countries. This would mean that the model would be able to explain tropical forest loss at least, under the assumption that Fit 2 Fit2 1 5 2(Ait 2 Ait2 1).

[5]

That is, as discussed in the introduction, in most developing countries the major cause of forest loss is presumed to be conversion to agriculture, although the relationship between deforestation and agricultural land expansion may not be as exact as implied by [5] (see Table 2). Second, any synthesis model should be able to test for more than one key factor explaining land use change identi® ed by the above four approaches, provided that the factors chosen are not mutually exclusive. For example, it should be possible to construct a model that can test for the EKC hypothesis

May 2001

as well as examine the possible in¯ uence of institutional factors. On the other hand, given the dif® culty in obtaining cross-country time series data on key variables, such as rural wages, roads, other input prices, it is dif® cult to include variables representing agricultural returns or ``accessibility’ ’ of forest lands in the model. Thus the problem of applying the forest land conversion model across countries, let alone combining this model with an analysis of a possible EKC in¯ uence or institutional factors, still remains. However, as some studies have demonstrated, structural agricultural, economic and geographic factors that vary from country to country are signi® cant in explaining the different land conversion trends across countries (e.g., Barbier and Burgess 1997; Deacon 1999 and Southgate 1994). These factors may be particularly signi® cant explanatory variables in a cross-country analysis, if variables representing agricultural returns or ``accessibility’ ’ of forest lands cannot be included due to data limitations. Thus, the synthesis model should include certain ``structural’ ’ variables (sit), such as agricultural yield, cropland share of land area, agricultural export share, and arable land per capita, to capture country-by-country differences in agricultural sectors and land use patterns, as well as other exogenous explanatory variables, zit. Thus a possible synthesis model might look like: Ait 2 Ait2 1 5 A(Yit, Y2it, sit, zit; q i).

[6]

Finally, as institutional factors (q i) tend to be invariant with time, two versions of the model can be tested, one without and one including q i. Model [6] was applied to a panel analysis of tropical agricultural land expansion over 1961± 1994, with the dependent variable being the percentage annual change in agricultural land area.6 The EKC variables (Yit, Y2it) are represented by gross domestic product (GDP) per capita in constant purchasing 6 Following Barbier and Burgess (1997), topical countries were de® ned as those countries with the majority of their land mass lying between the tropics.

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

power parity (1987 $) and by GDP per capita squared, respectively. The structural variables (sit) are cereal yield, cropland share of total land area, agricultural export share of total merchandise exports, and arable land per capita. The additional explanatory variables (zit) are population and GDP growth. The source of data used for these variables was the World Bank’ s World Development Indicators, which has the most extensive data set for key land, agricultural and economic variables for developing countries over the period of analysis. Table 4 indicates the results without institutional factors, q i. Both one-way and twoway ® xed and random effects models were

163

tested for the sample of all tropical countries, as well as for the regional sub-samples for Africa, Latin America, and Asia. Table 4 displays the results for the preferred models and the relevant statistics. In the table, the parameter estimates are reported as elasticities, in order to facilitate comparison of the effects of the different variables, which are in different units. Across all tropical countries, the structural variables appear to be the more important explanatory factors determining agricultural expansion. Growth in agricultural land area increases with cropland share of total land area and agricultural export share but declines with cereal yield. GDP per capita

TABLE 4 P anel Analysis of Tropical Agricultural Land Expansion, 1961± 1994 Dependent Variable: Agricultural Land Expansion (% annual change)a Elasticity Estimates:b Explanatory Variables GDP per capitac (PPP, constant 1987 $) GDP per capita squared GDP growth (% annual change) Population growth (% annual change) Cereal yield (kg per hectare) Cropland share of land (% of land area) Agricultural export share (% of merchandise exports) Arable land per capita (Hectares per person) Kuznets Curve (Turning point estimate) F-test for pooled model Breusch-Pagan (LM) test Hausman test Adjusted R2 Preferred model

a

All Countries (N 5 656)

Africa (N 5 168)

Latin America (N 5 319)

Asia (N 5 169)

0.270 (0.199) 20.708 (22.035)** 0.012 (0.177) 20.603 (21.027) 22.248 (22.846)*** 7.112 (3.550)*** 4.351 (2.079)** 0.355 (0.163) No ($858) 3.077*** 51.69*** 31.97***

20.980 (21.315) 0.271 (1.228) 0.023 (0.456) 0.259 (0.278) 20.250 (20.479) 0.923 (4.249)*** 0.078 (0.496) 20.695 (21.659)* No ($3,706) 1.398 5.90** 2.07

20.388 (20.109) 0.050 (0.036) 0.009 (0.127) 20.734 (21.234) 22.632 (22.028)** 5.565 (2.485)*** 0.462 (2.098)** 20.263 (20.163) No ($17,359) 3.323*** 11.23*** 15.52**

3.068 (2.667)*** 21.391 (23.370)*** 20.132 (20.502) 20.538 (20.445) 21.678 (21.644)* 20.452 (20.694) 0.395 (0.967) 0.406 (0.665) Yes ($6,182) 2.245*** 0.29 10.88

0.212 One way ® xed effects

0.183 One way random effects

.215 One way ® xed effects

0.176 One way random effects

Mean for all countries is 0.64%, for Africa 0.26%, for Latin America 0.75%, and for Asia 0.80%. t-ratios are indicated in parentheses. c Mean for all countries is $2,863, for Africa $1,230, for Latin America $3,654, and for Asia $3,029. PPP is purchase power parity. *** Signi® cant at 1% level; ** signi® cant at 5% level; * signi® cant at 10% level. b

Land Economics

164

squared also has a negative impact on agricultural expansion, although this effect is smaller than for the other signi® cant variables. These results are pretty much replicated for the Latin America sub-sample, with the exception that neither per capita income variable is signi® cant. In Africa, agricultural expansion is explained by cropland share of total land area only, although it is possibly negatively affected by the amount of arable land per capita. The regression for Asia is the only estimation that cannot reject the EKC

May 2001

hypothesis. However, the level of per capita income at which agricultural expansion peaks in Asia is estimated to be $6,182, which is approximately double the sample mean. Increases in cereal yields may also possibly slow agricultural land expansion in Asia. Table 5 repeats the same regressions, but with the inclusion of three institutional variables: a corruption index, a property rights index, and a political stability index. These indices were obtained from the Levine-

TABLE 5 P anel Analysis of Tropical Agricultural Land Expansion, 1961± 1994, Including Institutional Factors Dependent Variable: Agricultural Land Expansion (% annual change)a Elasticity estimates:b Explanatory Variables GDP per capitac (PPP, constant 1987 $) GDP per capita squared GDP growth (% annual change) Population growth (% annual change) Cereal yield (kg per hectare) Cropland share of land (% of land area) Agricultural export share (% of merchandise exports) Arable land per capita (hectares per person) Corruption index (high 5 0, low 5 10) Property rights index (high 5 5, low 5 1) Political stability index (high 5 0, low 5 1) Kuznets Curve (Turning point estimate) Durbin-Watson statisticd Adjusted R2 Preferred model a

All Countries (N 5 383)

Africa (N 5 48)

Latin America (N 5 233)

Asia (N 5 102)

4.125 (3.443)*** 21.519 (22.461)*** 20.066 (20.778) 1.152 (2.153)** 20.309 (20.577) 0.433 (2.064)** 0.654 (5.658)*** 20.043 (20.174) 0.840 (1.703)* 20.395 (20.590) 0.613 (1.896)*** Yes ($5,445) 2.009

225.214 (21.388) 13.984 (1.651)* 20.088 (20.463) 2.945 (0.803) 20.270 (20.180) 5.364 (0.648) 21.168 (21.217) 20.970 (20.334) 2.779 (0.105) 25.266 (20.214) 25.266 (0.150) No ($1,211) 2.372

7.943 (2.595)*** 23.499 (22.328)** 20.066 (20.862) 1.421 (2.271)** 20.623 (20.929) 0.789 (2.343)** 0.383 (2.948)*** 0.358 (0.83) 0.249 (0.345) 20.363 (20.425) 0.494 (2.220)** Yes ($4,946) 2.045

0.749 (0.200) 20.699 (20.528) 0.010 (0.032) 4.635 (2.510)*** 22.267 (20.783) 3.541 (1.474) 20.282 (20.421) 22.842 (21.087) 4.447 (2.251)** 9.440 (1.046) 21.429 (20.948) No ($1,815) 1.998

0.191 OLS with ACd

0.073 OLS with ACd

0.128 OLS

0.450 OLS with ACd

Mean for all countries is 0.61%, for Africa 0.14%, for Latin America 0.66%, and for Asia 0.72%. t-ratios are indicated in parentheses. Mean for all countries is $2,986, for Africa $1,211, for Latin America $3,675, and for Asia $2,246. PPP is purchase power parity. d After autocorrelation correction (AC) by Cochrane-Orcutt procedure, if required. *** Signi® cant at 1% level; ** signi® cant at 5% level; * signi® cant at 10% level. b c

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

Loayza-Beck data set used in Beck, Levine, and Loayza (1999) and Levine, Loayza, and Beck (1999), which are available from the Economic Growth Research Group of the World Bank. The corruption and property rights indices are directly from the LevineLoayza-Beck data set and are averaged over 1982± 1995. The political stability index was created as a composite index of the average number of revolutions and coups (averaged over 1960± 1990), average number of assassinations per million population (averaged over 1960± 1990), and an index of ethnic fractionalization (averaged over 1982± 1995). As these indices were not available for all tropical countries in the original sample, the inclusion of these institutional factors reduced the sample sizes of the regressions considerably. In addition, the three indices are time invariant, and with their inclusion in addition to the original explanatory variables of the model, ® xed effects regressions cannot be run.7 Table 5 indicates that the inclusion of institutional factors has a considerable in¯ uence on the analysis.8 For all tropical countries, the EKC hypothesis can no longer be rejected, although the estimated EKC turning point is nearly double the mean per capita income for the sample. Population growth, the ratio of cropland to total land area, the share of agricultural exports, and political instability all appear to have a signi® cant and positive impact on agricultural expansion across all tropical countries. The regression for Latin American countries yields similar results, which is not surprising as this region dominates the sample of all countries. However, the estimated EKC turning point for Latin America is only one third larger than the average per capita income of $3,675 for the sub-sample. This suggests, that if agricultural land growth does start to slow down as GDP per capita increases, we are likely to observe this phenomenon occurring in tropical Latin America ® rst. For Asia, population growth and lower corruption appear to have a signi® cant and positive in¯ uence on agricultural expansion. The latter effect may seem counter-intuitive, although LoÂpez (1998) has argued that reduced corruption and improved bureaucratic ef® ciency may

165

actually facilitate the implementation of land and credit policies that stimulate a ``race for property rights’ ’ to convert forest and other common resource land to agriculture. Finally, given the small number of observations for the African sub-sample, the separate regression for the African region has poor explanatory power and should be ignored. The above results for the synthesis model provide interesting additional results to existing cross-country analyses of tropical land use and deforestation. First, the pattern of agricultural development, as represented by such structural variables as cropland share of total land area, agricultural export share of total exports, and to some extent, cereal yields, appears consistently to in¯ uence tropical agricultural land expansion. Population growth could be an additional factor, especially in Asia. Corruption and political stability may also be important institutional in¯ uences, but their signi® cance may vary from region to region. The existence of an EKC effect for agricultural expansion appears to be highly sensitive to the model speci® cation, and the impact of changes in GDP per capita on agricultural expansion is likely to differ considerably across tropical regions. VI. THE CONTRIBUTION OF THIS SPECIAL ISSUE By focusing on speci® c case studies from tropical regions, the articles comprising this special issue explore in more depth the economic determinants of tropical forest land conversion. In doing this, several new contributions are made. 7 Including the three time-invariant institutional indices in a ® xed effects regression leads to collinear regressors (Baltagi 1995). As the institutional indices are in themselves `weighted’ country-speci® c dummy variables, including the indices in an OLS regression will essentially imitate a ® xed effects model. Of course, the estimated coef® cients on the institutional variables may also be including the in¯ uence of other slow-changing factors that vary across countries. 8 The failure of the property rights index to be signi® cant in any of the regressions reported in Table 5 may re¯ ect the fact that this variable indicates the degree of protection of private property rights across all sectors of the economy rather than the security of land tenure in the agricultural sector.

166

Land Economics

Studies on the economics of temperate and tropical forest land are beginning to take into account the location of forests (relative to roads) and spatial features (e.g., size, topography, soil conditions) as factors in¯ uencing land use decisions (Parks, Barbier, and Burgess 1998). As noted above, many forest land conversion models incorporate the ``accessibility’ ’ of forest land as an explicit determinant of the equilibrium level of land converted to agriculture. Previous studies of tropical deforestation have also included location-speci® c input and output prices to explain the in¯ uence of road-building on forest conversion. Several studies in this special issue further illustrate the importance of incorporating spatial and locational variability in the analysis of tropical deforestation and land use. For example, Cropper, Puri, and Grif® ths use a bivariate probit model to explain land clearing and the siting of protected areas in the 17 provinces of Northern Thailand in 1986. The results suggest that steeper slopes, higher elevations, and locations further from market centers increase the chance that forest land is designated a protected area or a wildlife sanctuary. However, whereas there is weak evidence to suggest that the creation of wildlife sanctuaries may have deterred deforestation in Northern Thailand, protected areas overall (i.e., wildlife sanctuaries plus national parks) did not reduce the probability of land clearing. Finally, the authors also used their model to demonstrate where further road building in Northern Thailand may have the greatest impact on the probability that forests are cleared, and to identify the impact of additional road building on protected areas through simulation (see Figures 5 and 6 in Cropper, Puri, and Grif® ths). Nelson, Harris, and Stone combine spatial and institutional analysis to investigate the hypothesis that providing land users with secure property rights will result in more sustainable land use and less deforestation. A multinomial logit model incorporating both socioeconomic and geophysical variables is used to evaluate land use patterns in three locations in DarieÂn Prvovince, PanamaÐ a national park where no human activity is supposed to occur plus two long-standing re-

May 2001

serves for indigenous peoples, Cemaco and Sambu Reserves. The authors employ their model to simulate the effects on land use of removing legal protection of the park and the two reserves. The results suggest that legal protection of the national park has made little difference to land use within the park, as its dif® cult terrain and remoteness limit any pro® table activity in this area. In contrast, as construction of the Pan American highway in the early 1980s increased accessibility to the Cemaco Reserve, legal property rights in the reserve clearly serve to limit encroachment from outsiders. The need for effective property rights is less important for Sambu Reserve, as this area is far from the nearest primary road. KoÈhlin and Parks develop a household spatial model to examine how the options of collecting fuelwood from the natural forest, from village woodlots, or from both locations impact on deforestation in Orissa, India. Using survey data from 742 village households, the discrete decision of whether to collect from the natural forest is ® rst estimated using a probit model, and then the decision to spend time collecting is analyzed using a sample selection procedure. The main ® nding is that village woodlots can reduce the pressure on natural forests for fuelwood collection, but woodlots closer to the forest appear to be less used than those located further away. However, plantations very far from the natural forest also do not reduce the use of the forest for fuel biomass. The result is an inverted-U relationship between the distance of the woodlot from the natural forest and the probability that a household will reduce its fuelwood collection from the forest (see Figure 1 in KoÈhlin and Parks). This implies that the location of a village woodlot will have a considerable impact on whether it is effective in reducing fuelwood collection by households from the natural forest. Another innovative approach that can be used to examine the effects of economy-wide and sectoral changes on tropical land use and deforestation is computable general equilibrium (CGE) modeling. Two papers in this special issue, Cattaneo and Bashassha, Kraybill, and Southgate, employ this approach. Cattaneo develops a CGE model to assess

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

the impacts on deforestation in the Brazilian Amazon of changes in the real exchange rate, agricultural taxes and price supports, road construction, land tenure, and technological change in agriculture. The differences in the short- and long-run deforestation rates predicted by the model illustrate how interregional ¯ ows of labor and capital are critical to the process of agricultural conversion in the Amazonian frontier forest. If recent trends in road building continue to lower transport costs in the region, deforestation is likely to accelerate. On the other hand, deforestation could be signi® cantly reduced if the Brazilian government adopts a balanced contraction policy in response to any real exchange rate devaluation. However, because signi® cant agricultural reforms are already in place in Brazil, the scope for abolishing subsidies or imposing a revenue-neutral tax in agriculture to reduce deforestation is limited. Within the Amazon, improving land tenure security would reduce deforestation signi® cantly, although implementing this policy in the frontier region would be extremely dif® cult. Finally, the impacts of technological change are mixed: in the livestock sector, deforestation would be increased, although the income of producers would also rise; in the annual crops sector, deforestation would also increase, but producer income would not; and in the perennial crops sector, deforestation would be reduced while producer income would rise. Bashaasha, Kraybill, and Southgate construct a CGE model to explore the market and land use impacts of agricultural intensi® cation and fuelwood taxation in Uganda. Their results suggest that technologically neutral improvements in agricultural productivity across all sectors of the rural economy increases cash crop exports, raises domestic food consumption, and reduces forest loss. In contrast, targeting a single agricultural sector with high demand elasticity, such as export crops, will cause an expansion of cropped area at the expense of natural forest. Fuelwood taxation proves to be an ineffective instrument for reducing deforestation, and would be an extremely unpopular policy if implemented in Uganda. The authors conclude that improvements in agricultural pro-

167

ductivity and overall development across Uganda are more effective approaches to mitigating deforestation than either targeting improvements to speci® c agricultural sectors or implementing a fuelwood tax. The next two papers in this issue are concerned with exploring upland land use decisions and forest clearing in the Philippines. In both cases, the authors focus on the economic linkages between upland activities and markets as determinants of the land clearing decisions of upland households. Coxhead, Rola, and Kim examine the hypothesis that national markets and policies play a signi® cant role in determining the land expansion and allocation decisions of upland farmers in Lantapan Watershed, the Philippines. Previous studies in this region have con® rmed that upland farmers do respond to relative prices and their variability in making their land use decisions. By employing vector auto-regression techniques, Coxhead, Rola, and Kim are able to demonstrate that there is also a strong degree of interconnectedness between corn, potato, and cabbage prices in the local Lantapan market and prices in regional and national markets. The results indicate that in the short run the markets for these crops are integrated, and that upland farmers are essentially price takers with respect to regional and national markets. This suggests that economy-wide and trade policies will affect land use decisions in the Philippine uplands, and in particular, trade liberalization is likely to reduce farm-gate prices of corn and vegetables, the two most environmentally damaging crops grown in Lantapan and similar watersheds in the Philippines. Shively investigates the impact of rural employment opportunities in lowland areas of the Philippines on the labor and land allocation decisions of upland farmers. In particular, the focus is on how changes in technology in lowland agriculture, including irrigation development, affect land-clearing decisions in the uplands. To do this, the author develops a model linking lowland and upland household activity, which was then applied to two upland and two lowland communities in Palawan. The results indicate that irrigation development in the lowlands does

168

Land Economics

lead to increased hiring of upland workers, and in response to such off-farm employment opportunities upland farmers will reduce their rates of forest clearing. Lowland agricultural intensi® cation, particularly in areas adjacent to uplands, can therefore reduce pressure on forests in marginal upland farming areas. Policies that encourage the use of rural labor in these lowland agricultural areas, either explicitly or through improvements in cropping intensity, could reduce upland deforestation signi® cantly. As noted above, institutional factors have been increasingly viewed as important determinants of deforestation. Of particular concern is the lack of property rights in frontier forested areas, which leads to a de facto open access situation and excessive clearing. As Angelsen suggests, under these conditions, possible land use and deforestation outcomes on the frontier can be modeled as a twoplayer game in which a given forest area can end up being allocated in three different ways: it can be converted to agricultural land by the local community; it can be converted to plantations, logging or other large scale projects by the state; or it can remain as natural forest. When the scarcity of and competition for forest land are very strong, and either the local community or the government acts as leader in a Stackelberg game to ``squeeze out’ ’ the other agent, then the result is increased deforestation. Evidence from Sumatra, Indonesia, and Brazil cited by the author indicates situations in which local communities ``race’ ’ to establish their claims by clearing forested land. Further evidence from Indonesia demonstrates how state-supported logging, transmigration, and plantation projects have in the past both supplanted local communities and led to excessive forest conversion. Finally, Angelsen (1999) also con® rms that state ``planned’ ’ deforestation can also promote further deforestation by local communities, particularly if state activities lead to increased infrastructure and roadbuilding in forested areas and thus lower the costs of local agricultural expansion. One of the key state-sponsored investments in tropical forests is logging activities, as governments are able to earn signi® cant revenues from harvesting royalties. How-

May 2001

ever, it has been argued that most royalty systems in tropical countries may actually increase forest degradation and conversion through incentives for poor logging practices. In the ® nal paper of this special issue, Amacher, Brazee, and Witvliet investigate this important linkage by examining how selective harvesting of high-valued species in timber concessions in Sabah, Malaysia, are affected by the royalty structure. The authors ® nd that, as harvesting of the high quality species is price elastic, forest degradation can be signi® cantly reduced per additional dollar of government revenue earned if royalty reform includes increasing the differentiation in tax rates for high- versus low-valued species. However, to reduce degradation and increase revenues, the government would clearly prefer a differentiated lump sum royalty, whereas to increase its share of rents, a concessionaire would prefer a differentiated ad valorem royalty. Nevertheless, royalty reform could promote more sustainable forestry in Malaysia and similar countries, especially if the gains in revenues through differentiated royalties were re-invested in reforestation programs or intensive management of secondary forests. VII. CONCLUSION The economics of tropical deforestation and land use has consisted of two distinct ``waves’ ’ of analysis. This introduction to the special issue has conducted a brief overview of the cross-country analyses of the causes of tropical deforestation that typify the ``® rst wave,’ ’ and has attempted a synthesis analysis of the factors in¯ uencing agricultural land expansion across countries. The results suggest that the pattern of agricultural development across these countries appears to affect the growth in agricultural land area, which tends to be the predominant cause of forest loss in tropical regions. Population growth may also matter, especially in Asia. Institutional factors are also important in¯ uences, although their inclusion in crosscountry analyses is still constrained by the lack of data for some countries as well as the limited appropriateness of those institutional indices that are available (e.g., the property

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

rights index used in the synthesis analysis of this introduction). Finally, the impact of changes in per capita GDP on agricultural expansion varies considerably from region to region, and does not always exhibit an EKCtype relationship. As the articles in this special issue demonstrate, ``second wave’ ’ country case studies are able to investigate in much more detail other key factors that in¯ uence the economics of tropical deforestation and land use. For example, the papers by Cropper, Puri and Grif® ths, Nelson, Harris, and Stone, and KoÈhlin and Parks illustrate through spatial analysis the importance of location, the ``accessibility’ ’ of forests and other geographical factors in explaining forest land use patterns, whether it be land clearing and the siting of protected areas in Thailand, the impact of property rights on land use in Panama, or the impact of the location of woodlots on fuelwood collection from natural forests by rural households in India. Similarly, Cattaneo and Bashaasha, Kraybill, and Southgate show how CGE modeling in Brazil and Uganda, respectively, can be an effective tool for investigating a variety of economy-wide and sectoral policy impacts on agricultural expansion and deforestation. Coxhead, Rola, and Kim and Shively both focus on how external market forces may affect the land-clearing decisions of upland farmers in the Philippines. Coxhead, Rola, and Kim demonstate that farmers are highly responsive to changes in regional and national crop prices, whereas Shively indicates how technological change and increased employment opportunities in lowland agricultural areas in¯ uence the labor and land decisions of upland farmers. The role of the state in in¯ uencing both local patterns of deforestation in remote frontier forested areas and forest degradation by major state-supported activities, such as the timber industry, also needs more careful analysis. Such analysis is the focus of the ® nal two articles in this special issue, by Angelsen on game-theoretic, land use decisions on the open-access frontier and by Amacher, Brazee, and Witvliet on harvesting royalties in the Malaysia. Both papers demonstrate that the state is not ``passive’ ’ with

169

regard to frontier forest loss and degradation, but can have considerable impact on deforestation through the way in which it ``sponsors’ ’ plantations, logging, and other largescale commercial activities on the frontier, or through the structure of taxes that it imposes on such enterprises, such as the royalties paid by timber concessionaires. In sum, the articles in this special issue illustrate how innovative economic models can be used effectively to investigate a range of important in¯ uences on tropical land use changes in a variety of representative developing countries. It is hoped that this issue will inspire others to pursue novel and insightful studies into the economics of tropical deforestation and land use. References Alston, Lee J., Gary D. Libecap, and Bernardo Mueller. 1999. Titles, Con¯ icts, and Land Use: The Development of Property Rights and Land Reform in the Brazilian Amazon Frontier. Ann Arbor: University of Michigan Press. Ð Ð Ð . 2000. ``Land Reform Policies, the Sources of Violent Con¯ ict, and Implications for Deforestation in the Brazilian Amazon.’ ’ Journal of Environmental Economics and Management 39 (2): 162± 88. Angelsen, Arild. 1999. ``Agricultural Expansion and Deforestation: Modelling: The Impact of Population, Market Forces, and Property Rights.’ ’ Journal of Development Economics 58 (Apr.): 185± 218. Antle, John M., and George Heidebrink. 1995. ``Environment and Development: Theory and International Evidence.’ ’ Economic Development and Cultural Change 43 (3): 603± 25. Baltagi, Badi H. 1995. Econometric Analysis of Panel Data. Chichester, U.K.: John Wiley. Barbier, Edward B. 1997. ``The Economic Determinants of Land Degradation in Developing Countries.’ ’ Philosophical Transactions of the Royal Society, Series B 352 (1356): 891± 99. Ð Ð Ð . 2000. ``Institutional Constraints and Deforestation.’ ’ Paper presented at the 2000 Royal Economic Society/Scottish Economic Society Conference, St Andrews, Scotland, July 10± 13, 2000. Barbier, Edward B., and Joanne C. Burgess. 1996. ``Economic Analysis of Deforestation in Mexico.’ ’ Environment and Development Economics 1 (2): 203± 40.

Ð

170 Ð

Land Economics

Ð . 1997. ``The Economics of Tropical Forest Land Use Options.’ ’ Land Economics 73 (May): 174± 95. Beck, Thorsten, Ross Levine, and Norman Loayza. 2000. ``Finance and the Sources of Economic Growth.’ ’ Journal of Financial Economics 58 (1± 2): 261± 300. Bohn, Henning, and Robert T. Deacon. 1997. ``Ownership Risk, Investment, and the Use of Natural Resources.’ ’ Department of Economics, University of California, Santa Barbara. Working Paper. Brown, Katerina, and David W. Pearce, eds. 1994. The Causes of Tropical Deforestation: The Economic and Statistical Analysis of Factors Giving Rise to the Loss of the Tropical Forests. London: University College London Press. Burgess, Joanne C. 2000. ``The Economics of Tropical Forest Land Use.’ ’ Ph.D. diss., Economics Department, University College London. Chomitz, Kenneth M., and David P. Gray. 1996. ``Roads, Land Markets, and Deforestation: A Spatial Model of Land Use in Belize.’ ’ The World Bank Economic Review 10 (3): 487± 512. Cropper, Maureen, and Charles Grif® ths. 1994. ``The Interaction of Population Growth and Environmental Quality.’ ’ American Economic Review, AEA Papers and Proceedings, 84 (2): 250± 54. Cropper, Maureen, Charles Grif® ths, and Muthukumara Mani. 1999. ``Roads, Population Pressures, and Deforestation in Thailand, 1976± 1989.’ ’ Land Economics 75 (Feb.): 58± 73. Deacon, Robert T. 1994. ``Deforestation and the Rule of Law in a Cross-Section of Countries.’ ’ Land Economics 70 (Nov.): 414± 30. Ð Ð Ð . 1999. ``Deforestation and Ownership: Evidence from Historical Accounts and Contemporary Data.’ ’ Land Economics 75 (Aug.): 341± 59. Ehui, Simeon K., and Thomas W. Hertel. 1989. ``Deforestation and Agricultural Productivity in the CoÃte d’ Ivoire.’ ’ American Journal of Agricultural Economics 71 (Aug.): 703± 11. Food and Agricultural Organization (FAO). 1993. Forest Resources Assessment 1990: Tropical Countries. Rome: FAO. Ð Ð Ð . 1997. State of the World’ s Forests 1997. Rome: FAO. Godoy, Ricardo, Marc Jacobson, Joel De Castro, Vianca Aliago, Julio Romero, and Allison Davis. 1998. ``The Role of Tenure Security and Private Time Preference in Neotropical

May 2001

Deforestation.’ ’ Land Economics 74 (May): 162± 70. Kaimowitz, David, and Arild Angelsen. 1998. Economic Models of Tropical Deforestation: A Review. Bogor, Indonesia: Center for International Forestry Research. Koop, Gary, and Lise Toole. 1999. ``Is There an Environmental Kuznets Curve for Deforestation? ’ ’ Journal of Development Economics 58 (July): 231± 44. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. ``Financial Intermediation and Growth: Causality and Causes.’ ’ Journal of Monetary Economics 46 (1): 31± 77. LoÂpez, RamoÂn. 1997. ``Environmental Externalities in Traditional Agriculture and the Impact of Trade Liberalization: The Case of Ghana.’ ’ Journal of Development Economics 53 (July): 17± 39. Ð Ð Ð . 1998. ``Where Development Can or Cannot Go: The Role of Poverty-Environment Linkages.’ ’ In Annual Bank Conference on Development Economics 1997, ed. B. Pleskovic and J. E. Stiglitz, 285-306. Washington D.C.: The World Bank. Myers, Norman. 1979. The Sinking Ark: A New Look at the Problem of Disappearing Species. Oxford: Pergamon Press. Nelson, Gerald C., and Daniel Hellerstein. 1997. ``Do Roads Cause Deforestation? Using Satellite Images in Econometric Analysis of Land Use.’ ’ American Journal of Agricultural Economics 79 (2): 80± 88. Panayotou, Theodore. 1995. ``Environmental Degradation at Different Stages of Economic Development.’ ’ In Beyond Rio: The Environmental Crisis and Sustainable Livelihoods in the Third World, ed. I. Ahmed and J. A. Doeleman. London: MacMillan Press. Panayotou, Theodore, and Somthawin Sungsuwan. 1994. ``An Econometric Analysis of the Causes of Tropical Deforestation: The Case of Northeast Thailand.’ ’ In The Causes of Tropical Deforestation: The Economic and Statistical Analysis of Factors Giving Rise to the Loss of the Tropical Forests, ed. K. Brown and D. W. Pearce, 192± 210. London: University College London Press. Parks, Peter J., Edward B. Barbier, and Joanne C. Burgess. 1998. ``The Economics of Forest Land Use in Temperate and Tropical Areas.’ ’ Environmental and Resource Economics 11 (3± 4): 473± 87. Sha® k, Nemat. 1994. ``Economic Development and Environmental Quality: An Econometric Analysis.’ ’ Oxford Economic Papers 46 (Oct.): 757± 73.

77(2)

Barbier: The Economics of Tropical Deforestation and Land Use

Southgate, Douglas. 1994. ``Tropical Deforestation and Agricultural Development in Latin America.’ ’ In The Causes of Tropical Deforestation: The Economic and Statistical Analysis of Factors Giving Rise to the Loss of the Tropical Forests, ed. K. Brown and D. W. Pearce, 134± 45. London: University College London Press.

171

van Kooten, G. Cornelius, Roger A. Sedjo, and Erwin H. Bulte. 1999. ``Tropical Deforestation: Issues and Policies.’ ’ In The International Yearbook of Environmental and Resource Economics 1999/ 2000, ed. H. Folmer and T. Tietenberg, 199± 248. London: Edward Elgar.

Predicting the Location of Deforestation: The Role of Roads and Protected Areas in North Thailand Maureen Cropper, Jyotsna Puri, and Charles Grif® ths ABSTRACT. Using plot level data, we estimate a bivariate probit model to explain land clearing and the siting of protected areas in North Thailand in 1986. The model suggests that protected areas (national parks and wildlife sanctuaries together) did not reduce the likelihood of forest clearing; however, wildlife sanctuaries may have reduced the probability of deforestation. Road building, by reducing impedance-weighted distance to market, has promoted clearing, especially near the forest fringe. We simulate the impact of further road building to show where road building is likely to have greatest impact and where it is likely to threaten protected areas. ( JEL Q23, Q28, R40)

I. INTRODUCTION Concern over the rate at which forests are being converted to agriculture has given rise to a literature that quanti® es the impact of forces that drive deforestation. The literature has focused on two questions: (1) What factors affect the location of deforestation? and (2) What factors affect the rate of deforestation? Each question has policy signi® cance. It is clearly important to know where deforestation is likely to occur, especially if it is in environmentally sensitive areas, and it is also important to know how fast the process is taking place. This paper focuses on the ® rst question. We estimate an equilibrium model of land use in North Thailand in the mid-1980s, using coarse-resolution (1: 1,000,000) plotlevel data. The purpose of the model is to predict where deforestation is likely to occur and to examine the impact of two government policies that can affect the location of deforestation: the establishment of protected areas, and road building. Protected areas are often suggested as a Land Economics · May 2001 · 77 (2): 172± 186 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

means of conserving tropical ecosystems and have, at least on paper, been created in many tropical developing countries. In 1985, Thailand declared that 15% of its land area should be set aside for conservation or protected forests. By 1986, 10% of the country’ s land lay within protected areas. Fifty-two percent of the land in protected areas was devoted to national parks and 42% to wildlife sanctuaries.1 Whether such areas can, in fact, protect biodiversity depends on their size and location, and on how they are managed. Protected areas are less likely to experience encroachment if they have the political support of surrounding communities, and if these communities can produce suf® cient income without encroaching upon the protected area. This suggests that understanding the reasons for the success or failure of protected areas requires on-the-ground knowledge, and is best evaluated using a case study approach. The contribution we make to the topic is to evaluate statistically whether protected areas have reduced the probability of deforestation in national parks and wildlife sanctuaries in Thailand. Other authors who have tackled this issue (Chomitz and Gray 1996; DeiMaureen Cropper is lead economist, The World Bank, professor of economics, University of Maryland, College Park, and university fellow, Resources for the Future. Jyotsna Puri is a consultant at The World Bank and a Ph.D. student at the University of Maryland. Charles Grif® ths is an economist at the Environmental Protection Agency. The authors would like to thank Ken Chomitz, Lou Scura, two anonymous referees, Tom Tomich, seminar participants at the Kennedy School of Government, and participants in a joint seminar hosted by CIFOR and ICRAF, Indonesia, for comments and helpful discussions. The authors also thank The World Bank for providing ® nancial support. The ® ndings, interpretations and conclusions are the authors’ own and should not be attributed to the EPA or the World Bank, its Executive Board of Directors, or any of its member countries. 1 The remaining 8% of protected areas included arboretums, botanical gardens, and reserved areas.

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

ninger and Minten 1996) have estimated a land use model that predicts the probability that land located in protected areas is cleared. The fraction of land predicted to be cleared is then compared with the fraction of land actually cleared to determine the impact of protected areas on clearing. This approach does not, however, allow one to determine whether the impact of protected areas on clearing is statistically signi® cant, or to test hypotheses about its magnitude. We estimate a bivariate probit model to explain the probability that a plot of land is cleared and the probability that it lies within a protected area. Protected area status enters the clearing equation, and variables that affect the designation of an area as protected (but not the clearing decision) are used to identify the coef® cient of protected area status. This allows us to control for the selectivity problem inherent in single-equation models of land use: In a single-equation model of clearing, the coef® cient of protected area status is likely to overstate (in absolute value) the impact of protected areas on clearing. This is because protected areas are likely to be located in places that have not yet been cleared. The second topic on which we focus is the impact of roads on the land-clearing decision. Qualitatively, the impact of roads on land clearing is well understood: Road building facilitates access to markets, and thus raises the probability that forests will be cleared for agriculture. Understanding the quantitative impact of road building on clearing is, however, crucial for policy. Suppose a government wishes to build a road to a proposed national park. Where should the road be located to reduce the likelihood of development en route to the park? As Chomitz and Gray (1996) emphasize in their study of the impact of roads on agricultural development in Belize, the impact of roads depends on the topography of the area, and on soil quality. One goal of our study is to show where road building in North Thailand is likely to have the greatest impact on the probability that forests are cleared, and to identify the impact of further road building on protected areas. To investigate the issues discussed above, we have assembled a GIS database on land

173

use, roads, physiographic variables (slope, elevation, and soil quality), populated places, and population density for the 17 provinces of North Thailand. The data also include protected area boundaries, and provincial and district boundaries. The model of land clearing and protected area status estimated with these data is described in section 2. Section 3 contains a more detailed description of the data and our sampling strategy. Econometric results are presented in section 4. We conclude the paper by showing how our model can be used to estimate the threat of encroachment in protected areas. II. A MODEL OF LAND CLEARING AND PROTECTED AREA STATUS Economic theory predicts that forested land will be cleared if the pro® ts from clearing land exceed the pro® ts from leaving land under forest cover. We follow Chomitz and Gray (1996) (see also Nelson and Hellerstein 1997) in assuming that the pro® t from land use k on plot i, Rik, may be de® ned as the difference between the value of outputs and inputs Qik and Xik at their respective locationspeci® c prices Pik and Cik, Rik 5 Pik Qik 2 Cik Xik .

[1]

Chomitz and Gray (1996) demonstrate that when output is a Cobb-Douglas function of Xij and plot characteristics, s1i ,s2i . . . Qik 5 Sik Xikbk with 0 , b k , 1

[2]

Sik 5 l 0 s l1i1k s l2i2k . . .

[3]

Rik may be written Rik 5

2

1

1 2 b k bk/(12bk) C (Pik Sik bk ) 1/(12bk). [4] bk

By taking logs and collecting coef® cients, this can be transformed into an expression of the form ln Rik 5 a k 1 d k ln Pik 1 q k ln Cik 1

^m n

nk

ln s ni.

[5]

174

Land Economics

Empirically, we distinguish between two forms of land use, agriculture and forestry and note that plot i will be devoted to agriculture if Ri1 . ln Ri0. In practice, data on input and output prices are unavailable at the plot level. We assume that both Pik and Cik vary with the impedance-weighted distance of the plot from the nearest market (Costi ), and, also, with the population density of the district in which the plot is located (Population densityi ). District population affects Pi1 by shifting the demand for agricultural output, and Ci1 by shifting the demand and supply curves of labor. We use district population density, rather than population, to control for the fact that districts vary in area. Plot characteristics {sni } that affect the pro® tability of clearing include slope, elevation, measures of soil quality, and the plot’ s protected area status. Since the government has the right to evict persons living in parks or wildlife sanctuaries, there is at least some threat of expropriation if output is grown in these areas. The province in which the plot is located is also likely to affect the pro® tability of agriculture. Provincial dummy variables capture differences in rainfall and may proxy differences in tenure security. Representing protected area status by Y2i 5 1, if a plot lies in a protected area (and 5 0 otherwise), and all other factors that in¯ uence the pro® tability of conversion (including distance to markets and population density) by vector Zi, a plot will be cleared if ZiB1 1 gY2i . 0.2 In our empirical model, Zi includes the slope of the plot, its elevation, population density in the district in which the plot is located, the natural logarithm of impedance-weighted to market, provincial dummy variables, and dummy variables for soil categories. There is no well-developed theory to explain which plots of land are designated protected areas; however, political and economic considerations suggest that land where the opportunity costs of protection are low (land of low agricultural value) would be more likely to be selected than land of high agricultural value. This suggests that the factors, Zi , that affect the pro® ts of clearing land (the opportunity cost of protection) should enter the equation to explain protected area

May 2001

status. The bene® ts of protecting a plot should, however, depend on different factors. Areas that serve as habitat to endangered species or that contain fragile ecosystems clearly yield higher bene® ts from preservation than areas that are ecologically unremarkable. Riverine forests constitute fragile ecosystems that are often home to diverse species. We posit that location near rivers increases the chance that a plot is protected. The econometric model that we estimate is thus given by Y* 1i 5 Zi B 1 1 gY2i 1 e 1i Y1i 5 1 if Y* 1 i . 0; 5 0 otherwise

[6]

Y* 2i 5 Zi B 2 1 aWi 1 e 2i Y2i 5 1 if Y* 2 i . 0; 5 0 otherwise

[7]

where the plot is cleared (Y1i 5 1) if the net pro® ts from clearing plot i (Y*1i ) are positive, and the plot lies in a protected area (Y2i 5 1) if the net bene® ts from protecting plot i (Y* 2i ) are positive. Wi indicates that the plot is located near a river (watershed dummy). We estimate this structural model as a bivariate probit model, assuming that e 1i and e 2i are jointly normally distributed.3 This allows us to estimate the impact of protected area status on the probability that a plot is cleared. The model is estimated for two de® nitions of protected area: national parks and wildlife sanctuaries (hereafter referred to as ``protected areas’ ’ ), and wildlife sanctuaries only. The focus on wildlife sanctuaries is prompted by anecdotal evidence that the Thai government has made stronger efforts to prevent encroachment in wildlife sanctuaries than in national parks. 2 If Pik and Cik are exponential functions of population density and distance to market, then these variables will enter Zi linearly. Likewise, if {s ni } are an exponential function of plot characteristics they will enter Zi linearly. 3 To reduce the problem of spatial autocorrelation, we sample plots at intervals of 5 km. We have also estimated the model including average values of slope, elevation, and distance to market within a 10-km radius of plot i. The coef® cients of these variables measured for plot i are robust to the inclusion of the average values of the variables on surrounding plots.

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

175

FIGURE 1 Forest and P rotected Areas Map of North Thailand, 1986

III. STUDY AREA AND DATA The areawe havechosenforthisstudyÐ the 17 provinces that constitute North ThailandÐ remains heavily forested, especially the Upper North portion of the region.4 Protected areas constituted 11% of the regionin 1986, the year of our study (see Figure 1), and continue to be established.NorthThailandis the secondpoorest of the four regions of Thailand, and road building is partof the government’ s strategy to reduce rural poverty. Between 1973 and 1985, extensive road building increasedroad density in North Thailand by 57% (Cropper, Grif® ths, andMani 1999). The policy issues raisedin the introduction are, therefore, relevant to North Thailand.

Data

We model the land clearing decision in North Thailand in 1986 using coarse resolution data. Land use information comes from a 1:1,000,000 Land Development Department map that originally contained 15 land use categories. Urban areas and water were omitted from the study area and the remaining land uses classi® ed as ``forest’ ’ or ``non forest.’ ’ The term ``clearing,’ ’ as used in section 2, is thus synonymous with ``non forest.’ ’ Physiographic factors that should in¯ u4 The Upper North consists of the provinces of Chiang Mai, Chiang Rai, Nan, Lampang, Lamphun, Mae Hong Son, Uthai Thani, Tak, and Phrae.

176

Land Economics

ence the pro® tability of clearing include the soil characteristics of the plot, its slope, and its elevation. All soils in North Thailand are classi® ed by the FAO Soil Map of the World as falling in one of 12 soil categories, de® ned on the basis of soil texture and slope class (Acrisol, Fluvisol, Gleysol, etc.). We represent these soil categories using a series of dummy variables.5 Elevation (in meters) was obtained at a resolution of 30 arc seconds, and the slope of each plot was calculated as the maximum difference between the elevation of the plot and the elevation of each of neighboring plot. (The sources of our data are described in the Appendix.) To compute ease of access to markets, we digitized a 1982 road map of Thailand (1: 1,000,000 scale), distinguishing between paved and unpaved roads. The locations of market towns were obtained from the Digital Chart of the World. To calculate the impedance-weighted distance from each plot to the nearest market town, travel along a paved road was assigned an impedance factor of 1, travel along an unpaved road an impedance factor of 2, and travel from a plot to a road a factor of [100 1 (Slope of Plot)2]. An algorithm was used to compute the shortest distance from each point to the nearest market town.6 River distances were computed in a similar fashion. Population, a proxy for the demand for agricultural products and for labor supply, is measured at the district level using 1990 census data. Population density is calculated using 1990 district boundaries. Because each district is large relative to the size of a plot, we treat district population density as exogenous to the pixel. Protected area boundaries, obtained from the IUCN, indicate that 14.4% of our sample points lie within protected areas (parks and wildlife sanctuaries), while 9.1% lie within wildlife sanctuaries. The percent of protected areas remaining under forest cover is 87% whereas it is 70% for all sample points. Sampling Strategy

All layers of the GIS database were converted to a resolution of 100 square meters, which resulted in over 28,000,000 data

May 2001

points. We sampled points systematically, at 5-km intervals, which yielded 6,550 observations. The three provinces that contained no protected areas were dropped from estimation of the protected area equations, while the ® ve provinces that contained no wildlife sanctuaries were dropped from those equations (see Table 1). Exact collinearity between protected areas and four soil categories (and between wildlife sanctuaries and the same soil categories) necessitated that observations in these soil categories also be dropped (see Table 1). The means and standards deviations of variables for each of the protected area and wildlife sanctuary samples are presented in Table 1. IV. ECONOMETRIC RESULTS Determinants of Land Clearing in North Thailand

We begin by examining how well our model explains land clearing in North Thailand (see Tables 2 and 3). North Thailand is a mountainous area, characterized by parallel hills and valleys that run north to south (see Figure 2). Steep slopes and high elevations have helped to protect much of the area from clearing. Indeed, 70% of the study area was classi® ed as forested in 1986. The model of Table 2 correctly predicts land use (Y1i 5 0) for 91% of the sample points under forest cover. The model predicts clearing less accuratelyÐ only 57% of cleared plots are correctly predicted to be cleared. When the model does predict clearing, however, it is correct 75% of the time (see Table 3). The quantitative impacts of factors that affect the probability of clearing are shown in columns (4) and (5) of Table 2. Phyisographic factors have a signi® cant impact on clearing: Calculated at the means of explanatory variables, the elasticity of probability of clearing with respect to the slope of the plot 5 The distribution of more familiar soil properties (nitrogen or phosphorous context) is known for all plots in a soil category; however, it is not known at the level of an individual plot. 6 Costdistance is a module in Arc/Infoä that calculates for each cell the least accumulative cost of travel from a set of source cells, over a cost surface.

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

177

TABLE 1 Summary Statistics

Variable Total no. of observations Cleared land Slope of plot (degrees) Elevation (meters) Population density 1990 (people/km2 ) Cost82 (impedance-weighted distance to nearest market) Watershed dummy Protected area dummy Wildlife sanctuary dummy Province dummy (Chiang Rai) Province dummy (Chiang Mai) Province dummy (Mae Hong Son) Province dummy (Phayao) Province dummy (Nan) Province dummy (Lampang) Province dummy (Phrae) Province dummy (Lamphun) Province dummy (Uttaradit) Province dummy (Tak) Province dummy (Sukhothai) Province dummy (Phitsanulok) Province dummy (Phetchaboon) Province dummy (Khamphaeng Phet) Province dummy (Phichit) Province dummy (Nakhon Sawan) Province dummy (Uthai Thani) Soil dummy (Af60-1/ 2ab) Soil dummy (Ag16-2a) Soil dummy (Ag17-2ab) Soil dummy (Ao107-2bc) Soil dummy (Ao90-2/ 3c) Soil dummy (I-Lc-Bk-c) Soil dummy (Je72-2a) Soil dummy (Lc100-c) Soil dummy (Lg39-3ab) Soil dummy (Ao108-2ab) Soil dummy (Nd65-3ab) Soil dummy (Vp64-3a)

North Thailand Sample

Protected Area Sample

Wildlife Sanctuary Sample

Mean or proportion (S.D) 6,550 0.425 3.54 (3.87) 472.54 (352.13) 63.44 (67.14)

Mean or proportion (S.D) 4,946 0.307 4.24 (3.94) 546.32 (645.06) 45.64 (53.78)

Mean or proportion (S.D) 4,355 0.263 4.46 (3.94) 578.93 (341.15) 42.56 (55.63)

546.92 (621.68)

636.45 (676.85)

652.96 (700.85)

0.600 0.108 0.069 0.062 0.134 0.077 0.037 0.069 0.075 0.040 0.026 0.046 0.103 0.040 0.062 0.072 0.047 0.026 0.045 0.039 0.119 0.007 0.086 0.062 0.479 0.029 0.045 0.012 0.046 0.068 0.043 0.005

0.569 0.144 0.091 province omitted 0.164 0.102 0.029 0.091 0.095 0.047 0.025 0.054 0.136 0.035 0.059 0.085 0.034 province omitted province omitted 0.044 0.147 0.009 category omitted 0.056 0.598 0.038 category omitted 0.016 category omitted 0.090 0.047 category omitted

0.562 0.263 0.151 province omitted 0.186 0.116 0.033 0.104 0.108 0.054 0.029 0.061 0.155 province omitted 0.067 province omitted 0.039 province omitted province omitted 0.050 0.136 0.010 category omitted 0.049 0.634 0.038 category omitted 0.018 category omitted 0.079 0.036 category omitted

is - 0.48, and the elasticity with respect to elevation is - 0.61.7 Soil quality also matters. Sixty percent of the observations in our sample lie in FAO soil category Ao90-2/3c, which is the omitted soil category in our models. This soil type is distinguished by shallow soils, with low potassium content found on steep slopes. The few pockets of better soil in North Thailand have a higher probability of being cultivated. For example,

the marginal effect of moving from FAO soil unit Ao90-2/3c to FAO soil unit Lc100-c is to increase the probability of cultivation by 7 If we calculate the elasticity at means of forested plots, the elasticities with respect to slope and elevation are much higher: -0.66 and -0.84, respectively. Our discussion here focuses on the models reported in Table 2. Results for the clearing equations in Table 4 are qualitatively and quantitatively similar to those in Table 2.

20.332 8.87

20.077 1.295 23,714.743

Rho Log Likelihood

20.059

20.027 20.0003 0.001 20.059 20.039 20.179 20.094 20.082 20.133 20.105 20.123 20.095 20.079 0.141 0.274 0.014 20.02 0.108 0.224 20.05 0.038 0.361 0.071 20.018

20.475 20.614 0.154 20.24

Elasticity

214.01 0.309

20.068

5.297 9.058 2.297 7.477 1.422 4.163 3.748 21.737 1.865 5.217 6.395 4.755 4.678 5.454 3.216 4.752 8.253 24.29 6.103 1.473 5.193 1.921 23.326 20.387 3.543

Z

24.098

0.034 0.001 0.001 0.192 0.363 1.052 1.042 20.574 0.501 1.381 1.851 1.197 1.331 1.446 0.855 1.334 2.176 20.452 1.397 0.175 0.573 0.309 20.52 20.06 0.188

Coef® cient 0.005 0.0001 0.0002 0.028 0.063 0.253 0.265 20.059 0.096 0.388 0.579 0.292 0.373 0.407 0.194 0.374 0.68 20.052 0.406 0.028 0.116 0.054 20.055 20.007 0.026

Marginal Effect

0.272 0.917 0.09 0.362

Elasticity

b

Marginal Effects calculated from univariate reduced-form equations. Watershed dummy 5 1 if the impedance-weighted distance to the nearest river is less 3 km, assuming no primary roads. Protected area dummy 51 if pixel lay in a Protected Area in 1986. Population density is measured at the district level. c Cost is measured as units of primary road traveled, in km.

a

4,946

210.652 28.095 4.532 29.729 21.085 25.573 22.249 22.453 24.4 23.099 22.769 23.061 22.101 2.916 6.46 0.213 20.687 4.773 2.263 21.677 0.761 5.75 2.536 20.599

20.088 20.001 0.003 20.191 20.12 20.725 20.341 20.278 20.493 20.394 20.491 20.343 20.288 0.384 0.746 0.03 20.107 0.326 0.563 20.17 0.101 0.947 0.215 20.062

Slope (degrees) Elevation (ms.) Population density 1990 (people/km2) b Log (Cost82) c Provincial dummy (Chiang Mai) Provincial dummy (Mae Hong Son) Provincial dummy (Phayao) Provincial dummy (Nan) Provincial dummy (Lampang) Provincial dummy (Phrae) Provincial dummy (Lamphun) Provincial dummy (Tak) Provincial dummy (Sukhothai) Provincial dummy (Phitsanulok) Provincial dummy (Phetchaboon) Provincial dummy (Kamphaeng Phet) Provincial dummy (Uthai Thani) Soil dummy (Af60-1/2ab) Soil dummy (Ag16-2a) Soil dummy (Ao107-2bc) Soil dummy (I-Lc-Bk-c) Soil dummy (Lc100-c) Soil dummy (Ao108-2ab) Soil dummy (Nd65-3ab) Watershed dummy b Protected area dummy (1986) b Constant

Marginal Effecta

Equation [7]

Land Economics

No. of observations

Z

Equation [6]

Coef® cient

Independent variable

Dependent variable Cleared Land (Y1 5 1)

TABLE 2 Bivariate P robit Model Estimated Using P rotected Area Sample 178 May 2001

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation TABLE 3 Accuracy of Bivariate P robit Model in P redicting Clearing (protected area sample) Actual ® Predicted ¯

Cleared

Forested

Cleared Forested Percentage correctly predicted

872 657 57%

296 3,133 91%

Note: Diagonal (bold) ® gures show correct predictions.

FIGURE 2 Elevation Map of North Thailand

Percentage of predictions correct 75% 83%

179

180

Land Economics

May 2001

FIGURE 3 Impact of P opulation Density on P robability of Clearing, Evaluated at Forest Means

36%. The latter soil is distinguished by ® nely textured soils that drain well, have good chemical properties, and are well-suited to growing sugarcane and rice. In general, the soil categories that signi® cantly increase the probability of clearing are loamy, occur at greater depth than soils in the reference category, and are found on ¯ at or moderately undulating plains. Deininger and Minten, in their study of deforestation in Mexico, note that physiographic factors alone explain land clearing almost as well as a model to which socioeconomic variablesÐ speci® cally, population density and market accessÐ are added. The same is true of North Thailand. If we exclude population density and impedance-weighted distance from the model, the percent of observations correctly predicted by the model hardly changes: the percent of observations correctly predicted by the clearing equation falls from 81.1% to 80.7%. Nonetheless, population density and market access do have a statistically signi® cant

impact on clearing. Figures 3 and 4 show the impact of changes in these variables on the probability of clearing, when all other variables are held at their mean values for plots in forest areas. In forest areas mean population density is approximately 40 persons per square kilometer. Doubling this density (and holding all other variables at their means in forest plots) increases the probability that a plot is cleared from about 0.15 to 0.18 (see Figure 3). This relatively modest effect can be explained by the fact that higher population density has two opposing effects on clearingÐ increases in population density may imply higher agricultural prices, which should encourage clearing, but may also re¯ ect higher wages, which should discourage clearing.8 The impact of roads is much larger, especially at the forest fringe. Consider a forest 8 As a referee noted, increases in rural population density may reduce agricultural wages through the factor proportions effect.

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

181

FIGURE 4 Impact of Impedance-Weighted Distance on P robability of Clearing, Evaluated at Forest Means

plot that is 2.5 km from the nearest paved road and 6 km along the road to the nearest market (i.e., with an impedance-weighted distance of 256). As Figure 4 shows, bringing this plot 1.5 km closer to a paved road (i.e., reducing impedance-weighted distance by 150) increases the probability of clearing from 0.18 to 0.23, that is, by 5%. The impact of changes in the road network is further explored in section 5 below. Determinants of the Location of Protected Areas and Wildlife Sanctuaries

As one would expect, the variables that increase the probability that a pixel is cleared in general reduce the probability that it lies within the boundary of a protected area (see Table 2) or wildlife sanctuary (see Table 4). Steeper slopes, higher elevations, and locations farther from market centers increase the chance that land is designated a protected area. The same is true for wildlife sanctuar-

ies, although slope and elevation have a smaller quantitative impact on the siting of wildlife sanctuaries than they do on all protected areas. Higher population density in a district increases the probability that a pixel within the district lies in a protected area, although the effect is quantitatively small. This may re¯ ect a desire to locate national parks near population centers. By contrast, higher population density reduces the probability of siting a wildlife sanctuary in a district. Our results in Tables 2 and 4 support Dixon and Sherman’ s (1990) observation that, in developing countries, areas of low agricultural value are more likely to be designated protected areas in order to avoid political con¯ ict. This point is brought home by estimating univariate probit versions of equation [6] (without either protected area or wildlife sanctuary dummy variables) and using them to predict the probability that plots in protected areas and wildlife sanctuaries are cleared. The average predicted probability of

21.296 8.055

20.334 1.211

20.077

20.026 20.0002 0.001 20.051 20.053 20.159 20.084 20.084 20.131 20.097 20.138 20.088 0.12 0.013 20.009 0.085 0.212 20.052 20.002 0.34 0.131 20.029

Marginal Effecta 20.561 20.02 0.163 20.25

Elasticity

0.018

24.037

0.019 0 20.008 0.292 0.321 0.868 1.438 20.472 0.047 0.55 1.398 0.94 0.835 0.778 1.972 20.572 0.057 20.343 0.729 0.377 20.15 0.266 0.133

Coef® cient

0.017

212.746

2.514 2.997 24.401 9.198 1.251 3.485 5.116 21.456 0.161 1.793 4.562 3.786 3.003 2.439 7.59 23.881 0.221 21.553 6.158 2.271 20.701 1.427 2.052

Z 0.001 0.00003 20.0005 0.017 0.061 0.17 0.388 20.006 0.032 0.113 0.378 0.179 0.175 0.17 0.578 20.022 0.003 20.015 0.077 0.03 20.008 0.017 0.007

Marginal Effect

Equation [7]

0.191 0.459 20.753 0.638

Elasticity

Marginal Effects calculated from univariate reduced-form equations. b Watershed dummy 5 1 if the impedance-weighted distance to the nearest river is less than 3 km, assuming no primary roads. Wildlife sanctuary area dummy 5 1 if pixel lay in a wildlife sanctuary in 1986. Population density is measured at the district level. c Cost is measured as units of primary road traveled, in km.

4,355

210.111 26.861 4.509 28.4 21.781 25.602 22.218 22.942 25.046 23.264 23.73 23.06 2.913 0.341 20.135 3.748 2.703 21.686 0.076 5.647 4.312 20.848

20.09 20.001 0.003 20.179 20.199 20.734 20.341 20.335 20.571 20.405 20.66 20.342 0.379 0.048 20.021 0.277 0.619 20.202 0.012 0.949 0.41 20.104

22890.715

Z

Coef® cient

Equation [6]

Land Economics

a

No. of observations

Rho Log Likelihood

Slope (degrees) Elevation (ms.) Population density 1990 (people/km2) b Log (Cost82) c Provincial dummy (Chiang Mai) Provincial dummy (Mae Hong Son) Provincial dummy (Phavao) Provincial dummy (Nan) Provincial dummy (Lampang) Provincial dummy (Phrae) Provincial dummy (Lamphun) Provincial dummy (Tak) Provincial dummy (Phitsanulok) Provincial dummy (Khamphaeng Phet) Provincial dummy (Uthai Thani) Soil dummy (Af60-1/2ab) Soil dummy (Ag16-2a) Soil dummy (Ao107-2bc) Soil dummy (I-Lc-Bk-c) Soil dummy (Lc100-c) Soil dummy (Ao108-2ab) Soil dummy (Nd65-3ab) Watershed dummy b Wildlife sanctuary dummy (1986) b Constant

Independent variable

Dependent variable Cleared Land (Y1 5 1)

TABLE 4 Bivariate P robit Model Estimated Using Wildlife Refuge Sample

182 May 2001

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

clearing is 0.165 for protected areas and 0.125 for wildlife sanctuaries. These numbers are much lower than the average predicted probability of clearing for all sample points, which are 0.308 for the protected area sample and 0.26 for the wildlife refuge sample. Impacts of Protected Areas and Wildlife Sanctuaries on Land Clearing

We turn now to the impact of protected areas on the probability that land is cleared. The coef® cient of the protected area dummy in the clearing equation in Table 2 is insigni® cant, suggesting that protected areas had no statistically signi® cant impact on forest clearing in North Thailand.9 A much different impression is obtained from a univariate probit model with the same variables as equation [6]. In the univariate probit model (not shown) the coef® cient of the protected area variable 5 -0.199, with a standard error of .076. The impact of switching Y2i 5 1 from Y2i 5 0 is to reduce the probability of clearing by 6 percentage points. This erroneous conclusion occurs because areas designated as protected are less likely to be cleared in the ® rst place. Measuring the impact of protected areas using the Chomitz and Gray/Deininger and Minten approach also leads to a different conclusion than Table 2. Their approach is to estimate a single equation probit model for clearing and then use this to predict the probability that pixels in protected areas are cleared. If we estimate a single equation model for clearing (without the protected area or watershed dummies) the average probability that protected areas are cleared equals 0.165. This is higher than the fraction of protected areas actually cleared (0.132). The analysis of Table 2 however indicates that this difference is not statistically signi® cant. The story is somewhat different for wildlife sanctuaries. In the single-equation version of equation [7] in Table 4, wildlife sanctuaries have a much larger impact on clearing (coef® cient 5 20.303 with standard error 5 0.104) than do all protected areas. In Table 4, the coef® cient of wildlife sanctuaries is ap-

183

proximately the same as in the single equation model (20.334), but has a larger standard error (0.257). Had we been able to identify a better instrument for wildlife sanctuaries than the watershed dummy, we would very likely have estimated the impact of wildlife sanctuaries with greater precision. We therefore conclude that there is weak evidence to suggest that wildlife sanctuaries may have deterred deforestation in North Thailand. These results are consistent with anecdotal evidence (Albers 1999). National parks in Thailand are designed without formal buffer zones to separate parks from adjacent land uses. Park boundaries often become de facto buffer zones, a result supported by our analysis. By contrast, anecdotal evidence suggests a deliberate policy to prevent encroachment in wildlife sanctuaries. V. POLICY IMPLICATIONS OF THE MODELS In this section we use the model to answer two questions of policy relevance for North Thailand. Which protected areas are under the greatest threat of encroachment? And what is the likely impact on protected areas of increased road building? We de® ne the areas of North Thailand under greatest threat of deforestation as those areas under forest cover in 1986 (Y1 5 0) for which the predicted probability of clearing exceeds one-half. Two hundred ninety-three sample points are so threatened, and are plotted on Figure 5. Most of these points are clustered in the low-lying portions of the lower half of the region. This is not surprising given the importance of slope and elevation in explaining clearing. Although only 8 of the 293 points lie strictly within the boundaries of protected areas, most of the points are clustered near protected areas. The national parks of Nam Nao and Thung Salaeng Luang, near the southeastern border of North Thailand are surrounded by areas un9 Following the suggestion of a referee, we also used the length of time a pixel had been designated protected to explain the probability of clearing. This variable was, however, insigni® cant.

184

Land Economics

May 2001

FIGURE 5 Areas P redicted To Be Cleared

der high threat of conversion, as are the Khao Sanam Phriang wildlife sanctuary and the Ramkamhaeng national park, located to the west. We note, in the case of Thung Salaeng Luang, that three-quarters of the area of the park under forest cover in 1986 had a probability of clearing greater than or equal to 0.4. To show how further road building might affect deforestation, we use equation [6] (Table 2) to compute the impact of a 100-unit reduction in impedance-weighted distance to market on the probability of clearing for all our sample points. This is equivalent to bringing a paved road one kilometer closer to each point. We then identify the areas where such an improvement in access raises the probability of clearing above 0.5. There are 207 such points. These points (along with the points predicted to be cleared in Figure 5) are plotted in Figure 6. Not surprisingly, the plots that we predict will be cleared as a result of road building are often clustered near the plots predicted to be cleared in Figure 5. In some cases we predict that road-

building will result in clearing within protected area boundaries. In other cases, road building will lead to development around a park or wildlife sanctuary, suggesting the likelihood of eventual encroachment. This is especially true for the national parks labeled in Figure 6. What are the policy implications of these exercises? Analyses such as ours can suggest where effort should be placed if the goal of protected area management is to prevent deforestation within park boundaries. While our work says little about what tools are likely to be effective in preventing encroachment, it suggests where these tools should be applied. Our models also suggest where road building is likely to increase the threat of encroachment in protected areas, but also where it will not. There are, for example, areas in Figure 6 where improved access to markets is likely to encourage land clearing (and may thereby achieve other objectives, such as reducing poverty), but where protected areas are not threatened.

77(2)

Cropper, Puri, and Grif® ths: Predicting the Location of Deforestation

FIGURE 6 Areas P redicted To Be Cleared after a 100 Unit Reduction in Impedance-Weighted Distance APPENDIX Sources and Layers Comprising the GIS Database Data Layer Land Use Political Boundaries Elevation Rivers Roads

Soil Population Populated Places Slope Protected Areas

Source Land Development Department Bangkok, Thailand University of New Hampshire Digital Elevation Model (EROS web site) http://edcwww.cr.usgs.gov Digital Chart of the World Digitized from paper maps provided by the Land Development Department, Thailand FAO Housing and Population Census, Thailand Digital Chart of the World Derived from the Elevation Map IUCN (World Conservation Union)/ The World Bank

Year

Attribute Categories

1986

15 land use categories

1990

17 Provinces and 168 districts 1 meter intervals

NA Unknown 1982

1972 1990 Unknown

1991

Perennial and nonperennial waterways Paved and unpaved roads

12 FAO soil categories Population at the district level 620 populated places in study area Derived using ``slope’ ’ module in IDRISI National Parks (IUCN category No. II ) & Wildlife Sanctuaries (IUCN category No. IV)

185

Land Economics

186

May 2001

P roperties of Soils of Thailand (%) FAO Soil Category Af60-1/2ab Ag16-2a Ag17-1/ 2ab Ao107-2bc Ao90-2/3c a I-Lc-Bk-c Je72-2a Lc100-c Lg39-3ab Ao108-2ab Nd65-3ab Vp64-3a

Too Wet

Infertile

Sandy

Loamy

Clayey

Slope 1± 8%

Slope 8± 30%

Slope .30%

Depth . 100cms

10 70 55 0 0 0 40 0 70 10 0 10

20 30 20 20 10 0 0 0 0 60 0 10

30 0 15 0 0 0 0 0 0 0 0 0

70 100 85 100 65 100 100 100 60 90 50 40

0 0 0 0 35 0 0 0 40 10 50 60

25 70 35 0 0 16 100 0 60 30 30 75

75 30 65 75 25 50 0 25 40 70 65 15

0 0 0 25 75 34 0 75 0 0 5 10

100 100 100 90 20 66 100 10 100 100 90 40

Source: FAO/UNESCO Soil Map of The World. a Is the comparison Soil Category. Note: These 12 categories of soils are an exhaustive list of soils occurring in North Thailand. The numbers in the table show the percentage of each soil category in all of Thailand with the property shown in the column.

References Albers, Heidi. 1999. Personal communication. Resources for the Future. Chomitz, Kenneth M., and David P. Gray. 1996. ``Roads, Land Markets, and Deforestation: A Spatial Model of Land Use in Belize.’ ’ The World Bank Economic Review 10: 487± 512. Cropper, Maureen L., Charles Grif® ths, and Muthukumara Mani. 1999. ``Roads, Population Pressures, and Deforestation in Thailand, 1976± 1989.’ ’ Land Economics 75 (Feb.): 58± 73. Deininger, Klaus, and Bart Minten. 1996. ``Deter-

minants of Forest Cover and the Economics of Protection: An Application to Mexico.’ ’ Working Paper Number 10. The Poverty, Environment and Growth Working Paper Series. Washington, D.C.: The World Bank. Dixon, John A., and Paul B. Sherman. 1990. Economics of Protected Areas: A New Look at Bene® ts and Costs. Honolulu: East-West Center and Island Press. Nelson, Gerald C., and Daniel Hellerstein. 1997. ``Do Roads Cause Deforestation? Using Satellite Images in Econometric Analysis of Land Use.’ ’ American Journal of Agricultural Economics 79 (2):80± 88.

Deforestation, Land Use, and Property Rights: Empirical Evidence from DarieÂn, Panama Gerald C. Nelson, Virginia Harris, and Steven W. Stone ABSTRACT. Economic conventional wisdom suggests that providing land users with more secure property rights will result in more sustainable land use and less deforestation. In this paper, we use spatial econometric techniques to evaluate quantitatively the effect on land use of designated property rights in three parts of Darie n provinceÐ a national park where no human activity is supposed to occur, and two reserves for indigenous peoples. Results suggest that legal property rights for an indigenous population can in¯ uence land use. Geography appears to be more important than legal protection for the national park. (JEL Q15, Q23)

I. INTRODUCTION Economic conventional wisdom suggests that providing land users with more secure property rights will result in more sustainable land use, preservation of biodiversity, and less deforestation (see Godoy et al. 1998 for a review of literature and a recent empirical examination of this idea). Provision of effective property rights lowers the discount rate of the operator of a parcel, making long-term investments with larger future payouts more desirable. These longer-term investments, which range from preserving existing forests for opportunistic collection of valuable plant materials to perennial crops, are sometimes thought to be more sustainable than shortterm alternatives. Preserving existing forests is almost by de® nition sustainable. But even cultivation of perennial crops is assumed to disturb the soil less often, reducing the potential for soil erosion. Potentially endangered species might be more likely to survive among perennial species than in locations with annual crop production. Annual crops such as vegetables or grains might increase the risk of soil erosion. If an operator of a Land Economics · May 2001 · 77 (2): 187± 205 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

parcel risks the loss of future output because of uncertain tenure, he or she may choose to plant a crop with quick returns that is more environmentally destructive. In this chain from designation of property rights to more sustainable land use, several assumptions are madeÐ of® cially designated property rights translate into effective rights, long-term investments are more sustainable than the alternatives, and operators maximize pro® ts. In this paper, we examine empirical relationships between designated property rights, and actual land use in the DarieÂn province of Panama, a remote and environmentally important region of Central America. A sustainable development project ® nanced by the Inter-American Development Bank (IDB) in 1998 collected spatially explicit data on land use, property rights, and culture. In the past, it has been dif® cult and expensive to investigate speci® c geographical impacts of property rights due to the high costs of collecting and analyzing spatially explicit data. However, new developments in geographic information software, new data sets, the rapidly declining costs of computational capacity, and the development of new analytical techniques are making spatially Gerald Nelson is an associate professor and Virginia Harris is a doctoral candidate, in the Department of Agricultural and Consumer Economics, University of Illinois, Urbana-Champaign. Steven Stone is an environmental specialist at the Inter-American Development Bank (IDB). This paper is based on research in support of an IDB-® nanced project in DarieÂn province, Panama. The authors would like to thank Robert Kaplan, Jack Hastings, Heli Nessim, Luis Zavaleta, Julieta Diaz, and Marco Fiero for their support, Luc Anselin for invaluable econometric advice and Alessandro De Pinto for research assistance. In addition, two anonymous reviewers provided valuable suggestions for improvements. The opinions expressed herein are the responsibilities of the authors, and do not necessarily represent the positions of the Inter-American Development Bank or the University of Illinois.

188

Land Economics

FIGURE 1 Darie´ n P rovince Overview

May 2001

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

explicit, quantitative assessments more feasible.1 The goal of this paper is to use spatial econometric techniques to evaluate the effect on land use of designated property rights in three parts of DarieÂn provinceÐ a national park where no human activity is supposed to occur, and two long-standing reserves for indigenous peoples. We use a multinomial logit model to estimate the effects of socioeconomic and geophysical variables on the choice of land use and present a spatially explicit technique for evaluating predictive power of the regressions. Finally, we simulate the effects on land use of changing property rights. This approach allows us to examine both the qualitative hypothesis of more sustainable land use with more effective property rights and make spatially explicit quantitative simulations of that effect. An important econometric issue is how to account for the possibility of spatial autocorrelation. We compare the effects of two easily computable spatial lags. II. A MODEL OF THE DETERMINANTS OF LAND USE2 Three sets of economic variables determine land use for a given parcel. The ® rst set is the location’ s geophysical characteristics. These might be vegetative (timber, productive soil), mineral, or even atmospheric (rainfall, evapotranspiration). A second set of characteristics is socioeconomic; the location-speci® c attributes such as prices of inputs and outputs and degree of operator control over the parcel.3 Finally, geophysical and socioeconomic variables combine with a set of production technologies that relate inputs and outputs. We assume the operator of the location (the person with effective control over the land) uses its resources to increase his or her (or their, in the case of common property) utility. In this section we equate utility and pro® t maximization. However, we relax that assumption in the empirical analysis. The operator chooses a particular land use by comparing the net present value of the pro® tability of all possible land uses. If we assume that a given land use has a single marketed

189

product, the net present value of the return to that land use, its rent (Rhl) at time T, is given by: RhlT 5

#

¥

t50

(PhlT1t QhlT1t 2 ChlT11 XhlT1t)e2i l tdt, [1]

where P is the output price, Q is the quantity of output, C is a vector of input costs, X is a vector of inputs under operator control and il is the location-speci® c discount rate, all for each land use h at location l at time t. The operator chooses the land use that has the highest RhlT for the parcel. 1 For a general discussionof spatial analysis and examples of its use in World Bank development projects, see Nelson and Gray 1997. 2 The static version of this model was originally developed in Chomitz and Grey 1996 and used to assess the effects of roads on land use in Belize. Nelson and Hellerstein 1997 extended the theoretical model to multiple time periods and used the approach to simulate the land use effects of complete removal of a road network in central Mexico. Other authors who have used this methodology to study determinants of land use in developing countries include Deininger and Minten 1999. Bockstael and associates at Maryland have used a similar methodology to study urban expansion in the Washington, D.C./Baltimore, Maryland corridor (Bockstael and Bell 1998). See Kaimowitz and Angelsen 1998 for a review of deforestation models. 3 A reviewer has suggested that some parcels are not subject to human control and that differences in land use, or more appropriately land cover, are determined solely by geophysical characteristics. In that case, the changes in a socioeconomic variable should not lead to a change from one ``natural’ ’ land use to another, and socioeconomic coef® cient values should be constrained to re¯ ect that fact. We agree with this proposition in the abstract. However, there is evidence that even the most remote parts of this region have had human intervention for (at least) hundreds of years. See Bush and Colinvaus (1994) for a fascinating discussion of research that demonstrates this intervention at a remote location. Hence, we have assumed that all observed land uses are the result of some human intervention and estimate unconstrained coef® cients. In addition, one must distinguish carefully between the effect of a change in a socioeconomic variable on absolute pro® tability of a land use and the actual land use choice. It is entirely plausible that an increase in the price of coffee in Panama City would make the pro® tability of coffee production in a DarieÂn wetland increase yet no coffee production would actually take place there because the pro® tability increase was from a large negative number to a slightly less large negative number. In essence the importance of the geophysical variables swamps that of the socioeconomic variables for some locations.

Land Economics

190

The theoretical derivation, presented in an appendix, makes several restrictive assumptions to arrive at a theoretically-consistent reduced-form equation that includes prices of inputs and output, a vector of geophysical characteristics (G), parameters of the production function (a and b), and a locationspeci® c discount rate, il. The discount rate is location-speci® c to capture differences in effectiveness of property rights and cultural values. In addition, data restrictions impose additional constraints. As discussed below, we use only cross-section information. Hence prices are assumed to have no temporal variation. However, we do have spatial variation in prices caused by transportation costs. The reduced form equation is:

3 p 5 b 3P G p C

RhlT 5 bh Phl Gl

2akh khl

a

2akh khl

a

C

k

h

hl

l

k

1 bh

4 #e 4 11i 2.

akh kh

akh kh

¥

2ilt

dt

t50

1 bh

[2]

goods are transported out of the province. From most of the province, goods move either along the Pan American highway north or to El Real, a river port, and onto a coastal seagoing vessel. We combined these two destinations, estimating the cost to the nearest one using a cost-minimization algorithm. Our fourth destination is a Paci® c coast port, the town of Puerto PinÄa in the southwest. Puerto PinÄa is a viable destination only for goods produced nearby since mountains surround the town and there are no access roads to the rest of the province. An important point to note is the existence of several navigable rivers. In particular, the Chucunaque River parallels the Pan American highway and is navigable in small boats from its mouth to a point at least 20 kilometers north of the highway’ s southern terminus in Yavisa. Substituting the price proxies anddoing additional manipulations (see appendix) gives: ln RhlT 5 h0h 1

l

^h i

1ih

Dl 1

1 h3h ln il 1 uhl.

Since we have no data on either locationspeci® c or ® nal destination prices, we follow Chomitz and Gray (1996) and proxy location-speci® c prices by measures of costof-access to different ® nal destinations with the following functional form: Phl 5 exp [g01 1 g11 D1] Chl 5 exp [d01 1 d11 D1]

May 2001

[3]

^h r

Gr

2hr

[4]

Parcel h will be devoted to land use k if RhkT . RhlT, " l ¹ k. If the u are Weibull distributed and uncorrelated across land uses, equation [4] is equivalent to a multinomial logit model where: Probhl 5

eVihh

^e

Vihj

.

[5]

j

Dl 2 cost of access measure from ® nal destination or source of input to location l. Note that this form assumes that the price proxies for all inputs (Ckhl) are the same.4 This assumption seems reasonable for bulk commodities that are transported in similar size lots, for example, in trucks or barges. It is probably less correct if the commodities produced range in variety. For this analysis, we identi® ed four distinct ® nal destinations that might be of relevance to land use choice. Some activities produce goods for home or local consumption. Two ® nal destinations for this type of activity are proxied by the nearest village and the nearest local population center. In addition, some

For estimation, the V vector consists of three sets of explanatory variables: G, site-speci® c geophysical variables, D, cost-of-access and other socioeconomic variables, and S, spatial effects geophysical variables (discussed below). To avoid identi® cation, the hs for land use 0 are set to zero. The remaining hs can be interpreted as the marginal effects of right-hand side variables on the ln of th/e the ratio of the probability of a land use choice to the zeroth land use. More generally, 4 This assumption does not mean the effect of a change in access cost is the same for all land uses. See Greene (1993) for a theoretical explanation and Nelson and Hellerstein for an example.

77(2)

12

¶ln

Pi Pj

¶Vi

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

5 hi 2 h j.5

[6]

We use the estimated hs to generate probability predictions for each land use at every location in the province.6 These probabilities can then be used to predict land use at each location. The predicted land use is determined by the highest probability value of the land uses. Static Versus Dynamic Considerations

An implicit assumption made above is that land use categories in 1997 are in equilibrium. Any land use change initiated by some form of intervention will take some time to complete. The completion of the Pan American highway in 1983 began a process of land use change as the cost of reaching the interior of the province by road dropped dramatically. The time series data available to us are land use categories in 1987 and 1997. We have assumed that in 1987 a post-road land use equilibrium had not been reached, but that by 1997 land use change from the construction of the road was ® nished. We have also assumed in our simulations that land use change is reversible; that is, a human intervention pixel can revert to a forested pixel. Assessing Predictive Power

The most frequently used method of assessing the predictive power of a logit regression is to calculate a ``prediction matrix’ ’ comparing actual and predicted categories. The matrix rows show the number of locations actually in a given category; its columns show the number of locations predicted to be in a given category, where the predicted land use is the one with the highest probability.7 Diagonal elements are correct predictions.8 This technique, however, does not give any measure of the power of the prediction. For example, with ® ve categories it is possible that the probability values for four of the categories are equal to 0.19, and the value for the ® fth category is 0.24. While the ® fth category has the highest probability and would therefore be the predicted category,

191

the strength of this prediction is small. We developed two new, related graphical measures of predictive power. The ® rst maps the maximum probability value, Pmax, at every location. This map gives a spatial representation of the power of the prediction but does not convey any information about prediction accuracy. The second measure maps Pdiff 5 Pmax 2 Pactual (the probability value for the actual land use). If the category with the highest probability value is also the actual category, Pdiff 5 0. Otherwise 0 , Pdiff , 1. Simulation Techniques

With estimates of the hs, we can simulate changes in the socioeconomic variables. We simulate removal of legal restrictions on land use in three areasÐ DarieÂn National Park and two reserves for indigenous groups by setting the respective dummy variables to zero. DarieÂn National Park is supposed to have no human intervention except for a preexisting mineral concession. There is no road access to the park and the Chucunaque River is located between the park and Yavisa.9 However, there is a small indigenous population living there, and resources for enforcement 5 One must interpret this effect carefully. An increase in the probability of a land use relative to the base land use (or any other for that matter) may have no signi® cance on its likelihood of being ``chosen’ ’ when compared to other possible land uses. 6 For example, we might ® nd the following land use probabilities at a location in the middle of the parkÐ forest without cuipoÐ 72%, forest with cuipoÐ 18%, agricultureÐ 3%, pastureÐ 6%, etc. The sum of probabilities for all ® ve categories is 100%. 7 Chomitz and Gray (1996) propose an alternate approach that involves assigning a location to a ``natural’ ’ land use only if its predicted probability is higher than the actual ratio of that land use to total land area. We have chosen not to implement this approach, in part because we have no single land use category or set of categories that ® ts their approach. 8 The prediction matrix is like a ``confusion matrix’ ’ in the remote sensing literature that compares categories identi® ed by a classi® cation scheme to categories identi® ed by ground observation (Richards 1993). 9 Vehicles do occasionally traverse the park. Companies offering off-road adventure advertise a trip through the park and its Colombian counterpart as an extreme adventure. In addition, it is likely that park lowlands provide a path for Colombian immigrants to reach the Pan American highway.

Land Economics

192

are extremely scarce. Three areas within the province have been set aside for indigenous groupsÐ the Kuna de WalaÂ, Mortõ  y Nurra Indigenous Reserve in the northwest, the Choco homelands Comarca Embera No. 1 in Cemaco District, and Comarca Embera No. 2 in Sambu District. The Kuna Reserve was created only in 1997, the year for which we have data, and has not yet been demarcated so we did not include it in our analysis. Information gathered in the ® eld suggested that the population of the Cemaco district exercised its rights more effectively than that in the Sambu district. For all alternatives, we generate transition matrices10 and land use change maps. The null hypothesis is that removal of protection will increase the extent of human intervention. With the simulation, we can also estimate the magnitude and location of changes. III. DATA SOURCES, MANIPULATION, AND ECONOMETRIC ISSUES The principal source of data was a spatial data set prepared as part of the IDB project by the engineering ® rm of Dames and Moore. The data consist of over 80 coverages11 stored in Arc Info and Arc View ® le formats. Variables include land use, location of population centers by ethnic group, temperature, elevation, and political boundaries. Land Use

Dames and Moore created land use data sets for 1987 and 1997 from a variety of sources, including interpretation of satellite

May 2001

images and ground-truthing (Table 1 and Figure 2).12 The original land use classi® cation had 15 categories, but some represent areasÐ interior waters and populated areasÐ that are essentially ``no data’ ’ for the purposes of our analysis. To simplify the interpretation, we aggregated the original land uses into 5 categories (0 to 4 in Table 1). The original data were in vector coverages; we converted these to raster coverages with each pixel representing an area 500 meters on a side (0.25 sq. km).13 10 A transition matrix has one state of nature (e.g., existing land use) along the vertical and a second state of nature along the horizontal (e.g., land use without a reserve). A matrix cell contains the number of members common to both the ® rst and second state. For example, if 9 square kilometers of cativo forest in 1987 were converted to human intervention in 1997, the intersection of the 1987 forest column with the 1997 human intervention row would be 9. 11 A coverage is a set of location-speci® c values for a variable. In a vector coverage, location information is given as a polygon or other vector format. In a raster coverage, location information is given in a grid of squares, sometimes called cells or pixels (short for picture element). Pixel and cell are used here synonymously with location of a parcel. 12 We have little information on quality of the data. Informal communications with the Dames and Moore staff who prepared the data set indicate that some ground truthing was done for the land use classi® cation process, and extensive use was made of data prepared by the Instituto ``Tommy Guardia,’ ’ a Panamanian organization somewhat similar to the U.S. National Geographic Society. 13 This choice of pixel dimension is somewhat arbitrary. The relevant unit of analysis would be a location with a single land use over which a single decisionmaker has control. This might be a 1/2 hectare ® eld in an agricultural area or several hundred hectares of a forested area. The Dames and Moore data set has land use polygons that range from very small to very large. However, these are not based on control characteristics.

TABLE 1 Land Use Categories and Areas, 1987 and 1997 1987 Category (number in parentheses) Bosque sin Cuipo (forest without cuipo) (0) Bosque con Cuipo (forest with cuipo) (1) Cativo (2) Marsh, fresh and saltwater (4) Human intervention (3)

1997 Area (sq. km)

4,631 8,085 246 729 2,420

Source: Calculations based on data collected by Dames and Moore.

% Change 4,613 6,608 234 717 3,946

20.4 218.3 24.9 21.6 63.1

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

193

FIGURE 2 Darie´n P rovince Land Use, 1987 and 1997

The original data set identi® es ® ve categories of land use involving predominantly tree cover. In 1997, forests accounted for about three-fourths of the province, a decline of about 10% from 1987. Two categoriesÐ forest with and without cuipo (bosque con cuipo and bosque sin cuipo)Ð account for most of the forest area. The principal distinguishing characteristic of the two is the presence or absence of cuipo trees. This tree resembles the African baobob in appearance. Its wood is unsuitable for most timber needs. The cuipo-dominated forest is located at lower elevations and relatively close to human activities. The non-cuipo forest is located in more remote locations and locations with higher elevation and steeper average slopes (Table 3). Some observers have suggested that cuipo-dominated forest is at least partly the result of past human intervention; that

the cuipo tree can become established after slash and burn agriculture, especially in the low lands. The third important forest category is cativo. This is the most important commercial species in the province, used principally in the plywood industry. Since it often grows in pure stands it is especially susceptible to clear cut logging. During the 1987± 1997 period, the largest decline in forest area was in the forest with cuipo category. Almost 20% of this category was converted to other uses. Brush, pasture, and agriculture each got about 5% (Table 2). The province Conversion to regular units (square pixels) simpli® es the analysis. The choice of 500 meter pixels was a compromise between data volume (smaller pixels mean more observations) and our very imperfect understanding of the size of units under effective control by a single decision-maker.

Land Economics

194

has about 700 sq. km of mangrove and other wetland areas. These are located almost exclusively around the Gulf of San Miguel. There was relatively little change in this area between 1987 and 1997. We have aggregated these into a ``marsh’ ’ category. The remaining original land uses re¯ ect human intervention in the landscape. The largest categories are livestock pasture (ganaderõ Â a) and brush (matorrales). Project staff described the brush category as the land cover following slash and burn cropping practices or where pastures were abandoned. There was a very large increase in both these categories with the brush area doubling and the livestock area increasing by more than 50% (Table 2). As can be seen in Figure 2, most of this increase came in the northwest part of the province, largely at the expense of the forest with cuipo category. Agricultural area increased by almost 75% between 1987 and 1997, but from a relatively small base. Much of the increase took place at the edges of existing cultivation and from cuipo forests (Table 2). One exception is a large area of agricultural activity near the northernmost corner of the province. It is also informative to look at the spatial distribution of changes in land area between

May 2001

1987 and 1997. The center graph in Figure 2 shows that most of the change took place in the northern part of the province. Land use changes took place on both sides of the newly built Pan American highway (completed in 1983). However, the changes on the east side were located predominately in the narrow strip between the highway and the Chucunaque River, outside the Cemaco Reserve. Changes on the west side of the highway took place throughout the northwest corner of the province. This differential pattern of development suggests that the existence of the reserve had some effect on land use. Identi® cation of the human-intervention land use categories was based on interpretation of satellite imagery and aerial photography. There is a tendency for land uses such as pasture and crop land to have similar spectral characteristics, increasing the likelihood of misclassi® cation. Cropping practices further increase this possibility. Forest is cleared for agriculture and/or pasture. After a few years, the cleared land is abandoned, changing gradually to brush. The predictive power of our results for these classes individually was much lower than for the other land use categories. As a result we aggregated them into a single human intervention category.

TABLE 2 Land Use Transitions, 1987 to 1997 (sq. km) 1997 Land Use 1987 Land Use (category number in parentheses)

a

b

c

d

a. Forest without cuipo (0) b. Forest with cuipo (1) c. Forest with cativo (2) d. Disturbed forest, secondary forest, forest plantations (3) e. Mangrove areas (4) f. Fresh water marshes (4) g. Scrubs or secondary growth (3) h. Pasture (3) i. Agriculture (3)

4,588 5 0 0

11 6,499 1 57

1 1 226 0

5 377 0 200

0 0 0

1 17 3

0 0 0

0 0

15 1

Total, 1997

4,593

6,604

e

i

Total, 1987

3 525 2 133

7 250 2 17

4,624 8,090 240 552

2 0 457

4 0 29

0 0 46

423 220 563

0 0

76 29

770 7

8 341

898 382

225

1,130

1,473

671

15,991

f

g

h

1 1 0 0

0 26 0 0

9 406 9 144

14 5 28

401 1 0

2 197 0

0 0

28 5

1 0

228

663

405

Note: Entries in a cell indicate the number of sq. km that were in the row land use in 1987 and the column land use in 1997. For example, 525 sq. km of forest with cuipo in 1987 had been converted to pasture by 1997. The entries along the diagonal are areas where land use has not changed. The totals differ from those in Table 1 because some locations were identi® ed as missing in one period, while other locations were missing in the other period.

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

Geophysical Data

Geophysical variables determine the potential productivity of different land uses at a location (see Table 3). We started with the basic data provided by Dames and Moore and then manipulated them to generate other variables for the analysis. Temperature. The temperature data set provided by Dames and Moore has values ranging from 21.5 C to 27.0 C degrees average annual temperature. As might be expected the lower temperatures are found at higher elevations. The correlation coef® cient between temperature and rainfall is 0.99; hence we exclude the rainfall variable from our analysis. Elevation and slope. Elevation values were derived from a radarsat raster image with pixel size of 92 meters. The average slope was calculated from the original data. Then both elevation and slope data were resampled to 500-meter cells. The highest point in the province is 1,800 meters but much of the province is close to sea level. The average elevation is 291 meters and the average slope is 6.7 degrees. However, some land use categories have very different average values. The average elevation of a forest without cuipo location is 658 meters and its average slope is 13.51 degrees. These values indicate the rugged topography where forest without cuipo grows; one reason it remains inaccessible and well conserved. Soil quality. A soil map was one of the data sets collected for the IADB project. Conversations with the staff assembling the data indicated the map was scanned from an existing soil map whose provenance was not clear. We experimented with a variety of ways of incorporating the soil data into the analysis. For the analysis here, we constructed a soil quality index from this map, with 1 being the lowest quality, and 7 the highest quality. Socioeconomic Data

The socioeconomic variables are chosen to re¯ ect the in¯ uence of humans on land use (see Table 3). Four 0-1 dummy variablesÐ for the national park, the Cemaco and Sambu reserves, and concession areasÐ are included

195

to indicate demarcation of legal property rights. Forty seven percent of forested pixels are in the park, and a further 27% are in the Cemaco reserve. Only 3% of the pixels with human intervention were inside the park, 7% in the Cemaco reserve, and 5% in the Sambu reserve. Four of the coef® cients indicating effects of socioeconomic variables on category 4 (marsh) have been set to zero because there was insuf® cient variation in the data set. These are for the location dummies for the park, concessions, and the Cemaco and Sambu reserves. Darie n National Park. DarieÂn National Park was established in 1980. It is located on the southern border of the province and makes up about one third of the provincial area. UNESCO designated the park a World Heritage Site in 1981 and a biosphere reserve in 1983. With a few exceptions (such as mineral concessions that preceded the park creation and a few indigenous groups with no formal rights of access), no economic activity is supposed to take place within these boundaries. However, the ability to enforce this restriction is limited by the small number of park personnel. Concession areas. The land use data set includes information on the two types of concession areas in the provinceÐ forest and mineral. Two mineral concessions are located in the park; the remaining concessions are just outside the park boundaries. The forest concessions are in three groupsÐ one near the park border, one in the middle of the province, and several in the northeast corner of the province. All concessions were aggregated into a single dummy variable. Cemaco and Sambu Reserves. Although three areas have been set aside for indigenous groups in the province, we have included only two in the econometric estimationÐ the Choco homelands Comarca Embera No. 1 in Cemaco District and Comarca Embera No. 2 in Sambu District. These reserves were set up in 1983. Discussions with project staff suggested that residents of these reserves, especially the Cemaco reserve had been most successful in using the reservation status to gain effective property rights. The third reserve, the Kuna de WalaÂ, Mortõ  y Nurra Indigenous Reserve in the headwaters area of the Chucunaque River, was created in

ID Area

0.77 0.02 0.22 0.13 126.68 256.59 72.56 87.81

4.32 1.58

0.27 0.05 0.23 0.07 37.00 242.80 16.30 26.17

4.84 0.40

1.79 26.2 4.80 4.86

1 6,604

Forest with cuipo

0.00 0.18 0.36 0.14 16.94 231.93 3.41 9.08

2.65 0.07

0.72 26.9 1.13 2.6

2 228

Forest with cativo

Source: Own calculations using Dames and Moore data set.

DARPARK CONCESSN CEMACO SAMBUÂ COST2PTS COSTPP COSTVILL COSTTWN

Socioeconomic variables

LSOIL LSLOPE

Spatial lag variables

ELEV TEMPERA SLOPE SOILINDX

6.58 24.8 13.51 4.39

0 4,594

Name

Geophysical variables

Forest without cuipo

0.03 0.04 0.07 0.05 15.1 221.80 1.10 3.56

3.96 0.33

1.03 26.5 3.05 4.01

3 3,936

Human intervention

0.00 0.00 0 0 14.05 223.58 3.70 5.38

1.92 0.19

.39 26.9 1.70 2.01

4 225

Marsh areas, fresh and salt water

0.34 0.04 0.18 0.08 56.16 240.67 28.03 37.24

4.34 .71

2.91 25.9 6.70 4.37

15,995

Province

1 2 in park 1 2 in concession 1 2 in Cemaco reserve 1 2 in Sambu reserve $/mt cost to nearest of northern border or El Real $/mt cost to Puerto Pina $ cost to nearest inhabited village $/mt cost to nearest town

Ave. of SOILINDX in 8 neighboring pixels Ave. of SLOPE in 8 neighboring pixels

100 meters meters degrees 0 is lowest quality; 7 is highest

sq. km

Units

TABLE 3 Mean Values for Explanatory Variables for the P rovince and Within Each Land Use Category, 1997

196 Land Economics May 2001

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

1997, the year of our analysis. A reviewer made the suggestion that comparison of land use in the newly created reserve, where any effects from protection should not yet have occurred, with land use in the older reserves would provide valuable insights into the effects of protection. Unfortunately, the precise location of the new reserve was not included in the data sets collected for the project. Hence, we include only the locations of the older reserves. These indigenous groups have extensive knowledge of the forest, and despite long contact with outside societies, have managed to keep some of their traditional forest cultures. The Kuna have been studied extensively (see D’ Arcy and Correa-A. 1985); their language and behavior are in part directed by their relationships with wild animals and plants and the symbolic and magical features they represent (Chapin 1991). (http://nmnhwww.si.edu/botany/projects/ centres/darien.htm#E). Hence, we might expect utility, rather than pro® t, maximization to determine behavior. On the other hand, research just north of the province comparing the land use patterns of indigenous and migrant groups could ® nd no statistical differences (Simmons 1997). Cost of access data. Ideally, we would use prices of important inputs and outputs to determine the most pro® table land use at each location. These data are unavailable. However, we do have indirect information on the cost of transporting wood, an important product of the region. Although imperfect, we use timber as a proxy for all transported outputs.14 In a study of logging in Brazil, Stone estimated the costs of transporting a cubic meter of wood over various land surfaces and navigable waters (Stone 1998). We calibrated these cost estimates to re¯ ect local conditions based on information gathered in the ® eld and in consultation with project staff. Rivers provide an important source of transportation. Project staff identi® ed several stretches of rivers in the province that are navigable by small boats. These include the Chucunaque River as far north as the Meteti area, the Tuira River as far south as Boca de Cupe, and the Sambu River to Boca de Sa-

197

balo. In addition, we assumed that the gulf waters were navigable at the same cost as the rivers. We added an additional ® xed cost to move goods onto or off the gulf to re¯ ect loading costs. An important implication of these assumptions is that the cost of moving goods out of the province via a water route (river and gulf) is similar to using the Pan American highway. Where cost information was not available from Stone, we used estimates based on discussions with project staff. Our per-metric-ton cost estimates of traversing a kilometer with the following land uses are: primary road, $0.10; secondary road, $0.15; navigable river, $0.08, forest, $3.00, human intervention, $0.2± 0.5; marsh, $3.00. These costs are then adjusted to re¯ ect the higher cost of moving over sloping ground.15 As might be expected, the lowest cost points are those nearest the destinations and along roads and navigable rivers that provide easy access to those destinations. The average cost of access to different land use categories varies quite dramatically (Table 3) and is indicative of their locational rents. The average cost from a forest without cuipo pixel to the lowest cost destination outside the province is $127 because these locations are both remote and have a steep slope. The average cost from a human-intervention location is only $15. Econometric Issues Endogeneity of roads. The use of road location in constructing cost-of-access explanatory variables implies a one-way causality from road location to land use choice. As many observers have pointed out (e.g., see 14 Timber is similar to other agricultural products in weight and volume. It may be the case that other agricultural products are exported from the province in less than full truckload lots and hence we are underestimating the transport cost for those commodities. In addition, as mentioned in the methodology section, the transport costs of some commodities such as cattle or perishable fruit might differ from timber and bulk commodities. 15 The formula for the ® nal cost was (1 1 slope2/50) times land cover cost. This formula is arbitrary. It has the desirable properties that as slope increases, cost increases more rapidly. When the slope is 10 degrees, the cost doubles.

198

Land Economics

Chomitz and Gray 1996), it is possible for the location of these socioeconomic variables to be in¯ uenced by the same variables that determine land use. Roads might be built to provide improved access to locations where soil and climate conditions favor agriculture. While these possibilities are in general possible, they are less likely to be a concern for this analysis. The Pan American highway construction was motivated less by regional access concerns (the region’ s limited population was already served by navigable rivers), than by supranational interests in connecting the Americas.16 Nonetheless, we include results using an instrumental variable approach to correct for this endogeneity, along the lines originally used in Chomitz and Gray. First, new cost-of-access variables are constructed with the cost of traversing a road pixel replaced by the cost of traversing the underlying land use. In essence, the roads are removed from the transportation cost calculation.Then,we runaregressionexplaining each cost-of-accessvariable using the newcostvariable as an explanatory variable and the remaining right-hand-side variables from the main regression. The coef® cients of almost all RHS variables differ signi® cantly from 0 because of the large number of observations, but the distance variable with roads removeddominates the regressions, with coef® cient values ranging from 0.93 to 0.99. The adjusted R2 is greater than 0.99 in all regressions. All results reported in this paper are based on the instrumental variables for cost of access. Spatial lag variables. The variables described above are derived from data where potentially important spatial relationships exist. For example, all else equal, a location is more likely to have human intervention if the neighboring locations have human intervention. There are two types of potential econometric problems from not incorporating these spatial relationshipsÐ spatial dependence and spatial autocorrelation (Anselin 1988). With spatial dependence, there is a spatial relationship in the error structure, possibly caused by an omitted explanatory variable. For example, rainfall will affect vegetation on all neighboring locations. The cost of not correcting for spatial dependence is inef® cient but unbiased parameter estimates.

May 2001

With spatial autocorrelation there is a causal spatial relationship between the dependent variable and its neighbors. For example, the existence of forest at a location is the result of seeds spread from forest at a neighboring location. Spatial autocorrelation results in biased estimates of the parameters. There are formal tests for spatial dependence and autocorrelation in the case of continuous left-hand-side variables. However, for qualitative dependent variables there are no formal tests.17 We experimented with two adhoc corrections for the possible existence of spatial relationships. Both have the advantage of easy calculation. The ® rst correction was to add two indices for latitude (values ranging from 1 in the north to 417 in the south) and longitude (1 in the east to 295 in the west). These variables capture only spatial effects that occur along the lines of the compass. A second regression was done using spatial lags of two right-hand-side variablesÐ slope and the soil index. Each lag variable is the average of the values of the original variable in the eight pixels surrounding the location. We refer to these as the latlong and xlag alternatives. IV. RESULTS Predictive Power of the Regression

We estimated the multinomial logit model (equation [5]) using 1997 data. We have 5 land use categories (see Table 5) and the 15 explanatory variables listed in Table 3. We estimated equation [5] both with the original cost-of-access variables and with instrumental variables. Since there was little difference in the results, all further analysis was done with instrumental variables. We also estimated equation [5] using the two alternate approaches to dealing with spatial correlation described above. There was relatively little 16 U.S. funding for construction of the remaining stretch to connect the southern and northern portions of the highway was not forthcoming because of U.S. concerns about hoof and mouth disease transmission. 17 Two working papers provide preliminary theoretical results for how such tests might be constructed but no working implementations of these tests are available (personal communication from Luc Anselin).

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

199

TABLE 4 Estimated Coefficients and Standard Errors 1 (forest with cuipo)

Constant ELEV TEMPERA SLOPE LSOIL LSLOPE SOILINDX DARPARK CONCESSN CEMACO SAMBU PCST2PTS PCOSTPP PCSTVILL PCOSTTWN

2 (forest with cativo)

3 (human intervention)

4 (marsh)

coeff

SE

coeff

SE

coeff

SE

coeff

SE

12.168 21.769 20.111 20.055 20.075* 0.087 0.055 0.006* 4.661 0.060* 0.421 20.218 20.029 20.153 0.064

0.897 0.027 0.033 0.004 0.055 0.025 0.053 0.048 0.151 0.054 0.072 0.009 0.007 0.020 0.018

241.955 24.263 1.757 20.239 20.576 0.208* 0.034* 0 6.705 0.119* 2.079 20.491 0.209 22.341 1.180

5.754 0.189 0.190 0.035 0.083 0.228 0.074 Fixed 0.195 0.114 0.216 0.144 0.082 0.131 0.186

9.141 21.739 20.101 20.023 20.211 0.637 0.043 20.328 4.372 21.143 0.010* 20.003* 0.206 26.403 20.979

1.544 0.042 0.050 0.007 0.061 0.044 0.058 0.084 0.168 0.069 0.089 0.034 0.033 0.098 0.072

18.089 28.748 0.785 20.040 20.518 1.416 0.028* 0 0 0 0 21.457 21.124 1.874 20.086*

3.901 0.171 0.135 0.016 0.070 0.097 0.064 Fixed Fixed Fixed Fixed 0.056 0.054 0.168 0.134

Note: Cost variables are instrumented. For coef® cients indicated with a *, the ratio of the coef® cient to its standard error is less than 2. Five coef® cients have been restricted to zero because there was insuf® cient variation of that variable for the respective land use. One commonly used measure of overall predictive power is R*2 5 12 ln L/ ln L0 (see, e.g. Greene 1993, 651), where L is the value of the likelihood function and L0 is the value of the likelihood function with coef® cients constrained to 0. For this regression, R*2 5 0.84.

difference between the two, and with regressions with no spatial lags. A pseudo R*2 measure18 was 0.677 for the latlong approach and 0.674 for the xlag approach (0.671 without lags). Another assessment of predictive power is the number of locations where land use is predicted correctly. Both approaches had essentially the same result overall; 87.2% for the latlong approach and 87.3% for the xlag approach. However, they differed somewhat in predicting individual categories. The xlag approach was marginally better than the latlong approach for all categories except cativo forest where it predicted only 6.8% of these locations correctly, compared to 9% for the latlong approach. These results suggest that land use may be affected in different ways by the different spatial processes, but we have not explored this possibility further. For the remainder of the paper we report xlag results. Table 4 reports the estimated coef® cients and standard errors with instrumental variables. It is useful to examine the coef® cients for property rights dummies (Cemaco, Sambu, and the park) on the human intervention land use. These can be interpreted as the effect of a one unit change in the

property rights dummy onln (Probhumanintervention/ Probforest without cuipo). The Sambu coef® cient is positive but very small and probably does not differ signi® cantly from zero. Both the park and Cemaco coef® cients are negative, and the Cemaco coef® cient is one of the larger human intervention coef® cients in absolute terms. These coef® cients suggest that the existence of the park and Cemaco reserve reduce the likelihood of conversion of forest without cuipo to human intervention. The remaining Cemaco coef® cients are either very small or ® xed at zero, implying that the addition of the Cemaco reserve does not change the ln odds of conversion to those land uses from forest without cuipo. Table 5 presents the prediction matrix for the regression. The predictive power is high (65% of locations or more predicted correctly) for four of the ® ve categoriesÐ forest with and without cuipo and human intervention. The human intervention category was pre18 R*2 5 1- ln L/ ln L0 (see, e.g., Greene 1993, 651), where L is the value of the likelihood function and L0 is the value of the likelihood function with coef® cients constrained to 0.

Land Economics

200

May 2001

TABLE 5 P rediction matrix, Five Land Use Categories, 1997 data (sq. km) Ratio, correct Actual

0

1

2

3

4

Total

to total

Forest without cuipo (0) Forest with cuipo (1) Forest with cativo (2) Human intervention (3) Marsh (4)

16,805 1,000 0 130 0

1,499 22,604 483 748 228

0 11 62 35 0

51 2,546 272 14,631 632

14 187 93 195 1,668

18,369 26,348 910 15,739 2,528

0.915 0.858 0.068 0.930 0.660

Total

17,935

25,562

108

18,132

2,157

63,894

Note: Diagonal (bold) cells indicate correct predictions.

dicted correctly 93% of the time.19 Prediction of the cativo forest locations is especially weak; the correct land use was predicted only 7% of the time. About two-thirds of the cativo forest locations were predicted to be in forest without cuipo, with the human intervention category about one-® fth. This result may be due to disequilibrium in cativo forest area. Since cativo is a valuable species, it may be that remaining areas simply have not yet been reached by harvesters and the misidenti® cation in fact re¯ ects a likely future outcome. Pixels that lie on the border between land use categories are frequently misclassi® ed. This can be seen most clearly in Figure 3 in combination with Figure 2. Forexample, in the southeast part of Figure 3, the border between the forest with and without cuipo is clearly delineated by misclassi® ed pixels. Since the transitionfromone type of foresttoanotheris likely to be gradual, this result is to be expected. Prediction power maps. Figure 3 presents maps and histograms for Pmax, the highest probability value for any land use and Pdif, the difference between Pmax and the probability value for the actual land use. As can be seen in the top half of the ® gure, most locations are predicted correctly (60,239) and the remaining locations are distributed throughout the province. There appear to be a few concentrations of misclassi® ed predictions. As mentioned earlier, the borders of land uses, especially at locations that lie between cuipo and non-cuipo forests, are often misclassi® ed. At these locations, Pmax values are small, suggesting that the strength of prediction is weak. This can also be seen in the bottom

half where the borders between land uses often have lower maximum probability values. In addition there are a few clumps of mispredicted locations (park patches in the upper image in Figure 3). These areas should be further scrutinized as the project advances. Possible explanations include misclassi® cation in the original land use data sets and areas that are about to change land use. The lower half of Figure 3 maps Pmax. For most (59,416) locations, the probability values are in the top quintile. There are some interesting deviations. As discussed above, the borders of forest land use categories have lower Pmax values, as do the agricultural areas along the rivers in the eastern part of the province. Simulating Changes in Property Rights

We simulate three types of changes in property rightsÐ elimination of legal protection to the national park, and removal of special status for the Cemaco and Sambu reserves. The simulation is done by setting the location dummy variable to 0. Table 6 presents the transition matrices for each simulation. Figure 4 shows the geographic distribution of those changes. 19 This is in sharp contrast to the results when the estimation was done with disaggregated human intervention categories including annual agriculture, pasture, and brush (Nelson, Harris, and Stone 1999). In that study, the agriculture, pasture and brush categories were predicted correctly only 44%, 84%, and 29% of the time. We surmise that the land use classi® cation process was not suf® ciently precise to distinguish among these human intervention categories adequately.

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

201

FIGURE 3 Spatial Distribution and Histograms for Measures of P redictive P ower, Pdif and Pmax

Land Economics

202

May 2001

TABLE 6 Transition Matrices for Land Use Simulations, 1997 (sq. km) Forest, without cuipo (0)

Forest, with cuipo (1)

Forest, with cativo (2)

Human intervention (3)

Marsh (4)

Total

17,935 0 0 0 0 17,935

16 25,445 0 101 0 25,562

0 0 108 0 0 108

0 0 0 18,132 0 18,132

0 0 0 0 2,157 2,157

17,951 25,445 108 18,233 2,157 63,894

Predicted to no Cemaco forest, without cuipo Forest, with cuipo Forest, with cativo Human intervention Marsh Total changed

4,484 0 0 0 0 0

6 6,163 0 222 0 228

0 0 13 14 0 14

0 0 0 4,533 0 0

0 0 0 2 537 2

4,489 6,163 13 4,771 537 244

Predicted to no Sambu forest, without cuipo Forest, with cuipo Forest, with cativo Human intervention Marsh Total changed

4,484 0 0 0 0 0

22 6,344 0 24 1 46

0 0 27 0 0 0

0 0 0 4,533 0 0

0 0 0 0 539 0

4,506 6,344 27 4,557 540 47

Predicted to no park forest, without cuipo Forest, with cuipo Forest, with cativo Human intervention Marsh Total changed

Note: Columns are predicted land use categories with base data. Rows are land use categories with simulated changes in property rights. Entries in a cell indicate the number of sq. km with column land use in 1997 and row land use in the simulation. For example, in the simulation with no Cemaco reserve, 222 square kilometers of the forest with cuipo category is converted to the human intervention category.

Removing the park. Simulating the elimination of the legal protection to the park has relatively little effect on land use there. Land use changes on only 29 sq. km out of a total of 5,453 sq. km, or less than one percent of the park area. These changes all occur in the forest without cuipo category with most of the changed area moving to the human intervention category. The small effect of this simulated change in legal status is not surprising. The park is rugged and remote and the cost of economic activity there is high. Although the effective protection provided by the park’ s legal status is uncertain, geography has provided an effective barrier to increased exploitation.20 Removing the Cemaco Reserve. It is obvious from a land use map that land use in the Cemaco reserve differs from the rest of the

province. Project staff also report that the inhabitants have been able to use the reserve status to exclude migrants from other parts of Panama. Removing this protection causes the largest change of any of the simulations. Land use change occurs on 244 sq. km, or 8.4% of the reserve’ s 2,901 sq. km. The largest change is to forest with cuipo where 228 sq. km are lost and 222 of these move to the human activity land use. Cuipo forest is also lost to human intervention. Examination 20 This result holds only for the relatively coarse level of land use categories available. It is possible that ``high-grading,’ ’ the selective removal of valuable species such as mahogany and cedar, has gone on. We cannot detect that kind of activity in this analysis. In addition, civil con¯ ict in northern Colombia has sometimes spilled over into the park and may be restricting economic activity there.

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

203

FIGURE 4 Simulation of Land Tenure Changes

of the lower left ® gure in Figure 4 suggests that for the most part these changes take place at the borders between existing land uses. The ® nger-like areas running northeast to southwest are rivers that drain the mountains that border the province on the east. Land use along the river edges is principally agriculture. When the reserve is eliminated, agriculture expands into the surrounding forest. Removing the Sambu Reserve. The elimination of the Sambu reserve has less absolute but greater relative effect on land use than elimination of the Cemaco reserve. Land use changes on 47 sq. km, or 9.2% of the 510 sq. km in the reserve. Of that change 24 sq. km is towards more human intervention and almost all of the rest to forest without cuipo.

V. CONCLUSIONS The simulations presented above suggest that legal protection of the national park seems to have made little difference to land use within the park. Its dif® cult terrain and remoteness make commercial economic activity unpro® table, at least as indicated by our land use categories. On the other hand, our analysis suggests that cultural land use practices, combined with effective property rights, clearly make a difference in the resulting landscape. The inhabitants of the Cemaco reserve use their land resources differently than those on the other side of the Chucunaque River. The simulations indicate that a signi® cant loss of forest cover would result if effective property rights

Land Economics

204

were removed. We cannot say de® nitively how much of this land use choice is due to the property rights regime per se rather than a different ethic about land use. On the other hand, the inhabitants of the Sambu reserve do not seem to be using cultivation practices that differ from those outside the reserve. There are two possible explanations for the difference in importance of legal property rights in the two reserves. The construction of the Pan American highway in the early 1980s opened up areas near the Cemaco reserve to easy access by immigrants from other parts of Panama. The legal property rights created by the reserve designation gave the indigenous population another tool to prevent encroachment from outsiders. The Sambu reserve, on the other hand, is far from the nearest primary road. It is also probably the case that the population outside the Sambu reserve has cultural values, and therefore cultivation practices, that are similar to those inside the reserve. Hence, the need for, and the effect of, effective property rights is less signi® cant for the Sambu reserve than the Cemaco reserve. In summary, then, we can say that while property rights exert a signi® cant impact in land use and forest cover in some parts of the DarieÂn, our research indicates that the causality is not universal and depends on location. Furthermore, cultural values are also likely a factor and are correlated to some extent with location. These ® ndings, coupled with advances in spatial econometric modeling, open the door for new research in other frontier areas and should inform the ongoing debate about land use change in the tropics. VI. THEORETICAL APPENDIX We assume that a Cobb-Douglas relationship governs all production technologies. Qh 5 Gl

p

X ak 0 , ak , 1;

[7]

^a ,1 k

k

S

Gl 5

1p 2 G

s51

Let b 5 1 2

^a

fs s

Ð

a multiplicative combination of location-specific geophysical features that affect land use.

k

k

The indirect net present value function is:21 RhlT 5

#

13

¥

t50

bh PhlT1tGl

p

a

C

1 bh

4 2e

2akh akh khlT1t kh

k

2il t

dt.

[8]

Because we have no time-series data on land use, we have no time variability in prices. Then [8] can be rewritten as:

3 p 5 b 3P G p C

RhlT 5 bh Phl Gl

2akh khl

C

a

h

hl

2akh khl

l

a

1 bh

4#e 4 11i 2.

akh kh

k

akh kh

¥

2il t

dt

t50

1 bh

[9]

l

k

Taking the log of [9] gives: ln RhlT 5 ln bh 1 1

^(2a

kh

3

1 ln Phl 1 ln G bh

4

ln Ckhl 1 a kh ln a hk) 2 ln il. [10]

k

Since we have no data on either location-speci® c or ® nal destination prices, we proxy prices by measures of cost-of-access to several ® nal destinations. We assume that location-speci® c price proxies take the following functional form: Phl 5 exp[g01 1 g11Dl] Chl 5 exp[d01 1 d11Dl]

[11]

Dl ± cost of access measure from ® nal destination (or source of input) to location l. Note that this form assumes that the price proxies for all inputs (Ckhl) are the same. We make this assumption because we do not have input-speci® c data. Substituting the price proxies into [10] and performing some additional manipulations yields: ln RhlT 5 h0h 1

^h i

k

0,

May 2001

Dl 1

1ih

^h

Gr

2hr

r

1 h3h ln il 1 uhl g 1 h0h ; ln bh 1 0l 1 (d01)(bh 2 1) bh bh 1 1 (akh ln akh) bh k

^

21

See Beattie and Taylor 1993, 248.

77(2)

Nelson, Harris, and Stone: Deforestation, Land Use, and Property Rights

1 [gil 1 (dil)(bh 2 1)] bh 1 hr h2h ; bh h3h ; 21.

h1ih ;

[12]

We have the following hypotheses about signs of the structural coef® cients: 0 , ak , 1;

^ a , 1; 1b 5 1 2 ^ a 2; k

k

k

k

from production function gh , 0; Location-specific output price declines with distance to final destination. dh . 0; Location-specific input price increases with distance from input source. These hypotheses imply h0 and the h1 vector are negative, h2 to be positive for G ordered so that an increase in G raises output per unit of land, and h3 to be negative (and equal to negative 1). Parcel h will be devoted to land use k if RhkT . RhlT, " l ¹ k. If the u are Weibull distributed and uncorrelated across land uses, then [12] is equivalent to a multinomial logit model where: Probhl 5

eV1hh

^

.

[13]

eV1hj

j

References Anselin, Luc. 1988. Spatial Econometrics: Methods and Models, Studies in Operational Regional Science. Dordrecht, The Netherlands: Kluwer Academic Publishers. Beattie, Bruce R., and C. Robert Taylor. 1993. The Economics of Production. Malabar: Krieger Publishing Company. Bockstael, Nancy E., and Kathleen Bell. 1998. ``Land Use Patterns and Water Quality: The Effect of Differential Land Management Controls.’ ’ In Con¯ ict and Cooperation on TransBoundary Water Resources, ed. R. Just and S. Netanyahu. Norwell, Mass.: Kluwer Academic Publishers. Bush, Mark B., and Paul A. Colinvaus. 1994. ``Tropical Forest Disturbance: Paleoecological Records from DarieÂn, Panama.’ ’ Ecology 75: 1761± 68.

205

Chapin, M. 1991. ``Losing the Way of the Great Father.’ ’ New Scientist 131 (1781):40± 44. Chomitz, Kenneth M., and David A. Gray. 1996. ``Roads, Land, Markets and Deforestation: A Spatial Model of Land Use in Belize.’ ’ World Bank Economic Review 10 (3): 487± 512. D’ Arcy, W. G., and M. D. Correa-A. 1985. The Botany and Natural History of Panama: la botaÂnica e historia natural de PanamaÂ. St. Louis: Missouri Botanical Garden. Deininger, Klaus, and Barton Minten. 1999. ``Poverty, Politics, and Deforestation: The Case of Mexico.’ ’ Economic Development and Cultural Change 47 (2): 313± 44. Godoy, Ricardo, Marc Jacobson, Joel De Castro, Vianca Aliaga, Julio Romero, and Allison Davis. 1998. ``The Role of Tenure Security and Private Time Preference in Neotropical Deforestions.’ ’ Land Economics 74 (May): 162± 70. Greene, William H. 1993. Econometric Analysis. 2nd ed. New York: Macmillan Publishing Company. Kaimowitz, David, and Arild Angelsen. 1998. Economic Models of Tropical Deforestation: A Review. Bogor, Indonesia: Center for International Forestry Research. Nelson, Gerald C., and David Gray. 1997. ``The Use of Spatial Analysis at the World Bank.’ ’ Environment Department Paper 56. Washington, D.C.: The World Bank. Nelson, Gerald C., and Daniel Hellerstein. 1997. ``Do Roads Cause Deforestation? Using Satellite Images in Econometric Analysis of Land Use.’ ’ American Journal of Agricultural Economics 79 (2): 80± 88. Nelson, Gerald C., Virginia Harris, and Steven W. Stone. 1999. ``Sustainable Development in Panama’ s Darien Province: Modeling Land Use Change with Spatial Econometric Analysis.’ ’ Washington, D.C.: Sustainable Development Department, Environment Division, Inter-American Development Bank. Richards, John A. 1993. Remote Sensing Digital Image Analysis: An Introduction. 2nd ed. Berlin: Springer-Verlag. Simmons, Cynthia. 1997. ``Forest Management Practices in the Bayano Region of Panama: Cultural Variations.’ ’ World Development 26 (6): 989± 1000. Stone, Steven W. 1998. ``Using a Geographic Information System for Applied Policy Analysis: the Case of Logging in the Eastern Amazon.’ ’ Ecological Economics 27 (1): 43.

Spatial Variability and Disincentives to Harvest: Deforestation and Fuelwood Collection in South Asia Gunnar KoÈhlin and Peter J. Parks ABSTRACT. A major strategy to combat deforestation caused by household fuel collection has been the establishment of plantations, especially in India. A household model is speci® ed with a number of collection possibilities and analyzed empirically using household, vegetation, and GIS data, and the potential decrease in collection from the natural forest is estimated. The results show reduced pressure on the natural forest due to the establishment of plantations. It also questions buffer zone plantations very close to natural forests. (JEL Q23)

Regional and country-level trends in deforestation result from decisions made at smaller spatial scales. At the household level, decisions that lead to deforestation can be considered land use decisions. Recent advances in spatial modeling of land use decisions have had much success after including location of land use choices in analyses. Interest in, and examples of spatial models applied to tropical forests is growing. Improved regionaland country-level results can be obtained by aggregating smaller analytical units: a spatial model must start with tracts of land whose location is known. For example, one of the most important reasons for long-term deforestation is the fuel collection decision made by millions of households every day. In the 1970s, dependence on tropical natural forests as a source of fuel led to simplistic models and projections of the growing gap between deforestation and reforestation (so called wood balance or gap models). Such models typically took into consideration population and income growth, but not the possibilities for substitute fuels from planted forests. Planted forests have been used as a major strategy to combat deforestation, especially in India, where the government spent 35 billion RuLand Economics · May 2001 · 77 (2): 206± 218 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

pees during the 1980s to afforest 13 million hectares. Unfortunately, some plantation projects were badly implemented. Poor results decreased interest in the potential for planted forests to reduce fuelwood collection in natural forests. However, failing to consider fuelwood collection from natural forests in policy design is a serious mistake that could limit or even prevent the success of policies to combat deforestation in many parts of the developing world. The majority of households in developing countries depend on biomass for cooking and heat. Scarcity of biomass is likely to lead to continued environmental degradation and ever increasing efforts to ® nd fuel, perpetuating the vicious cycle of deforestation for fuel. Although studies of household fuelwood collection are increasing in number (e.g., Mercer 1991; Amacher, Hyde, and Joshee 1993; Bluffstone 1995; Cooke 1998; and Mekonnen 1999), the potential for plantations to reduce deforestation from fuelwood collection in many developing regions remains unknown. For example, in the Indian state of Orissa, the Orissa Social Forestry Project (OSFP) was established with external support from Sweden. The OSFP was intended to develop a self-reliant and replicable system of forestry that could be applied in the villages of Orissa, and which would also eventually reduce the pressure on government forests. Unfortunately, it has not yet been possible to The authors are, respectively, assistant professor at the Environmental Economics Unit, Department of Economics, GoÈteborg University, and associate professor at Department of Agricultural, Food, and Resource Economics, Cook College, Rutgers University. The research was supported by the Swedish International Development Cooperation Agency (Sida). The paper has bene® tted from comments by Thomas Sterner, Fredrik Carlsson, Alemu Mekonnen, Priscilla Cooke, Ramon Lopez and two anonymous reviewers.

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

show whether this goal has been met. In evaluations of the OSFP it is alleged that the major environmental bene® t from the project is decreased pressure on natural forests as the result of the establishment of 100,000 ha of community plantations. However, no quantitative evidence has been developed. In contrast, there are those within the Orissa Forest Department who argue that the secure user rights of village woodlots (VWL) may make villagers stay away from the woodlots, that were established (under the OSFP) on land that had earlier provided at least some fuel. If this is true, then the result of establishing these VWL has been increased pressure on government forests, which are exploited under de facto open access. These con¯ icting claims regarding the ability of planted forests to relieve pressure on natural forests can only be resolved through empirical analysis. Despite the great global interest in deforestation over the last decade, not much has been done to analyze the impact of plantations to reduce deforestation, or on household modeling to explain deforestation. In a recent review of the literature on tropical forest land use, Parks, Barbier, and Burgess (1998) describe only a few studies that focus on the agents of forest land clearing for agriculture and livestock, and that take household characteristics into account. The work of LoÂpez on agricultural households in CoÃte d’ Ivoire and Ghana (LoÂpez 1986) is one example. Other comprehensive reviews of the deforestation literature (Lambin 1994; Kaimowitz and Angelsen 1997) give the same result: there are numerous cross-sectional country and province studies, but fewer studies that use household data. This paper applies a spatial perspective to evaluate deforestation pressures on natural forests and the potential ability of planted forests to relieve these pressures. The policy question considered is: Have plantations reduced the pressure on natural forests? In order to answer this question the household collection decision must be carefully modeled and empirically estimated. Our strategy is to start from a spatial household model where we specify a number of collection possibilities. These are empirically ana-

207

lyzed using data from Orissa. The collection functions include accessibility of different sources of biomass (location) and household characteristics. Shadow wages for different household categories collecting fuel from different sources of biomass are estimated. These wages are then used to estimate the potential decrease in collection from the natural forest, using a time allocation function for collection in forests at different locations. Community plantations have been found to provide substantial welfare gains by saving time spent in fuel collection (KoÈhlin 1998, Chapter 2). These welfare effects depend on spatial considerations. A comparison of villages with and without VWL indicates higher consumption of fuel in the villages with VWL (KoÈhlin 1998, Chapter 3). The results from this paper show a signi® cant reduction in pressure on the natural forest due to establishment of VWL. The location of the VWL must be considered also in this case, since it must provide a reasonable fuel substitute for fuel from natural forests. The reduction varies between villages and ranges from 0 to 29% of current collection in natural forests (NF). Analysis of the decision to collect shows that the availability and location of VWL in¯ uences the probability of collection in NF. The combined effect is evidence that the establishment of plantations has reduced the pressure on the natural forest, but this effect is crucially dependent on location. There is spatial variation in the different bene® ts from plantations (time saved, fuel consumed, and decreased extraction from natural forests). The decision to collect in the natural forest is affected by shadow wages from collection in plantation. The shadow wages, in turn, are affected by the distances to planted and natural forests. Plantations can decrease fuelwood collection in natural forests, but this potential depends on distance. Decreased collection in natural forests depends on the location of the planted forest: the effect is non-linear with an inverted Ushape that peaks when planted forests are located around three kilometers from the natural forest. This location is where the household collection decision is most sensitive to changes in relative shadow wages. These ® ndings argue that the optimal loca-

Land Economics

208

tion of plantations depends on the objective of the plantation, but that the popular concept of buffer zone plantations could be less ef® cient in achieving the environmental objective of decreasing pressure on the forest than plantations a couple of kilometers away from the forest. The paper concludes with a summary of such policy recommendations. I. A HOUSEHOLD MODEL OF FUELWOOD COLLECTION

[1]

Fuel consumption, C, is derived from natural forests or village woodlots. Fuel (Fi ) can be collected either in natural forest or VWL (i 5 N and V, respectively). Collection is a function of time input1 (Ti ), household composition (A) and resource availability (Ri ). For collection in the natural forest, physical capital (K) in the form of ox and cart can be used. We thus have the following household collection functions: F i 5 fi (Ti , K, A, Ri ).

The decision on whether to collect, sell, or buy fuel also depends on a household monetary constraint where the purchase of commodities (at price px) and rental cost of a cart (pK K) must be balanced by the income from fuel sold (the net of fuel production and consumption at price pF ), wage labor (Tw at wage w) and other non-fuel, non-labor income2 (Y), p XX 1 p K K # p F (å i Fi 2 C) 1 wTW 1 Y.

The purpose of this model is to explain fuel collection behavior, that is, the choice of collection/consumption of biomass from natural forests, VWL, and the market. The model assumes that households make decisions to maximize utility. Utility is nonseparable due to large subsistence production which makes production and consumption decisions interdependent. This interdependence is a common feature of studies in developing countries, including applications to agriculture (e.g. LoÂpez 1986; Strauss 1986; Jacoby 1993; Skou® as 1994) and fuel collection (e.g. Amacher, Hyde, and Joshee 1993; Mercer 1991; Cooke 1998; Mekonnen 1999). The model features a well behaved utility function with fuel consumption (C) as a separate argument together with consumption of a composite purchased good other than fuel (X), leisure (TL) and household characteristics, A, acting as taste shifters, U 5 u(C, X, TL; A).

May 2001

[2]

These collection functions could also be referred to as production functions for fuelwood.

[3]

Note that this is not the commonly used fullincome speci® cation which would also include the value of household labor used for subsistence production and leisure. Instead we use an explicit time constraint where total time (TT) is the sum of time spent collecting in NF and VWL, time spent working for wages, and leisure, å i Ti 1 TW 1 TL 5 TT.

[4]

Consumption, capital, and time are nonnegative, so that C, X, K, Ti , TW, TL $ 0.

[5]

The Lagrangian for the household’ s3 utility maximization problem with respect to Fi , C, X, K, Ti , TW, and TL is: L 5 u(C, X, TL; A) 1 l[ p F (å i Fi 2 C) 1 wTW 1 Y 2 p XX 2 p K K] 1 m[TT 2 å i Ti 2 TW 2 TL] 2 n[FV 2 fV(TV, K, A, RV)] 2 j[FN 2 fN(TN, K, A, RN)],

[6]

1 Time input from different household member categories (men, women, boys, girls) are used in the estimation procedure, but the analysis is carried out at the household level. 2 The most important sector, agriculture, is only included this way to keep the model simple and in line with the empirical analysis to come. Admittedly, a full model featuring participation decisions in wage labor, agriculture, and collection would be preferable. 3 With collection aggregated to the household level. Permitting a free range of nonnegative choice for TW implies a perfect labor market. A logical extension of this work might be to explore speci® c imperfections. We thank an anonymous reviewer for making this observation.

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

where l, m, n, j are the Lagrangian multipliers of the income, time, and fuelwood collection constraints, respectively. The optimum amounts of fuelwood to collect in VWL and NF are given by ¶L/¶FV 5 lp F 2 n # 0, FV $ 0, (¶L/¶FV)F V 5 0,

[7]

and ¶L/¶FN 5 lp F 2 j # 0, F N $ 0, (¶L/¶F N)FN 5 0,

[8]

respectively. Fuelwood will be collected from the VWL provided that marginal utility of added income from fuel, lp F , equals the marginal utility cost of resources required to collect fuel from the VWL, n. Similarly, fuelwood will be collected from the NF provided that marginal utility of added income from fuel, lp F , equals the marginal utility cost of resources required to collect fuel from the NF, j. Optimum allocations of time for VWL fuel collection, NF fuel collection, wage labor, and leisure are given by ¶L/¶TV 5 2m 1 n¶fV/¶TV # 0, TV $ 0, (¶L/¶TV)TV 5 0,

[9]

¶L/¶TN 5 2m 1 j¶fN /¶TN # 0, TN $ 0, (¶L/¶TN)TN 5 0,

[10]

¶L/¶ TW 5 lw 2 m # 0, TW $ 0, (¶L/¶TW)TW 5 0,

[11]

and ¶L/¶TL 5 ¶u/¶TL 2 m # 0, TL $ 0, (¶L/¶TL)TL 5 0,

[12]

respectively. Time will be allocated to VWL (NF) fuel collection provided that the marginal utility value of added fuel production, n¶fV/ ¶TV (j¶fN/¶TN), equals the marginal opportunity cost of time, m. Similarly, time will be allocated to wage labor provided that the marginal utility of added income, lw, equals the marginal opportunity cost of time, m. Finally, time will be allocated to leisure until the mar-

209

ginal utility of leisure time, ¶u/¶TL, equals the marginal opportunity cost of time, m. From these conditions, we can summarize when households (1) will collect in both VWL and NF; (2) will not collect in the VWL; (3) will not collect in the NF; and ® nally, (4) will not collect at all. The discussion can be facilitated by interpreting the multipliers l, n, and j, as the marginal utility of monetary income,4 shadow price of fuel collected from the VWL, and the shadow price of fuel collected from the NF, respectively. The economic conditions under which collection from the NF is reduced are the focus of our empirical work. Collection in Both VWL and NF

When households optimally collect from both VWL and NF, the marginal products from collection in VWL and NF would be the same for the households engaged in both activities. For those who have access to a market the collection is ``capped’ ’ by the price of fuel: p F 5 n/l 5 j/ l. No Collection in VWL (But Collection in NF and Possibly Market Purchase)

With no optimum collection in VWL, FV 5 0, suggesting that ¶L/¶FV , 0. If ¶L/¶FV , 0, then the shadow price of VWL collection, n, exceeds the marginal utility of added income from fuel, lpF . When this is true the household could be expected to purchase fuel. However, since the shadow prices for the VWL and NF are not necessarily the same, (n could not in general be expected to equal j), then the household might still collect in the NF when: p F 5 n/l , j/l. No Collection in NF (But Collection in VWL and Possibly Market Purchase)

When households do not collect in the NF the optimum is FN 5 0. Since FN 5 0 implies 4 A common assumption is that l 5 1. It gives intuitive interpretations easily tested in empirical analysis, but is de® nitely a simpli® cation. For an empirical estimation of the marginal utility of money for different income groups in India see e.g., Sharma, McGregor, and Blyth (1991).

210

Land Economics

May 2001

that TN 5 0, the ® rst order conditions for collection times TN and TV (see [9] and [10] above) show that n¶fV/¶TV . j¶fN/¶TN. Under these conditions, collection in VWL brings more utility than collection in NF. Whether this is enough to actually spur collection by a speci® c household depends on whether the price of fuel is greater than the shadow price of fuel collection/marginal utility of cash income (see above).

TN 5 t( f N¢ , f V¢ , w, Y, A, R)

No Collection

DF N 5 DTN f ¢N.

There is a possibility that there will be no collection at all. The obvious case is when the opportunity cost of time is too high, for example, because of wage labor. Reorganizing the optimum allocations of time to VWL, NF, and wage labor suggests that optimum TV 5 TN 5 0 when: w . (n¶fV/¶TV)/l and w . (j¶fN/¶TN)/l.

Household time spent collecting in the NF and the resulting changes in collection in the NF will both be analyzed using household data collected from villages in Orissa.

II. EMPIRICAL ANALYSIS OF FUEL COLLECTION FROM NATURAL FORESTS In each of these cases the decisive marginal returns from collection depend on household characteristics and resource accessibility, A and Ri. These characteristics differ between households and villages, respectively. KoÈhlin (1998, Chapter 3) found statistically signi® cant differences among VWL and NF collection when households and villages were strati® ed by caste, gender, and other categories. Our empirical focus is on the determinants of time allocated to NF fuel collection, and how the introduction of VWLs can decrease pressure on natural forests. From the household model and the discussion of the Kuhn-Tucker conditions we can conclude a reduced form labor supply function for collection in NF.5 Household time allocated to NF collection, TN, can be expected to depend on the marginal products (shadow wages) from collection in NF and VWL (to conserve notation, ¶fV/¶TV and ¶fN/¶TN for a household will be written below as f¢V and f¢N), market wages, w, exogenous income, Y, household characteristics, A, and resource availability, R,

[13]

The change in TN for a particular household can then be estimated using the marginal effect of f¢V on TN simply: DTN 5 (¶TN/¶f ¢V)D f ¢V.

[14]

If we assume that the marginal product f¢N is constant in this interval, then the change in collection in the natural forest, FN, is [15]

Data

The data were collected in Orissa in February 1995. The data set contains 742 household observations based on personal interviews by professional enumerators. The households were visited in 22 randomly selected villages in the vicinity of a common forest, the Dhani Reserve Forest. The data set features detailed information regarding collection in both VWL and NF with time spent per trip, quantity collected per trip, and number of trips over the last year. In the following analysis the modern fuels have been excluded and the different biomass fuels have been converted to ``leaf equivalents’ ’ (KoÈhlin 1998, Chapter 2). The sources of fuel have been aggregated to natural forest (NF) and village woodlot (VWL). Estimation of Time Spent Collecting Fuel in the Natural Forest

In order to identify the potential decreased pressure on NF we need to estimate an empirical representation of the reduced form time allocation function, t(f¢Nh , f¢Vh , wh, Yh , Ah , R). The analysis is carried out at the household level. Because less than half (279 of 742) of the households collect in the NF, a Heckman speci® cation with sample selection 5 The authors are indebted to Professor Ramon LoÂpez for this insight.

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

211

TABLE 1 Fuelwood Collection in Natural Forests by Households in 22 Randomly Selected Villages in Orissa, India, 1995 Variable Probability of collection in natural forest Time spent collecting in natural forest (hours per year) Age of head of household (years) Family size Men older than 15 as share of family Number of schedule caste/tribe women in household Schedule caste/tribe dummy High caste (Brahmin/Khandayat) dummy Total income including agricultural produce (rupees) Distance to forest (km) Size of Informal Protection Committee (IPC) area (acres per household) Informal Protection Committee (IPC) age (years of protection) Size of village forest (acres per household) Size of village woodlot (ha per household) Household marginal product in natural forest (kg per hour) Household marginal product in village woodlot (kg per hour) Natural forest wage 2 village woodlot wage (rupees per hour) Natural forest wage 2 market wage (rupees per hour)

is used. In this two-step procedure, the discrete decision of whether to collect in the NF is ® rst analyzed using a probit model, then the continuous decision of time to spend collecting is analyzed using a sample selection procedure. Descriptive statistics for the variables used are presented in Table 1. The age of the head of household is expected to reduce the probability of collection because old households have less mobility. Family size is expected to be a major determinant of household fuel needs and is therefore expected to increase both probability to collect and the time allocated to collecting. Preliminary analyses indicate that there are two groups in the household that are positively correlated to collection in NF, women in scheduled caste or scheduled tribe (SCST, or low caste) households and men in general. These are included as share of men in the household and an interaction term based on number of women in the household and a dummy for SCST households. Dummies for low and

Mean

Standard Deviation

0.38 484 50 5.9 0.35 0.29 0.18 0.44 15,850 2.75 4

0.48 1,271 14 2.7 0.17 0.72 0.38 0.5 18,300 1.78 11

Range 0± 3± 12± 1± 0± 0± 0± 0± 2400± 0± 0±

1 11620 90 24 1 5 1 1 252,000 6 86

3.3

5.5

0± 15

0.2

0.4

0± 1.8

3.9

8.1

0± 63

3.4

5.8

0± 61

20.2

6.8

254± 61

22.4

8.8

285± 49

high castes are also included in both steps, with ``general caste’ ’ as the reference point. Income is included and is expected to decrease collection since market purchase and modern fuels become more interesting for households with a high opportunity cost of time and a lower marginal utility of money. The income variable used is a composite of all monetary incomes from wage labor, sale of agricultural products, livestock and other products. To this has been added an estimation of the value of subsistence production by taking the quantity of the staple rice produced times its market value. A ® nal modi® cation is the inclusion of net transfers to and from the household. A number of resource variables have been included. The size of potential village forest per household is expected to be positively related to the decision to collect. Time spent collecting in village forests are included in the second stage of the estimation. The size of jurisdiction and age of existing informal protection committees (IPCs) for the forest is

Land Economics

212

TABLE 2 P robit Analysis of Households’ Decision to Collect in Natural Forest, Orissa, India, 1995 Variable Constant Age of head of household Family size Share of men in household Number of SCST women in household Low caste High caste Total income (1026) IPC area per household Age of IPC Village forest area per household NF wage 2 VWL wage NF wage 2 market wage

Marginal Effect

t-Value

20.29*** 20.002 0.02** 0.18 0.15***

23.26 21.10 2.10 1.42 2.62

0.07 0.08* 20.85 20.006** 0.014*** 0.07

0.60 1.70 20.60 22.10 2.53 1.52

0.027*** 0.01***

6.38 4.31

* indicates signi® cance at the 10% level; ** indicates signi® cance at the 5% level; *** indicates signi® cance at the 1% level.

also included. A village forest is a (remnant of a) natural forest close by the village. Not all villages have such nearby forest areas and their utilization is probably informally restricted. IPCs are organized to better manage natural forest areas. They typically imply both restriction of the forest area to ``outsiders’ ’ and regulate the use of the forest to the eligible villagers. This is why it is expected to affect forest utilization over and above its impact on marginal product. It has been shown that IPCs in the area have improved the quality of the forest. The results in Table 2 imply that as time goes by the restrictions on the members decrease which enables them to spend more time collecting in the natural forest. Finally, two variables emanate from the household’ s optimum decisions. The decision to collect should depend on the relative prices of different sources of fuel, such as NF, VWL, and the market. The relevant prices are the opportunity costs of time. We have therefore included the difference between the shadow wages for fuel from NF and VWL [NF wage 2 VWL wage] as well as the difference between the shadow wages

May 2001

for fuel from NF and wage labor [NF wage 2 market wage]. The shadow wages are calculated using the marginal products6 multiplied by the local market price for fuel. The market wage is based on reported wages and labor days. All wages are expressed in rupees per hour. In the second step of our Heckman speci® cation, the dependent variable is the natural logarithm of the household time spent collecting in the NF. Explanatory variables include marginal products from collection in NF and VWL, as well as selected household characteristics. Endogeneity stems from the fact that the time allocation decision and the marginal product for collection in NF are not independent. Because the marginal product for collection in NF is endogenous, the model is estimated using an instrumental variable for f¢N. The probit equation is estimated ® rst, then f¢N is regressed on a number of instruments7. The ® tted values are kept for use in the second step estimation. The original variable is used when computing the estimate of the disturbance variance (Greene 1995). The results from the probit step are presented in Table 2. Signi® cant variables have the expected signs. Worth noting is the signi® cance of household variables such as age of head of household and number of SCST women in the household. This is an empirical 6 Six separate collection functions were estimated for men’ s collection in NF, women’ s collection in NF, men’ s collection in VWL, women’ s collection in VWL, boys (5± 15) collection in VWL and, ® nally, girls (5± 15) collection in VWL (KoÈhlin 1998, Chapter 2). The estimation used a sample selection procedure and maximum likelihood estimator. The estimations enabled us to calculate the marginal product of collection in the natural forest and in village woodlots by multiplying labor input elasticities, by the ratio of predicted quantity produced to time spent (Mekonnen 1998; Jacoby 1993; Skou® as 1994). Here we use the marginal products aggregated to the household level in the form of weighted averages for collection in NF and VWL separately. 7 The instruments used were number of women, family size, sex of head of household, average education of family, total income, two caste dummies, distance to forest, size of VWL, village forest and IPC per household, a vegetation index for a radius of 1.5 km for each village from the GIS, use of cart, sale of collected fuels, and if the village has electricity. The adjusted R2 for the equation was 0.73.

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

213

TABLE 3 The Determinants of Household Time for Collection in Natural Forests, Orissa, India, 1995 Variable

Coef® cient

t-Value

Constant Marginal product in village woodlot Marginal product in natural forest Family size Number of SCST women Low caste High caste Village forest area per household Inverse Mill’ s Ratio

4.35*** 20.052*** 20.031** 0.09*** 0.35** 1.00*** 0.12 0.80*** 1.10***

8.08 23.40 22.27 3.16 2.10 2.69 0.66 4.63 2.94

Marginal Effect

t-Value

0.054* 0.62 0.85** 20.02 0.67***

1.76 0.32 2.08 20.10 3.58

Note: The results are from the second step of Heckman two-stage estimation. * indicates signi® cance at the 10% level; ** indicates signi® cance at the 5% level; *** indicates signi® cance at the 1% level.

reminder to be cautious of uniform approaches to extension community forestry programs: Different households have very diverse strategies to meet their basic needs. The relative wage measures have very strong explanatory power. The marginal effects indicate that collection in VWL have greater leverage than wage labor. This could be due to the fact that it is a closer substitute in household labor allocation. An indication of the size of the effect of VWL on the probability of collection in NF can be obtained by multiplying the marginal effect, at sample mean, of the ``NF wage 2 VWL wage’ ’ variable (0.027) by the mean of the VWLwage variable for those who collect in VWL (7.15). On average the establishment of the VWL has changed the ``NF wage 2 VWL wage’ ’ variable by 27.15. This implies a reduced probability of 19% for collection in NF for those who collect in the VWL (27.15 times 0.027). Results from the model of time allocated to collection in natural forests are given in Table 3. The variables suggested by the model have overall high signi® cance. The marginal product in VWL, f¢V, is negative and signi® cant, suggesting that the higher the marginal product in VWL, the less time is allocated for collection in NF. The negative sign of the marginal product in NF f¢N, implies that when accessibility of fuel decreases, households spend more time collecting in the NF. If fuel becomes more accessible in NF, then households take out at

least part of this improvement in terms of reduced collection time.8 The coef® cients for family size, low caste households, and low caste women have the expected positive and signi® cant signs, while high caste households are not signi® cantly different from the reference group. Finally, the signi® cant Inverse Mill’ s Ratio reminds us that these results are conditional on the collection decision. Table 4 builds on these results, and quanti® es the impact of village woodlots on collection times and amounts in natural forests. First, the reduction in household time allocated to collection in NF is calculated using the marginal effect of f¢V on TN and the actual f¢V for collecting households as proposed in equation [14]. Village averages for households (N) affected by this time effect are presented as ``Decreased Time in NF.’ ’ This is the reduction in time devoted to collection in NF per household and year due to the availability of VWL. It is reasonable to expect, as suggested in equation [15], that the affected households would have collected in the NF roughly at their marginal productivity also in this time interval. That would result in a quantity reported in Table 4 as ``Decreased Collection in NF’ ’ (kg/hh/yr). If we expect 8 The negative coef® cient on marginal product in the natural forest may be due in part to measurement error, to the extent that this explanatory variable is a function of the dependent variable. The authors thank an anonymous reviewer for making this observation.

Land Economics

214

May 2001

TABLE 4 Changes in Collection in Natural Forest (NF) Due To Village woodlot (VWL), Orissa, India, 1995

Village Name

Decreased Time in NF (hours/ household/ year)

Decreased Collection in NF (kg/household/ year)

Decreased Village Collection in NF (kg/year)

Percent Decrease in Collection in NF

Champapedi Krushnapur Khandisi Kerendatangi Nakithana Arjunpur Kiapalla Raipada Kadamjhola Mayurjhelia Hariharpur Narasinghpur Patharbandha Tangi Balarampur Chandapur

140 160 100 60 190 440 0 310 170 220 70 380 90 0 170 120

870 840 600 540 1,040 1,580 0 1,490 900 780 410 860 760 0 1,140 430

38,800 29,200 11,500 32,000 17,000 42,700 0 41,800 9,900 17,600 3,400 76,300 5,600 0 15,900 28,800

29 14 4 21 6 23 0 5 4 18 11 17 24 0 13 18

Average

170

790

21,400

13

our sample to be representative for the rest of the village we can aggregate the reduced pressure to the village level, shown as ``Decreased Village Collection in NF.’ ’ Because the estimates presented in Table 4 are functions of other parameters which are estimated with some error (Tables 2 and 3), they should be interpreted as rough estimates. The variation in impact is large between villages (as expected). This method underestimates the impact, because it does not take into account the decreased pressure resulting from households abstaining completely from collection in NF due to VWL. This is especially evident in the case of Tangi, where no one in the sample collected in the forest. Still, substantial reductions in collection in NF are found. From these villages alone, pressure on nearby forests such as Dhani Reserve Forest is decreased by roughly 340 metric tons of biomass per year. This amount constitutes about 13% of current collection. The estimated decreased collection can also be compared with actual collection in VWL, which is reported to be twice as large. This de® es any simplistic notion of a one-to-one

relationship between collection in VWL and reduced collection in NF. III. SPATIAL VARIATION OF FUELWOOD COLLECTION IN NATURAL FORESTS In the preceding analysis we have shown how the collection decision depends on the relative returns from different sources of fuel. We have also indicated that distance to the forest is a crucial factor behind this. Because shadow wages are dif® cult to estimate, we will reexamine the decision to collect in NF, focusing on underlying factors that can be of interest for policies regarding forest intervention. For example, distance to NF could be expected to affect the shadow wage and thus behavior. If VWL at different distances to the NF have different impacts on collection, this will have implications for optimum locations of VWL. We therefore introduce an interaction term between the availability of VWL, expressed as hectares of VWL per household, and the distance to NF. Since we cannot expect this relationship to

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

215

TABLE 5 P robit Analysis of Households’ Decision to Collect in Natural Forest, Orissa, India, 1995, with Special Focus on Spatial Variation Variable Constant Age of head of household Family size Share of men in household Number of SCST women in household Low caste High caste Total income (10 -5 ) IPC area per household Age of IPC Village forest area per household Distance to NF Size of VWL per household times distance to NF Size of VWL per household times distance to NF squared

Marginal Effect

t-Value

0.21** 20.004** 0.02*** 0.26 0.19*** 0.06 0.07 20.34*** 20.004 0.007 20.03 20.13*** 20.73*** 0.13***

2.00 22.20 2.53 2.04 3.12 0.54 1.56 22.47 21.46 1.07 20.49 27.98 23.86 3.26

* indicates signi® cance at the 10% level; ** indicates signi® cance at the 5% level; *** indicates signi® cance at the 1% level.

FIGURE 1 Reduction in Collection in Natural Forest (NF) Due to Village Woodlot (VWL) as a Function of Distance

Land Economics

216

be linear, different speci® cations were tested and the quadratic form performed by far the best.9 The results of this alternative probit speci® cation can be found in Table 5. Since the shadow wages are excluded, the underlying household characteristics become even more pronounced. We can also note that our spatial variables, distance and interaction terms between distance and VWL, are all signi® cant at the one percent level. Figure 1 is based on both the estimated reduction in collection in NF (Table 4) and the estimated change in collection in NF due to interaction between availability of VWL and distance to the natural forest. The curve in Figure 1 is an extrapolation within the range of our observations of marginal effects, calculated at the mean distance of 2.8 km. The ® gure shows that the decrease in probability reaches a maximum at around this mean distance from the natural forest. The trend is quadratic in both cases, and shows an inverted U shape. This trend implies that to decrease the collection in natural forests using community plantations, these plantations should not be located near villages that are very close or very far from the natural forest. In the ® rst case, the natural forest will almost always be a superior source of biomass, and in the latter case, the forest is so inaccessible that other sources are chosen in any case. The fact that plantations have the potential to reduce deforestation is important. It is also important that the degree of success depends on location. Location-speci® c effects help to explain the variable success of plantations. These results motivate against summarily rejecting all plantations as a means to reduce fuelwood collection in natural forests. Although poorly located plantations may have little effect, appropriately located plantations appear to offer considerable promise as a means to accomplish reduced deforestation in natural forests. IV. DISCUSSION Reduced deforestation is a common reason for investment in social forestry projects. In that tradition earlier evaluations of the Orissa Social Forestry Project (e.g., SIDA

May 2001

1992) have postulated that the major environmental bene® t from the project would be decreased pressure on natural forest as the result of the establishment of 100,000 ha of community plantations. No empirical evidence has been given to support this contention. On the contrary, some forest of® cers in Orissa even argued that the secure user rights of the VWL would even increase the pressure on the de facto open access forest resources. We have shown that in the villages around Dhani Reserved Forest, establishment of VWL has both decreased the probability for collection in NF and the pressure on the forest among those who collect. At this point it would have been nice to be able to compare the environmental bene® ts from plantations at different locations with the costs to establish them in a way that could guide future forestry interventions. However, an estimation of the welfare effect of the decreased pressure is beyond the scope of this paper. Even if it had been possible to carry out such a social cost bene® t analysis it would not necessarily have guaranteed successful implementation, since that rests heavily on the incentives and resulting commitment of the surrounding people. In order to test whether there was such a commitment a contingent valuation study was carried out. In many of the villages people revealed a substantial willingness to pay for a new plantation, high enough to pass a net present value test (KoÈhlin 2001). One reason for this could be the fact that the plantations reduce the pressure on the natural forest and that households are already heavily involved in collection of nontimber forest products. With reduced pressure on the forest these incomes can be expected to grow, as will the sale of valuable species such as bamboo, sal, and teak. Johansson (1996) made a separate study that focused on the value of protection of NF in ® ve of the villages in our sample, Kiapalla, Barapalli, Ardjunpur, Balarampur, and Panaspur. She found that after eight years of protection the villages received gross returns of half a mil9 Different speci® cations for distance were also tested.

77(2)

KoÈhlin and Parks: Spatial Variability and Disincentives to Harvest

lion rupees worth of nontimber forest products yearly, when valued at market price of closest substitute. The commercial timber species were expected to give the returns of the same order of magnitude to the villages, despite a 50-50 sharing arrangement with the Forest Department. It is easy to make conjectures between the availability of alternative sources of fuel such as VWL and protection of the natural forest. A relevant question is whether the villagers themselves see this connection. One third of the households that have access to VWL felt it improved the possibility of protecting the natural forest. The major reason stated for this, however, was not the increased availability of biomass, but rather that the management of the VWL had shown the way to collective action. V. IMPLICATIONS FOR PROJECT PLANNING The most important result of this research is that plantations can be used to decrease the pressure on forests. This could imply that the criticized strategies to subsidize plantations might have been more successful than expected. This is particularly true if there are not enough local bene® ts for plantations to emerge spontaneously by collective action and if there are signi® cant positive externalities emanating from the forest. A ® ner detail is that we have found that plantations close to the forest are used less than plantations located further away. The impact of VWL on the pressure on natural forests therefore seems to be best described by the inverted U-shape of Figure 1. This ® nding has important implications regarding the ubiquitous recommendation to establish buffer zones around areas worthy of protection. The reason for this is that ``buffer zones have become so popular, in fact, that they are part of virtually all proposals for protecting natural areas’ ’ (Wells and Brandon 1992, 25). This is to a large extent due to their intuitive appeal and the combined biological and social bene® ts that they ``promise’ ’ (Poore and Sayer 1987). Buffer zones of this type have been integral parts of social forestry projects, such as

217

the ``interface forestry’ ’ component of the SIDA-supported project in Tamil Nadu (Eggert and Carlsson 1995). Actual success in achieving the expected bene® ts has been less than forthcoming. These experiences led Wells and Brandon to sum up: ``current buffer zone de® nitions are inconsistent and overlook practical problems, and this precludes their implementation in all but very limited circumstances. The buffer zone concept, although deceptively simple and intuitively very appealing, thus faces considerable challenges. It remains, however, a high priority for many conservation programs’ ’ (Wells and Brandon 1992, 27). Our analysis of the spatial variation in how plantations affect the pressure on natural forests is one piece of evidence that could make conservation programs more ef® cient in the future. If we combine this with the incidence of other potential bene® ts of plantations such as reduced time for collection and expenses for purchase of fuel, then, as a rule of thumb, village woodlots seem to be more bene® cial further away from the natural forest, where biomass is scarce and market purchases of fuel are common. References Amacher, G. S., W. F. Hyde, and B. R. Joshee. 1993. ``Joint Production and Consumption in Traditional Households: Fuelwood and Crop Residues in Two Districts in Nepal.’ ’ Journal of Development Studies 30 (1): 206± 25. Bluffstone, R. 1995. ``The Effect of Labor Market Performance on Deforestation in Developing Countries under Open Access: An Example from Rural Nepal.’ ’ Journal of Environmental Economics and Management 29: 42± 63. Cooke, P. A. 1998. ``Intrahousehold Labor Allocation Responses to Environmental Good Scarcity: A Case Study from the Hills of Nepal.’ ’ Economic Development and Cultural Change 46 (4): 807± 30. Eggert, H., and F. Carlsson. 1995. ``Economic Assessment of Interface Forestry in India.’ ’ Studies in Environment and Development. Department of Economics, GoÈteborgs University. Greene, W. H. 1995. LIMDEP: User’ s Manual, Version 7.0. Bellport, N.Y.: Econometric Software, Inc.

218

Land Economics

Jacoby, H. J. 1993. ``Shadow Wages and Peasant Family Labor Supply: An Econometric Application to the Peruvian Sierra.’ ’ Review of Economic Studies 60:901± 23. Johansson, M. 1996. ``Without the Trees: An Economic Study of the Value for Informal Protection in the Forest in Orissa, India.’ ’ Minor Field Study 1996:1. Department of Economics, GoÈteborg University. Kaimowitz, D., and A. Angelsen, 1997. ``A Guide to Economic Models of Tropical Deforestation.’ ’ Centre for International Forestry Research, Jakarta, Indonesia. Mimeo. KoÈhlin, G. 1998. ``The Value of Social Forestry in Orissa, India.’ ’ Ph.D. diss., Department of Economics, GoÈteborgs University. Ð Ð Ð . 2001. ``Contingent Valuation in Project Planning and evaluationÐ The Case of Social Forestry in Orissa, India.’ ’ Environment and Development Economics 6: 237± 58. Lambin, E. F. 1994. ``Modelling Deforestation Processes: A Review.’ ’ TREES Series B: Research Report No 1. Luxembourg: Of® ce for Of® cial Publications of the European Community. LoÂpez, R. E. 1986. ``Structural Models of the Farm Household That Allow for Interdependent Utility and Pro® t Maximization Decisions.’ ’ In Agricultural Household Models, ed. I. Singh, L. Squire, and J. Strauss. Baltimore: The John Hopkins University Press. Mekonnen, A. 1999. ``Rural Household Fuel Production and Consumption in Ethiopia: A Case Study.’ ’ Journal of Forest Economics 5 (1): 69± 97.

May 2001

Mercer, D. E. 1991. ``Application of Household Production Theory to Selected Natural Resource Problems in Less Developed Countries.’ ’ Ph.D. diss., School of Forestry and Environmental Studies, Duke University. Parks, P., E. B. Barbier, and J. C. Burgess. 1998. ``The Economics of Forest Land Use in Temperate and Tropical Areas.’ ’ Environmental and Resource Economics 11 (3± 4): 473± 87. Poore, P., and J. Sayer. 1988. The Management of Tropical Moist Forest Lands: Ecological Guidelines. Gland: IUCN. Sharma, R. A, M. J. McGregor, and J. F. Blyth. 1991. `The Social Discount Rate for Land-Use Projects in India.’ ’ Journal of Agricultural Economics 42:86± 91. SIDA (Swedish International Development Cooperation Agency). 1992. An Evaluation of the SIDA Supported Social Forestry Projects in Tamil Nadu and Orissa, India. Stockholm. Skou® as, E. 1994. ``Using Shadow Wages to Estimate Labour Supply of Agricultural Households.’ ’ American Journal of Agricultural Economics 76:215± 27. Strauss, J. 1986. ``The Theory and Comparative Statics of Agricultural Household Models: A General Approach.’ ’ In Agricultural Household Models, ed. I. Singh, L. Squire, and J. Strauss. Baltimore: The John Hopkins University Press. Wells, M., and K. Brandon. 1992. People and Parks, Linking Protected Area Management with Local Communities. Washington, D.C.: The World Bank, W.W.F, and U.S. A.I.D.

Deforestation in the Brazilian Amazon: Comparing the Impacts of Macroeconomic Shocks, Land Tenure, and Technological Change Andrea Cattaneo ABSTRACT. The paper examines the current relevance of the set of variables reported in the literature as driving deforestation in the Brazilian Amazon. The analysis uses a computable general equilibrium (CGE) model adapted to capture regional economic structures and the environmental processes speci® c to tropical areas. The paper compares the impact on deforestation in the Brazilian Amazon of: changes in real exchange rate; modi® cations in agricultural tax and support policies; reductions in transportation costs arising from investment in infrastructure in the Amazon; changes in land tenure regimes; and technological change in agriculture affecting productivity and agronomic sustainability. (JEL Q15, Q23)

I. INTRODUCTION Since colonial times, the settlement of new frontiers has been undertaken to open access to land and other kind of natural resources. In Brazil, macroeconomic policies, credit and ® scal subsidies to agriculture, and technological change in Brazilian agriculture have all acted as push factors in the migration process. On the other hand, regional development policies have pulled economic resources through the expansion of the road network, colonization programs, and ® scal incentives to agropastoral projects (Binswanger 1991). High transportation costs between the Amazon and the rest of the country, leading to high agricultural input costs and limiting inter-regional trade, also affect deforestation rates (Pfaff 1997). The proposed drivers of deforestation occur at different geographic scales and are linked to economic processes that span from macroeconomic policies to Amazon-speci® c conditions such as technology and tenure reLand Economics · May 2001 · 77 (2): 219± 240 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

gimes. For this reason computable general equilibrium (CGE) modeling is the preferred methodology to compare the order of magnitude of these different mechanisms affecting deforestation. In particular, it is important to specify the regional characteristics of agricultural production to understand how the interplay between different regions in Brazil determines deforestation rates. The paper is structured so as to ® rst clarify the modeling strategy considered appropriate for the problem at hand, second describe brie¯ y the database adopted, and then present the simulations as a way of comparing the importance of the different drivers of deforestation. II. MODEL CHARACTERISTICS The model adopted in this paper builds on the approach introduced by Persson and Munasinghe (1995) for a study of Costa Rica. They include logging and squatter sectors and therefore markets for logs and cleared land. We extend their approach to include land degradation as a feedback mechanism into the deforestation process. A review of The author is with the Resource Economics Division of the Economic Research Service (U.S. Department of Agriculture). The author thanks Steve Vosti, Sherman Robinson, seminar participants at the International Food Policy Research Institute (IFPRI), participants at the workshop on ``Technological Change in Agriculture and Deforestation,’ ’ held in Costa Rica in March of 1999, and two anonymous referees. On the Brazilian front, I would like to thank Eustaquio Reis and all the staff at IPEA for making this research possible and for their comments. This research was supported by IFPRI and DANIDA via the ICRAF-led Alternatives to Slash-and-Burn (ASB). Support was also given by the U.S. Environmental Protection Agency in the form of a graduate fellowship. The views expressed here are those of the author and do not necessarily re¯ ect those of the Economic Research Service or USDA.

Land Economics

220

May 2001

TABLE 1 Mapping of Activities, Commodities, and Factors Activity

Commodities Produced

Factors Used

Annuals production

Corn, rice, beans, mandioca, sugar, soy, horticultural goods, and other annuals

Arable land, unskilled rural labor, skilled rural labor, agricultural capital

Perennials production

Coffee, cacau, other perennials

Arable land, unskilled rural labor, skilled rural labor, agricultural capital

Animal products

Milk, livestock, poultry

Grassland, unskilled rural labor, skilled rural labor, agricultural capital

Forest products

Non-timber tree products, timber, and deforested land for agricultural purposes

Forest land, unskilled rural labor, skilled rural labor, agricultural capital

Other agriculture

Other agriculture

Arable land, unskilled rural labor, skilled rural labor, agricultural capital

Food processing

Food processing

Mining and oil

Mining and oil

Industry

Industry

Construction

Construction

Trade and transp.

Trade and transportation

Services

Services

CGE model applications to deforestation can be found in Kaimowitz and Angelsen (1998). This research is centered on the role of land as a factor of production. If land has different qualitative characteristics, which are perceived as distinct by economic agents, these characteristics will identify types of inputs into the production functions. In order to better describe the approach undertaken here, it is useful to de® ne some terms and concepts. Land is differentiated in land types on the basis of cover into the following categories: (1) forested land; (2) arable land; and (3) grassland/pasture. These distinctions are based on the qualitative characteristics which economic agents perceive as making these factors ® t for use in distinct economic activities. Land Transformation is de® ned as a transition between land types due to physical processes, given certain economic uses. An example of this is the transformation of arable land cultivated in rice into grassland/pasture. Land Conversion describes a transition

Urban skilled labor, urban unskilled labor, urban capital

between two land types brought about intentionally by economic agents. In the simulations presented below we allow for (1) farmers clearing forest to obtain arable land; and (2) farmers using arable land for pasture. The bio-physical component of the modeling framework affects the equilibrium stocks of the different land types given the land uses arising from the simulation. This framework is a ® rst step in linking bio-physical changes to the economic incentive for agents to modify existing land use patterns. These processes are a major constraining factor for regional development in the Amazon region. Our assumption is that natural transformation processes can be modeled as ® rst-order stationary Markov processes with land use entering as an exogenous variable (Baker 1989; Van Loock, Ha¯ ey, and King 1973). The results presented here rely on data collected through farm surveys by researchers at the International Food Policy Research Institute (IFPRI) in Acre and RondoÃnia.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

The activities considered in the model are presented in Table 1, along with the factors employed in production and the commodities being produced by these activities. Agricultural production is disaggregated by region (Amazon, Center-West, Northeast, rest of Brazil), by activities (annuals, perennials, animal products, forest products, and other agriculture), and, by size of operations (smallholder, large farm enterprise). Regional agricultural producers sell their products to a national commodity market. All factors employed by agriculture are regionspeci® c. Households are speci® ed at the national level and are disaggregated into ® ve categories: urban low income, rural low income, urban medium income, rural medium income, and high income. Price equations. The ® rst set of equations in Table 3 de® nes prices in the model. On the import and export sides the model incorporates the ``small country’ ’ assumption which states that world prices are exogenous. TABLE 2 Model P arameters and variables Sets A C F H LAND FCON FMIG

Activities Commodities Factors Households Land (ÌF) Factors involved in conversion (ÌLAND) inter-reg. mobile factors (ÌF)

FSf glesc htaxh i itaxaa ma, f1, f2 T tm c wfratf1, f2 yhfc f,h zlesc

TABLE 2 (Continued ) Variables ABSORB CDc EXR FDSCf,a FSAV FSf GDc GDTOT GR HREMITh IDc INVEST MPSh PAa PDc PEc PMc PQc PWEc PWMc PXc PXACa,c QAa QDc QEc QFCONf1, f2 QFMIGf QMc QQc QXc QXACa,c SAVING UESHf WFf,a WFAVGf YFCTRf YHh

Total absorption (sum of ® nal demand components) Final demand for private consumption Exchange rate (R$ per $US) Factor demand by sector Net foreign savings Factor supply Final government demand Total government demand Government revenue Remittances Final investment demand Total investment Marginal propensity to save Domestic activity price Domestic commodity price Domestic price of exports Domestic price of imports Price of composite good World price of exports World price of imports Average output price Price of commodity c from activity a Domestic activity output Domestic Sales Exports Factor conversion from factor f1 to f2 Net migration of factor f Imports Composite goods supply Domestic commodity output Domestic output of commod. c from activity a Total savings Share of factor f going unemployed Sectoral factor price Average factor price Factor income Household income

Functional dependen cies

Parameters clesc

221

Share of consumption allocated by commodity Factor supply in initial equilibrium Share of government exp. allocated by commodity Household tax rate Discount rate Indirect tax rate Land Transformation: factor f1 into f2 Planning horizon Tariff rate Wage ratio for ``connected’’ factor markets Share of factor income to household Share of investment allocated by commodity

CES CET TRANSLOG FOC1 FOC2 FOC3 FOC4

Constant elasticity of substitution Constant elasticity of transformation Translogarithmic ¯ exible functional form First order condition (FOC) for CES produc. FOC for translog commodity production FOC for CET transformation between products for export and domestic markets FOC for CES substitution in consumption between import goods and domestically produced goods

Land Economics

222

May 2001

TABLE 3 Model Description Price equations 1.

PMc 5 PWMc × (1 1 tmc) × EXR; PEc 5 PWEc × EXR PQc 5

PDc × QDc 1 PMc × QMc QQc

3.

PXc 5

PDc × QDc 1 PEc × QEc QXc

4.

PAa 5 TRANSLOG(PXACa,c, QAa)

2.

Import price and export prices Composite commodity prices

Producer prices Activity prices (multi-output activities)

Quantity equations 5.

QAa 5 CES(FDSCf,a)

Activity Production (CES)

6.

FDSCf,a 5 FOC1(WFf,a, PAa) QAa

Demand for primary factors

7.

QXc 5 CES(QXACa,c)

8.

QXACa,c 5 FOC2(PXc, PXACa,c) QXc

9.

QAa 5 TRANSLOG(QXACa,c)

Commodity demand (CES aggregation)

10.

QXACa,c 5 FOC2(PXa, PXACa,c) QAa

11.

QXc 5 CET(QEc, QDc)

12.

QEc 5 FOC3(PEc, PDc) QDc

13.

QQc 5 CES(QMc, QDc)

Disaggregated commodity demand Activity production (translog aggregation) Disaggregated multi-commodity production by activity a Output transformation (CET) for exporting sectors Export supply for exports Armington assumption: Composite commodity aggregation (CES)

QMc 5 FOC4(PMc, PDc) QDc Income equations 14.

15.

YHh 5

^ ^ yhfc fÎF

16.

GR 5

f,h

Import demand

× WFf,a × FDSCf,a 1 HREMITh

^ htax × YH 1 ^ itaxa h

hÎH

Household income

aÎA

h

a

^ tm

× PAa × QAa 1

c

aÎA

Government revenue

cÎC

× PWMc × QMc × EXR 17.

SAVINGS 5 GOVSAV 1 FSAV × EXR 1

^ MPS

h

hÎH

× YHh × (1 2 htaxh)

Total savings

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

223

TABLE 3 (continued ) Expenditure equations 18.

PQc × CDc 5

^ cles × (1 2 mps ) × (1 2 htax ) × YH c

h

h

Household consumption demand

h

hÎH

19.

GDc 5 glesc × GDTOT

Government consumption demand

20.

IDc 5 zlesc × INVEST

Fixed investment demand

Factor Supply and Demand, and Migration Relationships 21a.

FSf1 5 FSf1 1

^m

a, f1, f2

× FDSCf2,a 1

^ QFCON

Factor supply (no migration): Includes factor transformation for physical causes and factor conversion (such as deforestation)

f1, f2

f2 Îf

aÎA f2 Îf

21b.

FSf 5 FSf 1 QFMIGf for f Î FMIG

22a. 22b.

fÎFMIG

23.

WFAVGf 5

^ QFMIG 5 0; ^

Factor supply (with migration)

QFCONf1, f2 5 0

f

Conservation of total factor supply (for factors that are ``connected’’ through migration or conversion)

f1, f2 ÎFCON

^ WF

a, f

@ ^ FDSC

× FDSCa, f

aÎA

Average factor wage

a, f

aÎA

24.

WFAVGf1 5 wfratf1, f2 × WFAVGf2

25a.

FSf 5

For ``connected’’ factor markets the wage ratio is ® xed

^ FDSC

Factor market equilibrium (fully employed factors)

a, f

aÎA

25b

FSf 5

^ FDSC

a, f

[WFf . wf min f ]

Factor market equilibrium (potentially unemployed factors)

aÎA

26.

27.

3

UESHf 5 FSf 2

PX``def’ ’ 5

^ FDSC 4 @ FS a, f

Share of factor going unemployed (for potentially unemployed factors)

f

aÎA

WFAVG``gr’ ’ [1 2 e2iT ] i WFAVG``ar’ ’ 2 WFAVG``gr’ ’ 1 [1 2 e2(i 1m)T ] i1m

Deforestation demand: price is the expected NPV of returns to land

Macroeconom ic Closures 28.

QQc 5 CDc 1 IDc 1 GDc

29.

^ PM × QM 5 ^ PE × QE 1 FSAV 1 ^ HREMIT c

c

c

cÎC

30.

ABSORB 5

Commodity market equilibrium c

h

cÎC

^ PQ (CD 1 ID 1 GD 1 DST ) c

External account balance

hÎH

c

c

c

Total absorption

c

cÎC

31a. 31b. 32.

^ PQ × GD c

GOVABS 5

cÎC

ABSORB

SAVING 5 INVEST

c

; INVABS 5

^ PQ × ID c

cÎC

ABSORB

c

Government consumption and investment demand (® xed share of absorption) Saving-Investment balance

224

Land Economics

In equations [1a] and [1b] the domestic price of imports and exports is the world price times the exchange rate with tariff rates entering the domestic price of imports. The prices of composite commodities (made up of imports and commodities from domestic producers) are de® ned as a weighted average of domestic and imported commodity prices (eq. [2]). In a parallel manner, for any commodity, the aggregate producer price is a weighted average of domestic sales and export prices (eq. [3]). The (gross) price paid for any activity (revenue per unit of the activity) is a function of output and commodity prices (eq. [4]). Quantity equations. Equations [5] to [14] show the quantity equations for commodities and factors that are related to production and foreign trade (the latter only for commodities). Equation [5] de® nes the Constant Elasticity of Substitution (CES) production function which, for each activity, determines the relationship between the quantity produced and the use of primary factors. Equation [6] is the demand function for factors, derived from the ® rst-order condition for pro® t maximization subject to equation [5]. Equation [7] de® nes the demand at the national level for the commodities produced at the regional level. Equation [8] is the ® rst order condition for cost minimization and captures competition between multiple activities (distinguished by their speci® c technologies) producing a single commodity. Outputs from different activities are imperfect substitutes, an application of the Armington approach (commonly used for international trade) in a domestic setting. In addition to the standard one-to-one mapping between activities and commodities, equation [9] permits multiple outputs for any given activity. Agricultural technologies by sector are speci® ed as two-level production functions assuming separability between the two levels. At the lower level, real value added is a CES function of the primary factors of production; output by activity is a ® xed-coef® cients function of real value added and intermediate inputs. The output of the agricultural activity is transformed, at the second level, into commodities according to

May 2001

a smooth concave transformation frontier described by a translog function obtained as a production-side analogy of the Almost Ideal Demand System (Deaton and Muellbauer 1980). Convexity of the production set was checked according to Hasenkamp (1976). In effect, each agricultural activity produces a number of agricultural commodities. Equation [10] is a ® rst-order condition derived from cost-minimization subject to equation [9] and a ® xed aggregate output demand level. Equation [11] provides the Constant Elasticity of Transformation (CET) function that transforms domestic output to commodities to exports and domestic sales. Equation [12] is derived from pro® t maximization subject to equation [11] and a ® xed level of domestic output; it de® nes export supply as a function of relative prices. Equations [13] show how imports and domestic output sold domestically generate the composite commodities that are demanded by all domestic users. Equation [13] is the Armington function, that is, the CES aggregation function for imports and domestic output sold domestically. Equation [14] expresses import demand as a function of the relative prices of imports and domestic commodities; it is derived from cost-minimization subject to equation [13] and a ® xed level of composite commodity demand. It should be noted that the commodities QXAC, QX, QD, and QE are distinct, and associated with separate prices (PXAC, PX, PD, and PE, respectively). Imports (QM) and domestic goods (QD) are also distinct from their composite (QQ) with separate sectoral prices. Income equations. The model institutions are households, government, and the rest of the world. Factor income, as a function of factor demand and factor prices, is channeled completely to the household (eq. [15]). The household also receives part of its income as remittances from abroad. Government obtains its revenue from collecting income taxes, indirect taxes, and tariffs levied on imported goods (eq. [16]). Total saving, de® ned in equation [17], is made up of foreign, government, and household savings.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

Expenditure equations. Domestic ® nal demands are composed of private consumption, government consumption, and investment demand. For each household, consumption is determined by a Cobb-Douglas function, distributing marginal budget share across commodities (Eq. [18]). Similarly, Equations [19] and [20] assure, respectively, that government and investment demand are allocated across commodities in ® xed value shares. Factor Markets and Macroeconomic Closure

The supply of non-migrating factors depends on the initial stock, physical transformation, and conversion (eq. [21a]). Transformation is allowed from arable land to pasture/grassland. Conversion is allowed from forested land to arable land, and from unemployed arable land to pasture/grassland. In the long run scenarios we assume inter-regional mobility of labor and rural capital; therefore, the supply of these factors (eq. [21b]) depends on the migration required to maintain a ® xed-wage ratio between regions (eq. [24]). Equations [22a] and [22b] assure that, where migration and/or conversion are allowed, total factor supply is conserved. The equilibrium conditions for factor markets are de® ned in equations [25a] and [25b]. We assume in the short run that all factors except capital may go unemployed. In the long run only arable land may go unemployed, in which case it is converted to grassland/pasture. Flexible average factor prices perform the task of equilibrating each market; if the lower bound for a factor price becomes binding, a share of the factor will not be employed (UESHf de® ned in eq. [26]). To the extent that it is demanded by different sectors, each factor of production is assumed to be sectorally mobile inside its region. Equation [27] expresses the demand for agricultural land as an investment good and is derived in the next section. Equation [28] is the equilibrium condition for composite commodity markets: supply is set equal to the sum of ® nal demands; ¯ exible composite commodity prices assure that

225

this condition is satis® ed. Equation [29] speci® es the equilibrium condition for the current account of Brazil’ s balance of payments. The domestic price index was chosen as numeraire. Foreign savings is ® xed (the current account de® cit), and the real exchange equilibrates the current account. Absorption is de® ned in equation [30] as the sum of ® nal demands. This de® nition is drawn upon in equations [31a] and [31b] which determine the nominal values of government and investment spending as ® xed shares of absorption. Equation [32] de® nes the ® nal macro-closure condition, imposing equality between the values of total savings and total investment. Demand for Deforested Land

The price for arable land, Par, is determined by the return to agricultural land. In an in® nite horizon framework, the ¯ ow return from an asset divided by the asset price must be equal to the rate of interest in the steady state. This implies that the price of arable land the forest product activity (which produces arable land) would face, assuming a ® xed rental rate rar, would be: Par 5

#

T

0

rar e2it dt 5

rar [1 2 e2iT ]. i

This expression takes into consideration that an agricultural producer’ s decision to buy arable land depends on the tenure regime: if the land is subject to insecure property rights, the planning horizon will be ® nite. A limitation of the expression is that it does not take into account that the rental rate may vary with time due to decreasing or increasing productivity. For the purpose of this analysis it is reasonable to assume that arable land is transformed, through degradation, to grassland which can be used only for pasture. Let the degradation rate equal (the indices are dropped to simplify notation) and let rgr equal the rental rate of grassland, then the price for one hectare of newly deforested land, if we

Land Economics

226

assume the planning horizon to be T, is given by: dAar 5 2m Aar dt

so that Aar 5 A0 e2mt

with A0 5 1 (hectare)

then T

Par 5

#r 0

ar

e2it × e2mt 1 rgr e2it × (1 2 e2mt)dt

so the solution is Par 5

rgr (r 2 rgr) [1 2 e2iT ] 1 ar [1 2 e2(i1m)T]. i i1m

The deforesters, being the suppliers of arable land, are faced with this price and the amount of land that will be deforested will depend on Par and on the squatters’ pro® t maximizing behavior and technology. The behavior of agents carrying out the land clearing can be differentiated according to whether forest is an open-access resource, or whether property rights governing the use of the forest resource are well-de® ned. For the purpose of this paper we assume the forest is an open access resource by not taking into consideration the returns from standing forest in computing the pro® ts of deforesters. By assuming an in® nite planning horizon when using arable land, we allow agents to acquire full property rights through deforestation. While a broad consensus exists that the expansion of cropped area and pasture constitutes a major source of deforestation, there is no similar consensus with regard to logging. It appears, however, to be a direct source of deforestation in some contexts and to play an indirect role in others, by facilitating access to forested areas for farmers (Burgess 1993; Uhl and Vieira 1989). Logging in the Amazon, due to its selective nature,

May 2001

rarely leads to complete land-clearing. While it is well-known that logging can severely degrade forest and expose it to ® re risk, this analysis considers only land clearing for agricultural purposes as deforestation.1 In this paper we will take the approach that deforesters provide agricultural land to be sold to whatever agricultural entity is expanding, and logging, while not directly causing deforestation, is a complementary activity to land clearing. III. DATA BASE AND MODEL ASSUMPTIONS The data used in this model were drawn from Cattaneo (1998). The original sources used to construct the Social Accounting Matrix were the 1995 IO table for Brazil (IBGE 1997a), National Accounts (IBGE 1997b). These source were integrated with the Agricultural Census data for 1995± 96 (IBGE 1998) to yield a regionalized representation of agricultural activities. Household data was obtained from the national accounts and the household income and expenditure surveys. Total labor, land, and capital value added were allocated across the agricultural activities based upon the Agricultural Census. Labor was disaggregated into agricultural and non-agricultural labor, and further differentiated as Skilled or Unskilled. Gross pro® ts in agriculture were allocated in part to land based on the return to land being used by the activity (FGV 1998), and, for the remaining part, to capital. Regional marketing margins were estimated by calculating the average distance to the closest market, and using the ratio of these values relative to the industrial South to multiply the trade and transportation coef® cients of each agricultural sector as obtained from transportation cost surveys (SIFRECA 1998). Deforestation in 1995 was assumed to equal average deforestation rate between 1993 and 1996 (approximately 20,000 km2 yr21). 1 Logged areas, by maintaining some forest cover, are not perceived as deforested by the satellite imagery data on which this analysis is based. It would therefore be inappropriate to include logged areas as deforested.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

The coef® cients for deforestation technology were obtained from Vosti, Witcover, and Carpentier (2000). Timber production in the Amazon, and in the rest of Brazil, was obtained from the agricultural census. The economic rent to timber was based on a technological speci® cation proposed by Stone (1998). Elasticities of substitution between production factors were taken for industry from Najberg, Rigolon, and Vieira (1995). For agriculture, the substitution elasticity between land and capital was set at 0.4 for smallholders, and 0.8 for large farm enterprises. These values are judgment-based estimates, assuming farm enterprises can substitute more easily between factors. The speci® cation of multi-output production functions allows for the possibility that farmers consider certain agricultural commodities as substitutes, and others as complements, in the production process. The technology captures both price responsiveness, through own-price elasticities, and technological constraints in transforming agricultural output from one commodity to another through substitution elasticities. Values for these elasticities were obtained by distributing a survey among IFPRI and EMBRAPA researchers with expert knowledge about the production process in Brazilian agriculture. The results are presented in Table 4. The default option assumes high substitutability in production, and replicates the linear programming farm model approach to production in shifting production to the most pro® table crop. If, alternatively, the experts believed that farmers weigh price signals with other factors when making this decision, then substitution elasticities would be lower. Possible factors being considered were: (1) relative risk associated with the crops; (2) subsistence requirements; (3) crops requiring similar soil characteristics (substitutable) or different soil characteristics (less substitutable); (4) common practice (habit); (5) whether inter-cropping is common for two crops (in this case, at the extreme, there would be very low substitutability). On the bio-physical side, arable land is assumed to sustain annual production for four years before being transformed into pasture/ grassland. Livestock can be sustained for

227

eight years on pasture/grassland before degrading land completely.2 This implies that, on average, 25% of arable land in annuals and 12.5% of pastureland in livestock is transformed through biophysical processes. In concluding this section, we note two limitations in the data and model formulation. Due to the uncertainty surrounding the elasticities, the results of the simulations are meant to clarify the sign and order of magnitude of impacts of regime shifts and should not be interpreted as precise quantitative measures. For this precise reason, the results are presented as a range of possible outcomes given the range of possible parameters. Second, for the purpose of this paper, in which we want to compare the impact policy shocks in a controlled environment, a comparative statics framework is adopted, thereby barring any dynamic considerations. The CGE model was constructed using the General Algebraic Modeling System (GAMS) and solved using the PATH mixed complementarity solver available in GAMS. The model size was 1,417 variables and 1,417 equations (the number of variables and equations depends strongly on the regional speci® cation of the model: in this analysis Brazil was subdivided in four regions). IV. MODEL RESULTS In this section, a comparison is made between policy shocks at different levels of aggregation to provide guidance to policymakers concerning the order of magnitude of the impact of different scenarios. More speci® cally, should policymakers interested in slowing deforestation in the Amazon look at regional (Amazon), inter-regional, or national policies? Macroeconomic Shocks (I): Crisis and Structural Adjustment

In January 1999, the widespread rumor that states might default on their debt with 2 The degradation rates are conservative estimates based on personal communications with agricultural extension agents and are in line with values reported in the literature (Southworth, Dale, and O’ Neill).

Land Economics

228

May 2001

TABLE 4 P roduction technology: substitutability between commodities Technology

Commodity 1

Commodity 2

Substitutability

Annuals production

Corn Corn Corn Rice Rice Rice Beans Beans Mandioca Sugar Horticultural goods

Rice, beans Mandioca Sugar, soy, horticulture, other annuals Beans Mandioca Sugar, soy, horticulture, other annuals Mandioca Sugar, soy, horticulture, other annuals Sugar, soy, horticulture, other annuals Soy, horticulture, other annuals Other annuals

Low Low-medium Medium-high Low Low-medium Medium-high Low-medium Medium-high Medium High Medium-high

Perennials production

Coffee Coffee Cacau

Cacau Other perennials Other perennials

High Medium Medium-high

Animal products

Livestock Poultry

Milk Livestock, milk

Medium Medium-high

Forest products

Deforested land (agric.) Deforested land (agric.) Non-timber tree products

Timber Non-timber tree products Timber

Low-medium High High

Note: The elasticity values are: low 5 0.3, Low-medium 5 0.7, medium 5 1.0, medium-high 5 2.0, high 5 4.0.

the Brazilian federal government sent foreign investors ¯ eeing from the Brazilian capital market. The government, having to choose between making a stand for its overvalued currency or deciding not to intervene, opted to ¯ oat the exchange rate. The effect was an 80% nominal devaluation. In this section, we simulate a series of real exchange rate devaluations that range from 5% to 40%. The scenarios are distinguished by the time horizon of the adjustment process to the crisis in the following way: (1) Short-Run (SR)Ð this kind of scenario assumes that wages are rigid, and therefore, excess supply in the labor market is possible; we also assume that in the short run, labor and capital migration between regions is not possible; (2) LongRun (LR)Ð assumes wages are ¯ exible and migration between rural areas is allowed. Since the mechanisms underlying equilibrium in the labor and capital markets are complex and uncertainty abounds concerning how these markets react to differentials in factor returns, the results are presented as a domain of possible outcomes bounded by two extremes represented here by the shortrun and long-run scenarios. The rationale for

this choice is that this domain of possible outcomes will capture intermediate situations such as factor markets reacting only after a threshold in inter-sectoral, or inter-regional, differential in factor returns is overstepped. Throughout the results section the analysis is presented according to this general objective of bracketing possible outcomes relative to critical parameter values. In so doing, reference will be made to upper and lower boundaries in deforestation rates to shed light on the extreme situations that can occur. Changes in the exchange rate reverberate through the economic system by affecting the relative prices of goods. On the supply side of the economy, prices of export goods rise relative to non-traded goods sold domestically (e.g., services and construction). This implies that production shifts toward sectors that produce goods with a high export share. Conversely, on the demand side, the rise in price of imported goods leads to a greater demand for domestic substitutes of the imported goods. Given enough microeconomic detail in the CGE model, it is possible to follow the reverberations of a macroeconomic shock throughout the economy, for example,

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

FIGURE 1 Change in Deforestation Rates Following a Real Exchange Rate Devaluation

in our case to regional agricultural production sectors. The scenario presented, with a balanced reduction of private consumption, government demand, and investment would lead to a reduction in deforestation which would be substantial in the short run but this effect would be attenuated in the long run (see Figure 1). The mechanism underlying the decrease in deforestation for the balanced contraction scenario is linked to the performance of Amazon agriculture relative to agriculture in the three other regions of Brazil. A devaluation is usually thought to favor agriculture because it produces exportable goods, therefore one would expect the incentive to deforest for agricultural purposes would increase with the devaluation. This does not occur in the balanced contraction scenario for two reasons: (1) the Amazon has a smaller share of its production allocated to exports therefore, while agriculture as a whole does expand, Amazon agriculture suffers a contraction relative to the other regions; and (2) since the Amazon produces more for the domestic market, the contraction in private consumption affects Amazon agricultural production more than production in the other regions.3 Macroeconomic Shocks (II): Fiscal Reform in Agriculture

Care was taken to describe sectoral and regional subsidies and taxes for agricultural ac-

229

tivities using detailed data from the Ministry of Agriculture and from the Organization of Brazilian Cooperatives. Even so, the Brazilian tax system is extremely complex and considerable uncertainty remains over the average regional tax and subsidy rates by activity. In addition, the subsidies are often in the form of subsidized credit which is not well represented in a static framework. The sectoral data was reconciled with national accounts data on agriculture. Two separate simulations were carried out: (1) abolishing agricultural subsidies nationwide; and (2) imposing a revenue neutral uniform tax on agricultural activities. What emerged, even with sensitivity analysis, is that the ® scal incentives leading to deforestation had already been reduced by 1995. The impact on deforestation rates was limited compared to the other policy shocks presented in this paper. The change was minimal when the subsidies were removed (1% change in deforestation rates) and somewhat more pronounced for the harmonization of the agricultural tax system with a decrease of 3.4% in deforestation rates in the short run which turns into a 2.4% increase in the long run. Evidently little impact can be expected on deforestation rates from any foreseeable ® scal reform in agriculture.4 Inter-Regional Processes: Infrastructure Improvement in the Amazon

In this section we analyze the impact of a reduction in transportation costs. The policy relevance of such a scenario stems from the changes occurring in infrastructure in the region. A road through the Amazon to the Pa3 A caveat concerning the long-run results is appropriate. The effect on deforestation for agricultural purposes is extremely dependenton the migration ¯ ows being allowed. If migration is allowed only among the rural laborers of different regions, we obtain assuming a 30% devaluation, a decrease in deforestation of - 5%. Alternatively, if we believe that the crisis is such that urban labor is willing to migrate to the Amazon (and no other region), then we obtain a 35% increase in the deforestation rate. 4 The tax rates on agricultural production vary sectorally from zero to 10% and the uniform tax solution, depending on the reforms speci® cation, ranges between 1.7% and 2.7%; therefore, the simulated reform is not minimal although the impact on deforestation is limited.

230

Land Economics

ci® c is under construction through RondoÃnia, and a port facility is also under construction in RondoÃnia, in order to reduce transport costs for soybeans and other products produced in the region. On the eastern side of the Brazilian Amazon, the Center-North Multimodal Transportation Corridor, covering an area that includes the Southeast of ParaÂ, eastern Mato Grosso, and the south of MaranhaÄo, will reduce the transportation costs of grains considerably with investments in roads, railways, and waterways. The incentives that shape current land use patterns in the area may therefore undergo considerable shifts. We assume here that costs are reduced uniformly for all agricultural products produced in the Amazon.5 In all cases, a reduction in costs for transportation between the Amazon and the rest of Brazil increases deforestation rates. For small decreases in transport costs, one can ignore the uncertainty surrounding how elastic the response of the national commodity market is to increased agricultural products from the Amazon; however, for a large decrease in costs it will be important to know how the agricultural commodity markets react to such a shock. Since data to estimate such elasticities are not available, the results provided here are based on sensitivity analysis: simulations were performed with values for these elasticities ranging between 1 and 12. Since agricultural products are generally good substitutes, this range should bracket the true, but unknown, elasticity values. A 20% reduction in transportation costs for all agricultural products from the Amazon causes an increase in deforestation in the range of 21% to 39%. (see Figure 2).6 As agricultural production in the Amazon becomes more pro® table, the price of arable land increases, thereby increasing the incentive to deforest. The increase in pro® tability leads, in the long run (with mobile agricultural labor and capital), to a 6%± 23% increase in production by smallholders and a 3%± 9% increase in production by large farms. However, welfare effects at the national level are very limited (rural households at the national level gain 0.5± 0.9% in real income). This is because the increase in Amazon production, except for what share of

May 2001

FIGURE 2 Change in Deforestation Rates if Infrastructure to Amazon is Improved

it is exported, replaces previous production from other regions; therefore, the positive regional impact on Amazon development is offset by the negative impact on other agricultural areas of Brazil. Amazon-Speci® c Processes (I): Tenure Regime Regulation

The economic literature linking deforestation to tenure regimes has either adopted a 5 The population in the Brazilian Amazon is 62% urban: reduction of transportation costs to the Amazon of externally-produced goods could have a signi® cant impact on deforestation rates. We do not consider this possibility in our simulations because transportation costs to the major urban centers would not be affected by the proposed infrastructure schemes since these centers are already positioned on major waterways; the proposed schemes aim mainly at providing improved transport for grains out of the Legal Amazon with the secondary bene® t of transporting fertilizers, processed foods, and consumer goods into the Amazon region (da Costa Marques and Caixeta Filho 1998). Since trade into and out of the Amazon occurs for different goods in the two directions, and given that the infrastructure improvement in the Amazon would bene® t producer nodes rather than consumption nodes, we feel that considering solely the effect on deforestation associated with reduced costs of production is an acceptable approximation for this policy scenario. 6 The degree of complementarity in production between logging and deforestation activities appears to be crucial in determining the magnitude of the increase in deforestation. If it is assumed that producers view these as substitutable activities, the deforestation rate following the reduction in transportation costs increases even more dramatically.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

partial equilibrium approach (Mendelsohn 1994) or an econometric approach based on the explanatory power of measures of tenure security using cross-country data (Alston, Libecap, and Schneider 1996; Deacon 1999). The approach adopted here is similar to Mendelsohn’ s partial equilibrium description, however, the context in our case is one of general equilibrium. While in the partial equilibrium setting deforesters had the choice between sustainable forest uses and a destructive agricultural process with decaying physical output, in a general equilibrium framework squatters have an array of additional choices, from wage labor on large farms, to migrating to urban areas, to simply cultivating the already cleared land. First, the assumptions made in simulating changes in tenure regimes have to be laid out. We assume deforestation occurs exclusively to clear land for agricultural purposes. The second important assumption is that in the reference equilibrium the deforestation activity leads to the acquisition of property rights to unclaimed land. This implies that there is a speculative component linked to deforestation for the acquisition of informal tenure security on top of the value of deforestation associated with higher returns in agriculture of land in its arable state relative to the forested state. In the case of de facto property rights being acquired through deforestation, it is interesting to analyze the impact of a change in tenure regimes such that these property rights are made to be insecure through eviction if deforestation is carried out on untenured land. This change will affect the speculative motive behind deforestation. The change in tenure regimes can be represented in one of two ways: (1) as an increase in the discount rate equal to the probability of eviction (Mendelsohn 1994); or (2) as a decrease in the expected time of residence on the plot before eviction. In the analysis that follows the latter of the two options is adopted.7 The results, presented in Figure 3, show the percent change in deforestation rate as a function of the expected time to eviction. The shaded area represents the domain of possibility described by a variability range in the discount rate of 15% to 30%, believed to bracket the true discount rate of farmers in

231

FIGURE 3 Impact on the Deforestation Rate of Regulating Access to P roperty Rights

the Amazon. The lower boundary of the region occurs when the discount rate is 15% and shows a slow decrease in the deforestation rates for reducing the expected time of residence on the plot from 22 years up to 16 years (220%) and decreases markedly from there on (230% for 14 years). The deforestation rate levels off at around 32% of its original value when the expected time of residence is reduced to 10 years. The leveling off is due to the fact that as the risk of being evicted increases, it becomes more convenient to deforest previously tenured forestland rather than unclaimed land. A regime switch occurs from deforesting as capitalization on property right acquisition (even if unsecured) to deforesting exclusively for the value added that comes from agricultural activities. We therefore observe an ``optimal’ ’ level of deforestation rate (given the 1993± 96 average) would be around 6,700 km2 yr2 1. This value, far from arresting deforestation, is still much lower than the current trend, suggesting that 7 The difference between the two approaches is that the ® rst assumes an equal probability of eviction over time, thereby additionally discounting for the risk at every point in time beginning at t 5 1, instead the second option calculates the returns based on the expected time before eviction and therefore maintains the discount rate unchanged but does not include any revenues that could be obtained after the expected eviction date. The rationale underlying our choice is that probability of eviction is unlikely to be constant over time. The two methods can be compared, if the probabilities are constant, by using the formula E(T) 5 1/pevic.

232

Land Economics

the mode of tenure acquisition and its enforcement should be top priority issues. If, on the other hand, the discount rate is higher than 15%, the leveling off will be reached for expected times lower than 10 years (the upper boundary, using a discount rate of 30%, reaches the leveling-off value at 6 years). Decreasing the expected time of residence on untenured land will have different impacts on different crops. In particular, it will prevent investment in perennial crops. Since perennials make a very ef® cient use of land in terms of value of output per hectare, one might expect that an expansion of activities that make more extensive use of land would counterbalance the deforestation rate reduction stemming from the tenure regime change; however, this is not the case. To simulate a situation where perennials production is adversely affected by the change in tenure regimes (assuming that perennials are on untenured land) a total factor productivity shock of 235% was imposed to mimic the decreased expected returns from perennials when the expected residency on untenured land is 10 years. The impact of such a scenario was to further decrease deforestation rates (to 4,900 km2 yr2 1). This counterintuitive result is associated with the decrease in returns from arable land resulting from the lower expected returns from perennials.8 The assumption that all current deforestation occurs on unclaimed land may cause the results to overemphasize the impact of regulating tenure. If a share of the deforestation is already occurring on tenured land then this will raise the ``¯ oor’ ’ in the deforestation rate because this component will not be affected by changing tenure regimes. Since by construction, we begin from an equilibrium point, we can neither validate nor contradict the hypotheses put out in the literature that tenure leads to more deforestation (Vosti, Witcover, and Carpentier 2000) or to less deforestation (Deacon 1999). All this analysis can say is that relative to the 1995 base structure of the economy, assumed as an equilibrium, if unclaimed land is being deforested then increasing the probability of eviction will decrease the deforestation rate to the point where it is pro® table to clear only previously tenured land. In this respect, the results contradict the

May 2001

partial equilibrium results of Mendelsohn (1994), which state that the possibility of eviction leads to destructive land uses. Amazon-Speci® c Processes (II): Technological Change in Amazon Agriculture

At the local level, much has been done on technologicalchangeinthe Amazon.Different farming and cattle raising systems have been analyzed (Toniolo and Uhl 1995; Mattos and Uhl 1994). The approach taken in this paper refers to the impact of exogenous technological change as it occurs at the Amazon Basin level and it expresses a modi® cation in the structure of a producing sector as a whole.9 We simulate technological change in annuals production, perennials production, and animal products. Separate simulations are run for smallholder technological improvements and large farm improvements in technology.10 For each activity, different types of 8 In the present version of the model, farmers have no alternative source of newly arable land other than land with insecure tenure. In reality farmers would buy fully tenured arable land to produce perennials since they are a high value crop. Modeling this would require two parallel markets for land, one for tenured and one for untenured land (for each land type). Unfortunately, no data are available on the crop areas on tenured and untenured land, making this approach dif® cult to implement. The shift in production of perennials towards tenured lands would raise the price of tenured land and increase the incentive to deforest. However, with perennials occupying only 1.4% of deforested land in the Brazilian Amazon and large areas of tenured arable land already available the upward pressure on the deforestation would be very limited. The conclusion is that the imperfect approach followed in this paper is an acceptable approximation of reality. 9 Different levels of technological change at the sectoral level can be associated with either the technology shift or with the extent to which a technology is adopted. For example, if all producers adopt a technological innovation causing a 50% improvement in total factor productivity (TFP), this is equivalent in our framework to half the producers adopting a technology leading to a 100% improvement in TFP at the plot level (ignoring non-linearities). 10 Cattaneo (2000) presents results for simultaneous technological change in both smallholder and large farms: the consequences on deforestation differ in some respects from the results presented here. This highlights the importance, in determining deforestation outcomes, of who adopts technological innovation since innovation alters the production decisions of adopters and nonadopters alike.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

technological change are analyzed: we have a reference run in which total factor productivity is increased up to 70% in 10% increments (disembodied technological change). The other simulations replicate the productivity increase of the TFP case by acting on the productivity of speci® c factors (embodied technological change: labor productivity improvement, capital productivity improvement, land intensi® cation through combined labor and capital productivity improvement). In the factor-speci® c cases, the extent of the factor productivity increase is calculated as inversely proportional to the factor’ s value share in production (to replicate the TFP case). For comparison purposes, the TFP Index associated with an instance of embodied technological change is de® ned as the TFP increase used as reference to obtain the factor productivity increase. Comparison across simulations of the different types of technological change are performed by representing the results relative to their TFP index. Given that it is dif® cult to state that innovation at the Amazon-wide level is clearly labor improving or capital improving, the results are presented as a range of possibilities covering all four types of technological change and their combinations. We carry out simulations for the short run (1± 2 years), in which agricultural labor and capital are con® ned to their regions, and for the long run (5± 8 years) by allowing these factors to migrate inter-regionally. Improving technologies: the short run impact.

In the short run, increasing the productivity in annuals cultivation will tend to cause an increase in the deforestation rate. The TFP case, in which factor productivity in annuals is increased by the same amount for all factors, appears to lead to the greatest deforestation (upper boundary in Figures 4a & 4b), both if the technology change is adopted by smallholders and by large farms. The lower bound of the shaded area represents, in the smallholder case, the land-saving technological change, while for large farms the lower bound is given by capital-intensive technological change (which attracts resources away from capital-intensive livestock which has a lower effciciency in land use). Technological change in annuals leads to a greater

233

increase in deforestation if applied by large farms due to the fact that by shifting their resources away from livestock into annuals, besides transmitting the yield improvement to the returns to arable land, they crowd out smallholders from annuals and push them into livestock which makes extensive use of land. This phenomenon does not occur if the innovation is adopted by smallholders because large farms have a very limited production of annuals. Increasing productivity in perennials cultivation has a generally positive potential for reducing the deforestation rate (Figures 4c & 4d). In the short run, any technical change embodied in capital and/or labor has the effect of lowering the price of pasture and the counter-intuitive effect of decreasing the demand for arable land (lower bound for arable land rent becomes binding), letting arable land be used as pasture. The underlying cause of this shift is that perennials make an intensive use of labor and capital per hectare cultivated (more than annuals). This result implies that as resources are drawn to perennials there will be less overall demand for arable land. A second reason for the decrease in deforestation is that perennials, as opposed to annuals, do not cause transformation of arable land to grassland; therefore, there is a stock effect whereby the amount of available arable land increases, tending to reduce the demand for deforestation. Smallholders and large farms react differently, based on their factor endowments, to different types of technological change in perennials: smallholders adopt more widely innovations that are labor-intensive and, conversely, large farms prefer capital intensive changes. The preferred options lead to a greater expansion of perennials meaning a greater decrease in deforestation; therefore, in Fig. 4c and 4d the lower bounds of the shaded area represent, respectively, labor-intensive innovation for smallholders and capital-intensive change for large farms. There is an expectation that improved livestock technologies in the Amazon, by allowing for a more land-intensive production system (combined with appropriate regional policies), will reduce deforestation (Mattos and Uhl 1994; Arima and Uhl 1997).

234

Land Economics

May 2001

FIGURE 4 Short Run Impact on Deforestation of Technological Change Presented as a range of possibilities covering the different types of technological change, including changes in total factor productivity (TFP), labor productivity (LAB_PRD), capital productivity (CAP_PRD), and land intensi® cation (LANDSAV). Only the upper and lower bounds of each region are labeled.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

235

FIGURE 5 Long Run impact on deforestation of technological change Presented as a range of possibilities covering the different types of technological change, including changes in total factor productivity (TFP), labor productivity (LAB_PRD), capital productivity (CAP_PRD), and land intensi® cation (LANDSAV). Only the upper and lower bounds of each region are labeled.

236

Land Economics

The results concerning technological change in the livestock sector are considerably different if the innovation is adopted by smallholders or by large farms. In the case of smallholder adoption the impact is always an increase in deforestation, the magnitude of which depends on the type of innovation: from a substantial increase for TFP changes to nearly no change in the land-saving scenario (see Figure 4e). The increase in deforestation rates can be traced back to the transfer of smallholder resources from annuals and perennials into the livestock sector that has higher requirements of land per unit value of output. Even arable land is employed in part for pasture as the livestock sector becomes more pro® table. This is the least cost solution in the short run. In fact, with a TFP index equal to 3, demand for arable land is reduced by 43± 53% in all scenarios except the TFP case. Technological improvement in livestock for large farms appears to have great potential to reduce deforestation rates (see Figure 4f). The difference relative to smallholders adopting this form of innovation is that large farms already have a large share of their resources allocated to livestock production, therefore by adopting new land-saving technologies large farms move resources within the livestock sector leading to a considerable reduction in their land needs. Combine this with arable land being employed in part for pasture, and the value of grassland decreases due to excess supply, therefore the incentives to deforest decrease. Improving technologies: the long-run impact.

In the long run, allowing for migration of labor and capital between regions, technological improvement in annuals production would lead to higher deforestation rates than in the short-run case (see Figures 5a and 5b). The basic tenet is that with mobile factors, land becomes the scarse factor. This implies that the returns to arable land are higher than in the short run case, creating incentives to deforest. The increase is more marked if the technology is adopted by large farms. Underlying this difference is the same mechanism described for the short-run case. In the long run, the results are still encouraging for perennials. As pertains to

May 2001

smallholders, the labor-intensive innovation is still the best option in the long run in terms of deforestation because with migration the shift to producing more perennials and less annuals and livestock remains; however, it is attenuated relative to the short run (see Figure 5c). The underlying process is unchanged, but with migration there is no surplus arable land to be used as pasture; in fact, arable land increases in value. However, deforestation is still reduced due to the dampening effect of lower returns to pasture land due to factors shifting towards the production of perennials. This dampening effect is present also in the TFP and the more capital-intensive scenarios. However, it is not enough to offset the prospect of higher returns from arable land, so deforestation increases in the long run if these types of innovations are adopted by smallholders. On the large farm front, increasing productivity in perennials is a safe bet to reduce deforestation. The upper boundary in Figure 5d is given by the capital-intensive innovation, which was found to be very good in the short run and the lower boundary is now the labor-intensive technological change. The reason for this reversal is that in the short run labor is scarce and capital abundant for large farms, and therefore perennials production by large farms is favored by capitalintensive change. However, perennials technology is very intensive in labor to begin with, therefore, in the long run where labor availability is no longer an issue, large farms favor labor-intensive innovations. In each case, the preferred option is the one that leads to the greatest expansion of perennials with the ensuing decrease in deforestation. The idea that improved livestock technologies in the Amazon would reduce deforestation rates by allowing for a more landintensive productionsystemappears tocapture a short-run view. This viewpoint does not take into consideration the long term effects of a more pro® table ranching sector in the Amazon. In fact, in the short run, all improvements not directly impacting the productivity of land do lead to a reduction in deforestation, but this does not hold true in the long run (Figures 5e and 5f). In the long run, as resources can be at-

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

tracted from outside the Amazon, the increased demand for pasture is met by increasing deforestation. In all scenarios, improving livestock productivity in any way will substantially increase deforestation in the long run. The increase in deforestation rates is particularly strong if the adoption of technological change in the livestock sector is carried out by the large farms (Figure 4f). The reason for this dramatic increase is that, in the case of large farm adoption, not only the return to pasture land increases substantially, but also the price of arable land increases. The increased price of arable land comes about because production of annuals leads to land degradation and subsequent use of the land as pasture; therefore, as keeping the land in pasture becomes more attractive, the demand for arable land increases in expectation that it will be used as pasture in the future. In fact, in all the long-run scenarios, production of annuals increases alongside that of livestock (although at a lower rate). Perennials, which are also produced on arable land but do not cause degradation, do not expand, and may actually be reduced. To conclude this section, we try to summarize and compare the impact of different types of technological change in the Brazilian Amazon to determine the possible tradeoff between developing agriculture and reducing deforestation.11 The best option, in deforestation terms, is technological change in perennials, which also has positive income distribution effects by favoring small holdings. However, from a purely revenue-driven perspective, livestock is the best alternative for both small and large farms. This leads to an unfortunate dilemma because any form of technological improvement in livestock will lead to greater deforestation rates in the long run. Improvement in annuals production, while possible in certain parts of the Amazon, would probably bring higher deforestation rates with returns of the same magnitude as perennials; therefore, improvement in annuals does not appear particularly appealing.

237

pact on deforestation of the variables discussed in the literature as driving the process. In this section we draw general conclusions and determine what policy levers are available to policymakers. With the present changes occurring in transportation costs, deforestation rates will probably tend to increase in the future. How much of the Amazon products can be absorbed by the markets determines how dramatic this process may be, and second, the link between logging and deforestation will also affect the deforestation outcome. On the macroeconomic side, it is important to note that if appropriate policies are adopted to address a devaluation a decrease in deforestation rates can be obtained. A 40% devaluation in real terms causes, in the long run, a 12% decrease in deforestation, meaning approximately a 2,400 km2 yr2 1 difference in the deforestation rate. This implies an interesting opportunity for policymakers at the national level because the balanced contraction policy which is aimed at adjusting the national accounts, has a bene® cial effect on the deforestation rate, and could offset, at least in part, the increase in deforestation rate associated with decreasing transport costs. At the Amazon policy level of analysis, regulating tenure regimes is the best option to reduce deforestation, assuming that current deforestation is, in large part, occurring at the hands of untenured deforesters who acquire tenure in the process. Unfortunately, new tenure regimes are very dif® cult to implement and enforce in a region the size of the Brazilian Amazon. This result however, supports initiatives that aim at creating buffer zones with integrated participatory management, creating clear property rights in the buffer zone, and discouraging any infringement on protected areas. The adoption of technological change in the Amazon is also of interest to policymakers to the extent that it can be directed through funding and agricultural extension. It is encouraging that the order of magnitude of the inter-regional effects on deforestation can

V. CONCLUSIONS The intended purpose of this paper was to compare the order of magnitude of the im-

11 The impact on value added generation is not reported here for reasons of space but is available upon request from the author.

Land Economics

238

be partially offset by technological change if the technologies are carefully chosen. The results point out that there is a signi® cant trade-off between forest conservation objectives and agricultural growth. Livestock technology improvements appear to have the greatest returns for all agricultural producers in the Amazon; however, deforestation increases dramatically in the long run for all improvements in livestock technology. The alternative would be to pursue improvement in perennials’ technology, especially in labor-intensi® cation, which would have a small bene® cial effect if applied by smallholders and could reduce deforestation rates considerably if adopted by large farms. Equity effects of improving perennials would be progressive because small farmers would be the ones to gain the most. Although this option has theoretical potential, the nonadoption by large farms (which would have small gains), combined with the riskaverseness of smallholders, would probably limit the effectiveness of this solution. Nonetheless, even if adopted only in part, it would still contribute to reducing deforestation rates. Improvement in production of annuals appears to have little potential: in the long run it would increase deforestation, and income effects would be quite small.

May 2001

The striking difference in deforestation rates between the short run and the long run points out that inter-regional ¯ ows of labor and capital play a crucial role in determining the expansion of the agricultural frontier. On the technology front, we also observe that the agents being affected by the improvement, whether smallholders or large farms, also makes a difference given their different factor endowments. A ® nal point to conclude the paper is that, unless deforestation occurs for subsistence needs in an area isolated from markets, the transmission mechanisms from the nonfrontier part of the economy to the agricultural frontier are many and intertwined. Understanding these mechanisms is important in predicting the impact of policies on deforestation, something partial equilibrium analyses are not always well-equipped to do.

APPENDIX A crucial component in determining the relevance to deforestation rates of tenure regime regulation is the impact it has on the prices paid for deforestation services provided on tenured and untenured land. From Table A1 one can observe that the price paid for deforestation on tenured land increases as the supply of deforestation on

TABLE A1 Impact of Regulating Access to P roperty Rights on P rices for Newly Deforested Land Expected Years of Residence Before Eviction 22

20

18

16

14

12

10

8

6

4

172 254

173 249

174 242

175 233

177 222

180 208

185 190

190 Ð

190 Ð

190 Ð

88 132

88 132

88 131

88 130

89 127

89 124

91 118

93 108

93 93

i 5 0.15 Net price if tenured Net price if untenured i 5 0.30 Net price if tenured Net price if untenured

93 Ð

Note: Deforestation is in 10 3 km2/yr and net price for deforestation is in R$/ha. Two cases are presented: 15% and 30% discount rate.

77(2)

Cattaneo: Deforestation in the Brazilian Amazon

239

TABLE A2 Amazon Agricultural Technologies: Factor Intensities in the Base (in terms of output) Small Farms

Labor Capital Arable lnd. Grass lnd. Forest lnd.

Annuals

Perren.

546 209 717

354 431 456

Animal 277 459

Large Farm Enterprises Other Agric.

Annuals

226 325 179

336 159 1168

3,268

Perren. 483 421 469

Animal 25 638

Other Agric. 291 355 139

Forest Products and log/defor 232 205

9,182 21,634

Note: The units express: for labor the number of workers involved in producing 1 million R$ in the activity; for capital the monetary equivalent in thousands of R$ of physical capital involved in producing for 1 million R$; for land the number of hectares required to produce 1 million R$ worth of output. Due to the different units, only cells in the same row can be compared. Source: computed by author based on a database made available by the Empresa Brasileira de Pesquisa AgropecuaÂria.

untenured land decreases (due to increased possibility of eviction); however, the price increase is not dramatic.12 The factor intensities of the different agricultural technologies in the Amazon are central in determining how policy shocks and technological change affect deforestation rates. More speci® cally in terms of cleared land, perennials require the least hectares per unit value of output while large farm animal production requires the most.

References Alston, L. J., G. D. Libecap, and R. Schneider. 1996. ``The Determinants and Impacts of Property Rights: Land Titles on the Brazilian Frontier.’ ’ Journal of Law Economics and Organization 12 (2): 25± 61. Arima, E. Y., and C. Uhl. 1997. ``Ranching in the Brazilian Amazon in a National Context: Economics, Policy, and Practice.’ ’ Society and Natural Resources 10:433± 51. Baker, W. L. 1989. ``A Review of Models of Landscape Change.’ ’ Landscape Ecology 2 (2): 111± 33.

12 As mentioned in footnote 8, the model does not have a market for deforestation on tenured land; therefore, the price paid for deforesting tenured land presented in Table A1 is hypothetical. It was computed expost once the changes associated with the tenure reform are observed in the model. The price can be interpreted as an upper bound on the price of deforestation services on tenured land had there been a market for it in the model.

Binswanger, H. P. 1991. ``Brazilian Policies that Encourage Deforestation in the Amazon.’ ’ World Development 19 (7): 821± 29. Burgess, J. C. 1993. ``Timber Production, Timber Trade, and Tropical Deforestation.’ ’ Ambio 22 (2± 3): 136± 43. Cattaneo, A. 1998. ``The Interaction Between Economic Incentives, Deforestation, and Land Degradation in Brazil’ ’ . In: The Impact of MacroeconomicPolicy on Deforestation: A Comparative Study of Indonesia and Brazil. Progress Report. Washington, D.C.: International Food Policy Research Institute. Ð Ð Ð . 2000. ``A General Equilibrium Analysis of Technology, Migration, and Deforestation in the Brazilian Amazon.’ ’ In Agricultural Technologies and Tropical Deforestation, ed. A. Angelsen and D. Kaimowitz. New York: CABI Publishing. Da Costa Marques, R. W., and J. Caixeta, Filho. 1998. ``Ferronorte e Transporte de GraÄos.’ ’ In Precos Agricolas June Brazil: Escola Superior de Agricultura ``Luis de Queiroz.’ ’ Deacon, R. 1999. ``Deforestation and Ownership: Evidence from Historical Accounts and Contemporary Data.’ ’ Land Economics 75 (Aug.): 341± 59. Deaton, A., and J. Muellbauer. 1980. ``An Almost Ideal Demand System.’ ’ American Economic Review 70 (3): 312± 26. FundacË aÄo Getulio Vargas (FGV) 1998 PrecËos de Terra (ARIES online database: www.fgv.br/ cgi-win/aries.exe). Rio de Janeiro: FundacË aÄo Getulio Vargas. Hasenkamp, George 1976. ``Speci® cation and Estimation of Multiple-Output Production Functions.’ ’ Lecture Notes in Economics and Math-

240

Land Economics

ematical Systems. No. 120. Berlin: SpringerVerlag. Instituto Brasileiro de Geogra® a e Estatistica (IBGE) 1997a. Matriz de Insumo-Produto Brasil 1995. Rio de Janeiro: Instituto Brasileiro de Geogra® a e Estatistica. IBGE 1997b. Sistema de Contas Nacionais Brasil 1990± 1995/ 96. Rio de Janeiro: Instituto Brasileiro de Geogra® a e Estatistica. IBGE 1998. Censo AgropecuaÂrio 1995/ 1996. Rio de Janeiro: Instituto Brasileiro de Geogra® a e Estatistica. Kaimowitz, D., and A. Angelsen. 1998. Economic Models of Tropical Deforestation: A Review. Bogor, Indonesia: Center for International Forestry Research. Mattos, M. M., and C. Uhl. 1994. ``Economic and Ecological Perspectives on Ranching in the Eastern Amazon.’ ’ World Development 22 (2): 145± 58. Mendelsohn, R. 1994. ``Property Rights and Tropical Deforestation.’ ’ Oxford Economics Papers 46:750± 56. Najberg, S, F. Rigolon, and S. Vieira. 1995. ``Modelo de Equilibrio Geral Computavel Como Instrumento de Politica Economica: Uma Analise de Cambio e Tarifas.’ ’ Texto para Discussao n. 30. Rio de Janeiro: Banco Nacional de Desenvolvimento Economico e Social. Persson, A., and M. Munasinghe. 1995. ``Natural Resource Management and Economywide Policies in Costa Rica: A Computable General Equilibrium (CGE) Modeling Approach.’ ’ The World Bank Economic Review 9 (2): 259± 85. Pfaff, A. S. 1997. ``What Drives Deforestation in the Brazilian Amazon? Evidence from Satel-

May 2001

lite and Socioeconomic Data.’ ’ Policy Research Working Paper No. 1772. Washington, D.C: World Bank. SIFRECA. 1998. ``Sistema de InformacËoÄes de Fretes para Cargas Agrõ Â colas: Soja (9/ 97).’ ’ Piraacicaba-SP, Brazil: Escola Superior de Agricultura ``Luiz de Queiroz’ ’ (ESALQ). Southworth, F., V. H. Dale, and R.V. O’ Neill. 1991. ``Contrasting Patterns of Land Use in RondoÃnia, Brazil: Simulating the Effects on Carbon Release.’ ’ International Social Sciences Journal 130:681± 98. Stone, S. W. 1998. ``Evolution of the Timber Industry along an Aging Frontier: The Case of Paragominas (1990± 95).’ ’ World Development 26 (3): 433± 48. Toniolo, A., and C. Uhl. 1995. ``Economic and Ecological Perspectives on Agriculture in the Eastern Amazon.’ ’ World Development 23 (6): 959± 73. Uhl, C., and I. C. G. Vieira. 1989. ``Ecological Impacts of Selective Logging in the Brazilian Amazon: A Case Study from the Paragominas Region in the State of Para.’ ’ Biotropica 21: 98± 106. Van Loock, H. J., W. L. Ha¯ ey, and R. A. King. 1973. ``Estimation of AgricultureÐ Forestry Transition Matrices From Aerial Photographs.’ ’ Southern Journal of Agricultural Economics (Dec.): 147± 53. Vosti, S., J. Witcover, and C. Line Carpentier. Agricultural Intensi® cation by 2000. Smallholders in the Western Brazilian Amazon: From Deforestation to Sustainable Land Use. Draft Research Report. Washington, D.C.: International Food Policy Research Institute (IFPRI).

Land Use Impacts of Agricultural Intensi® cation and Fuelwood Taxation in Uganda Bernard Bashaasha, David S. Kraybill, and Douglas D. Southgate ABSTRACT. The market and land-use impacts of agricultural intensi® cation and fuelwood taxation in Uganda have been estimated using a computable general equilibrium model. Across-theboard yield growth causes total agricultural output to rise and commodity prices to fall. Forest area increases, as does land planted to cash crops. But other uses of rural natural resources decline, including the area planted to food crops. Taxing raw material inputs to fuelwood production leads to an increase in price and a decrease in consumption and production. However, forest area does not expand. Instead, it contracts a little. (JEL Q11, Q23)

I. INTRODUCTION As in many other parts of the developing world, deforestation in Uganda has reached an advanced cumulative stage. From 7.10 million hectares one hundred years ago, tree cover has declined markedlyÐ to 1.35 million hectares (World Bank 1993), or 7% of the national territory, as of the early 1990s. Habitat loss has coincided with agriculture’ s geographic expansion as well as depletive fuelwood extraction. The total area planted to crops rose from 3.25 million hectares in 1980 to 4.60 million hectares in 1993 (World Bank 1993). At present, annual consumption of fuelwood, which is favored by the vast majority of urban and rural households as well as industrial enterprises, amounts to 18.3 million cubic meters. This exceeds sustainable output, which is estimated to be 15.6 million cubic meters, by 20% (MNR 1994). Agriculture’ s geographic expansion and increasing fuelwood collection are both linked to population growth, which is currently averaging 2.8% per annum in Uganda. The demographic impulse for land use conversion, in particular, is strong because agLand Economics · May 2001 · 77 (2): 241± 249 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

ricultural yields are low. Because investment needed to boost productivity has long been lacking, wide gaps exist between output-perhectare on experimental plots (a measure of potential yields) and what the average Ugandan farmer actually harvests. To cite three examples, actual yields as a percentage of potential levels are 51% for maize, 68% for soybeans, and 55% for sun¯ owers (OSU, 1993). With productivity low, the agricultural sector has been obliged to meet burgeoning commodity demands almost exclusively by using more land for farming and ranching. During the past ten years, yield-enhancing investment has started to occur. For example, the National Agricultural Research Organization (NARO) was created in 1992 and its budget has grown substantially, from $7.27 million in 1993/94 to $15.48 million in 1996/97. However, the increased productivity that improved science and technology make possible might actually accelerate agricultural land clearing. Among others, Kaimowitz and Angelsen (1998) contend that the land use impacts of agricultural intensi® cation are ambiguous. It is to be expected, for example, that higher yields will result in deforestation if commodity demand is priceelastic. This would certainly be the case for a small nation that sells much of its output in global markets in which it is a price-taker. But if demand is inelastic, as it is for many food items, some landowners are bound to The authors are, respectively, acting chairman, Department of Agricultural Economics, Makerere University, Kampala, Uganda, and associate professor and professor, Department of Agricultural, Environmental, and Development Economics, The Ohio State University. Research funding provided by the Rockefeller Foundation. Salaries and research support provided by state and federal funds appropriated to the Ohio Agricultural Research and Development Center, The Ohio State University.

242

Land Economics

® nd their earnings adversely affected by technological change, the revenue losses associated with lower prices outweighing what is gained in the form of lower costs and higher output. Over the long run, these farmers switch real estate and other assets out of agriculture. Southgate (1998) points out that this consequence of intensi® cation tends to be especially pronounced near agriculture’ s extensive margins. Regression analyses carried out by Barbier and Burgess (1997) indicate that, as a rule, the relationship between yields and forest loss is, indeed, negative. By no means is deforestation in the face of inelastic commodity demand and agricultural intensi® cation inconceivable. If property rights are attenuated, individuals take advantage of any and every opportunity to seize resources by occupying or harvesting them. Agricultural colonization of unclaimed hinterlands is a case in point. Depletive harvesting of forest products is another. In Uganda, for example, there are no effective restrictions on the harvesting of fuelwood in what are supposedly state forests, over which the government has no effective control. Accordingly, the rent dissipation and resource overexploitation characteristic of open access are prominently on display. To say the least, then, arresting deforestation in a place like Uganda is a major challenge. This paper addresses two possible remedies. One is agricultural intensi® cation, resulting from improved support for agricultural research and extension. The other is to impose a tax, re¯ ecting resource replacement costs, on the raw material inputs to fuelwood production. Since most Ugandan households are poor and therefore spend large shares of their earnings on food and fuel, either policy initiative is bound to have signi® cant economy-wide impacts. Accordingly, a computable general equilibrium (CGE) model has been used to analyze the effects of improved agricultural technology and fuelwood taxation. II. THE COMPUTABLE GENERAL EQUILIBRIUM MODEL Our CGE model, described in detail in Bashaasha (1998), computes a long-run equilibrium based on the assumption that re-

May 2001

sources, including land, are allocated among different sectors of the economy so as to maximize net social returns. Comparison of different long-run equilibria, each corresponding to a particular policy scenario, does not allow the possibility of path-dependency of adjustment from one equilibrium to another to be investigated. For example, our model cannot be used to determine whether the ultimate land-use impacts of, say, higher agricultural yields might be affected by the speci® c sequence of responses that households and farmers make to this change. Like many others, this model has no spatial dimension. Only the aggregate land areas allocated to different uses are examined. In addition, we assume that the aggregate supply of land is ® xed and its allocation among different sectors is determined by relative net returns. For example, an increase in agricultural rents, resulting perhaps from higher commodity prices, causes farming to take place where forests would otherwise stand, all else remaining the same of course. Treating the Ugandan economy’ s land endowment as ® xed is reasonable. No part of the country, other than a few well-guarded parks, is uninhabited; economic activityÐ at the very least, herding and fuelwood collectionÐ occurs everywhere. In such a setting, expanding the area where the ``human footprint’ ’ cannot be detected is not a viable option. Instead, environmental goals must be achieved by promoting economic land uses, like forestry, that yield more environmental services than do others, such as agriculture. True, this approach rarely yields instantaneous results. For example, reforestation of land where agriculture is no longer pro® table takes time. This sort of lag is not captured, of course, in our CGE model. The structure of the model relates to the focus on land use, especially the extent of forests. In particular, the Ugandan economy is divided into six sectors. One, which occupies a small part of the national territory, encompasses manufacturing and services, including services provided by government. In addition, there are separate forestry and fuelwood sectors, the latter using negligible amounts of land and drawing raw materials from the former. The remaining sectors are cash crops (including coffee and other ex-

77(2)

Bashaasha, Kraybill, and Southgate: Policy and Land Use in Uganda

ports), food crops, and other agriculture (mainly livestock). This disaggregation of the national economy, generally, and its rural parts, speci® cally, is needed to investigate how policy changes, which affect living standards, affect consumption patterns and ultimately land use. Households, which supply factors of production and receive income in return, are disaggregated into three categories: rural, urban poor, and urban af¯ uent. In addition to consuming goods and services, households save a portion of their earnings. All these savings are used to purchase investment goods. Government, which collects taxes from industries and households, spends its revenues on goods and services. It is assumed that government savings are ® xed at the level recorded in the base year. Production technology is speci® ed as a nested function that combines primary as well as intermediate inputs. To be speci® c, a primary input aggregate, which is a CobbDouglas function of capital, land, and labor, is combined with intermediate inputs in ® xed proportions to produce output. Having convex isoquants and being a good approximation of production processes for which factors are imperfect substitutes, the CobbDouglas function can be calibrated using information contained in a social accounting matrix (SAM). This formulation is appropriate for this study since the data needed for econometric estimation of other functional forms are not currently available. In this model, trade ¯ ows respond to endogenous changes within Uganda as compared to the rest of a world. Since the country is a very small player in the global economy, international prices for all traded commodities are ® xed. As in other CGE models, the Armington (1969) assumptionÐ that imports and domestic goods are imperfect substitutesÐ is adopted. Likewise, export sales and domestic sales are also treated as imperfect substitutes. Foreign savings (borrowing) are ® xed and the foreign sector is brought into equilibrium through movements in a SwanSalter real exchange rate, which is a ratio of prices for traded and non-traded goods (Salter, 1959). The model is calibrated to a SAM prepared using data obtained from the Statistics

243

Department (SD) of the Ministry of Finance and Economic Planning (MFEP). The six production sectors were aggregated from the SD’ s 1992 input-output (I-O) table, which features thirty sectors. The single household sector in the I-O table was disaggregated into three household groups using expenditure and income shares based on two SD household surveys, one carried out in 1990 and the other in 1992 and 1993. Along with the SD’ s National Income and Product Accounts (NIPAs), these same household surveys were used to ® ll in the ® nal payments quadrant, which is typically left blank in an I-O table, to transform it into a complete SAM. To ensure accounting consistency in the data, I-O row and column totals were maintained and fourth-quadrant values in the SAM were adjusted to be consistent with these totals. The category ``operating surplus’ ’ in the input-output table expresses the returns to entrepreneurship, management, and natural resources. Land was isolated from this category using World Bank (1993) and Ministry of Agriculture, Animal Industry and Fisheries (MAAIF) estimates. The basic source of data for the fuelwood account was MFEP (1993), although personnel of the Forestry Department of the Ministry of Natural Resources (MNR) and more than a dozen landowners and private producers of fuelwood were interviewed to verify the accuracy of information about sales and expenditures. Additional data on biomass energy and insights about forestry and fuelwood were obtained from ESD (1996) and the National Environmental Management Authority (NEMA). Elasticities of trade substitution between domestic and foreign goods and services do not currently exist for Uganda and could not be estimated due to a lack of data. The trade elasticities are borrowed from other studies, including one carried out in Kenya by Tyler and Akinboade (1992) as well as research in Cameroon by Devarajan, Lewis, and Robinson (1991). Sensitivity analyses showed that the general equilibrium results of the model are fairly stable over a wide range of trade elasticities since Uganda imports and exports relatively little food. Data on agricultural research spending and information on technology development and dissemination strategies were obtained

244

Land Economics

from NARO and annual reports of af® liated institutions. The MNR’ s Forestry Department was the source of data on forestry production. A national biomass study (ESD 1996) contained useful information on biomass energy in general, and fuelwood supplies in particular. Additional information and insights were obtained from discussions with the following institutions: the Agricultural Secretariat of Bank of Uganda, the Economic Policy Research Center (EPRC), the World Bank country of® ce in Uganda, and MAAIF. III. RESULTS AND DISCUSSION Using the CGE model, we have examined the market and land-use impacts resulting both from improved agricultural productivity, which is made possible by recent increases in support for agricultural research and extension, and from a tax on the raw material inputs to fuelwood production. This tax equals resource replacement costsÐ to be speci® c, the expense of producing raw materials in a plantation setting. In particular, two scenarios relating to Hicks-neutral technological change are investigated. The ® rst corresponds to a 3%, one-time productivity increase, as modeled by an upward adjustment in the intercept of the Cobb-Douglas function. The second scenario introduces a 30%, one-time increase of the same intercept. Ugandan production of cash and food crops is currently growing at around 3% annually, so the ® rst scenario is a very conservative approximation of the productivity improvements that the country’ s farmers and ranchers can achieve. If 3% annual productivity growth were sustained for a little less than ten years, the overall improvement would be 30%, which is what is supposed in the second scenario. In a series of sub-scenarios, productivity growth of 3 or 30% is investigated separately for each of the three parts of the agricultural economy: cash crops, food crops, and other (mainly livestock, as indicated above). The price currently paid for raw material inputs to fuelwood production, which currently averages Ush 6,818 per ton, only re¯ ects private logging and hauling expenses.

May 2001

None of the costs associated with forest resource depletion are internalized. The best available measure of social costÐ private extraction expenditures plus resource depletion costsÐ is the expense of producing wood on a plantation: Ush 16,168 per ton (Bashaasha 1998; ESD1996). To examine the impacts of internalizing resource values, we run the CGE model assuming that a tax equal to the difference between social cost and private expensesÐ Ush 9,350 per tonÐ is applied, with all the revenues obtained spent on the forestry sector (i.e., the source of raw materials). Many of the simulations carried out with the CGE model con® rm a priori notions about what is required to change land use and to improve economic performance in the countryside. This is certainly true of many of the impacts of higher agricultural productivity. However, some of our ® ndings are counter-intuitive. For example, fuelwood taxation, which is supposed to end the neglect of resource user costs, actually ends up leading to a decline in forest area. The Impacts of Agricultural Productivity Gains

As is to be expected, technologically neutral improvements in agricultural productivity lead to lower commodity prices as well as increased crop and livestock output. The consequences for land use, in particular the area devoted to farming and ranching versus the area where trees and other natural vegetation remain in place, depend on demand elasticities. If, for example, purchases of a crop are highly sensitive to price changes, then yield growth causes the area planted to that crop to increase. By contrast, planted area declines because of improved productivity if commodity demand is price-inelastic. Results reported in Table 1 show that an increase in productivity in all agricultural sectors leads to lower agricultural prices. This lowers the relative return to factors of production employed in agriculture and induces the reallocation of factors to other sectors of the economy. The decline in relative factor returns in agriculture due to technological change and the freeing of resources for use in other sectors is consistent with observed

77(2)

Bashaasha, Kraybill, and Southgate: Policy and Land Use in Uganda

TABLE 1 P rice, Output, and Land Use Changes Induced by an Across-the-Board Increase in Agricultural P roductivity With 3% Productivity Shift

With 30% Productivity Shift

20.3 21.6 21.5

22.3 212.0 211.1

5.1 2.9 2.8

52.5 28.9 27.6

3.0 20.2 20.3 0.1 0.1

26.0 21.3 22.5 1.0 0.6

Relative Change (%) in: Producer Prices Cash Crops Food Crops Other Agriculture Output Cash Crops Food Crops Other Agriculture Land Use Cash Crops Food Crops Other Agriculture Forestry Manufacturing & Services

long run trends in many countries. The resources freed from agriculture are shifted to other sectors, and incomes rise as a result. The importance of demand elasticity is revealed in an assessment of the impacts of uniform and technologically neutral productivity gains throughout the agricultural economy. As indicated in Table 1, output rises across the board and there is a general decline in prices. Since practically all food crops and livestock products are consumed domestically, demand for those goods is much less price elastic than is export demand for cash crops. Accordingly, supply increases associated with technological progress have an appreciable effect on prices in the former sectors. Of course, price reductions enhance purchasing power, and therefore cause consumption to rise throughout the economy. In contrast, output from the cash crop sector is destined for foreign markets, in which changes in the volume of coffee, tea, and other Ugandan exports have no signi® cant effect on world prices. Accordingly, techno-

245

logical change at the national level has only a very small effect on prices. Changes in inter-sectoral resource allocation also re¯ ect differences in demand elasticity. Less land is used to raise domestically consumed food crops and livestock products. Meanwhile, the area planted to cash crops increases dramatically, especially for the case of a 30% productivity gain. Perhaps the most signi® cant result, though, is that tree-covered area does not decline; to the contrary, there is actually a small increase. General technological improvement, then, has a number of desirable consequences. Consumers bene® t because the food items they purchase are cheaper and more readily available. Production of coffee, tea, and other commodities in which Uganda holds a comparative advantage expands dramatically. Since those commodities’ prices are not much affected, export earnings likewise experience signi® cant growth. Finally, all of this is accomplished without deforestation. If productivity gains are not experienced in the entire agricultural economy, the impacts may not be universally bene® cial. Consider the case of technologically neutral improvement in the cash crop sector, alone. As indicated in Table 2, change of this sort would lead to sizable increases in cash crop production as well as in the resources devoted to that activity. Forest area would not be affected. However, expansion of the cash crop sector would be accomplished at the expense of production and land use in the rest of the agricultural economy. Consequently, domestic prices of food items would rise. Since formal mechanisms for income transfer are very rudimentary in places like Uganda, the adverse welfare impacts associated with this change could easily outweigh the bene® ts of increased exports. There are other negative consequences of productivity-enhancing investment targeted exclusively on the cash crop sector. For example, international commodity prices ¯ uctuate, which leads to corresponding variation in the fortunes of an economy, like Uganda’ s, that specializes in the production and export of one or a few commodities. This sort of dif® culty, which this study does not address, is stressed by those concerned about

Land Economics

246

TABLE 2 P rice, Output, and Land Use Changes Induced by Increased P roductivity in the Cash Crop Sector Alone With 3% Productivity Shift

TABLE 3 P rice, Output, and Land Use Changes Induced by Increased P roductivity in the Food Crop Sector Alone

With 30% Productivity Shift

Relative Change (%) in:

Relative Change (%) in:

Producer Prices

Producer Prices

Cash Crops Food Crops Other Agriculture

20.4 1.1 1.1

22.4 11.4 11.1

Output Cash Crops Food Crops Other Agriculture

Cash Crops Food Crops Other Agriculture

With 3% Productivity Shift

With 30% Productivity Shift

0.0 22.7 0.2

0.4 220.1 2.3

20.5 3.0 0.0

25.0 29.6 0.0

20.7 20.1 0.0 0.1 0.2

27.0 20.5 0.1 0.6 1.3

Output 5.8 20.1 20.1

61.1 20.5 20.5

Land Use Cash Crops Food Crops Other Agriculture Forestry Manufacturing & Services

May 2001

Cash Crops Food Crops Other Agriculture Land Use

3.9 20.1 20.1 0.0 20.2

36.1 20.8 20.9 20.2 21.5

food security and economic development. They would advocate a different sort of sectoral preference, one emphasizing productivity gains in food crops over gains related to exportable commodities. However, this strategy also involves sizable welfare losses and does not save any additional forests. Reported in Table 3 are the results of simulations of technological improvement in the food crop sector alone. As one should expect, production in that sector goes up dramatically and there is a signi® cant decline in prices. There is a modest reduction in sectoral land use as the returns to food crop production decline relative to returns in other sectors. Land use and output declines in the cash crop sector. Also, forestry and manufacturing and services experience geographic expansion. However, the increase in forest area is no larger than what is observed for across-the-board agricultural productivity improvement (Table 1). Thus, the habitat conservation bene® ts of a ``food ® rst’ ’ approach to agricultural development are no greater than those of the non-

Cash Crops Food Crops Other Agriculture Forestry Manufacturing & Services

targeted strategy, which has the additional merit of resulting in increased output of commodities in which Uganda holds a comparative advantage. Bashaasha (1998) has simulated the effects of other improvements in agricultural productivity. In general, his ® ndings are consistent with those presented here. The Consequences of Taxing Fuelwood

In this scenario, a tax of Ush 9,350 is collected on every ton of raw materials used to produce fuelwood and tax revenues are spent on investment goods (domestic and imported manufactures) destined for the forestry sector. Since the model is static, this allocation of revenues leads to an increase only in ® nal demand and not in productive capacity of the forestry sector. The simulated impacts of this policy change are at odds with cursory thinking on the subject. In particular, taxing raw material inputs to fuelwood production to internalize resource values does not lead to an

77(2)

Bashaasha, Kraybill, and Southgate: Policy and Land Use in Uganda

increase in tree-covered area. Rather, it induces a modest decline. Two market realities underlie this result. First, as the price of fuelwood rises, consumption declines. In part this re¯ ects the diminished purchasing power of households. But there is also a substitution effect, with other goods and services now being relatively cheaper. Second, substitution occurs among inputs to fuelwood production. Higher raw material prices, which are a direct consequence of the tax, induce more use of labor and other non-resource inputs. This raises conversion ef® ciencies in, for example, the production of charcoal, which is the preferred fuel in urban areas. The changes in price and production caused by fuelwood taxation re¯ ect the differences between demand and supply elasticities. In the CGE simulation, the market value of fuelwood increases by 22%, which is much less than the relative difference between the tax on raw material inputs (Ush 9,350 per ton) and price before introduction of the tax (Ush 6,818 per ton). The change in equilibrium output is smaller: 14%. Forest area declines under this scenario because substitution from raw materials to other inputs occurs in the fuelwood sector. With usage of forest products diminished, a small amount of land ends up being diverted to the cash crop sector (Table 4). Bashaasha (1998) has investigated other fuelwood taxation possibilities. Among other things, he ® nds that the transfer of land from forestry to cash crops is larger if revenues generated by the tax are spent on activities, like techTABLE 4 P rice, Output, and Land Use Changes Resulting from a Tax on Fuelwood Consumption Relative Change (%) Fuelwood Producer Prices Fuelwood Output Land Use Cash Crops Forestry Other Sectors

22.0 213.5 0.3 20.4 0.0

247

nology development and transfer, that make the forestry sector more productive. The lesson to be drawn is straightforward. In that sector, as in other parts of the rural economy, lower output prices combined with enhanced productivity result in diminished land use. Sensitivity Analysis

We conducted sensitivity analysis to determine the effects of trade elasticities on model results, focusing on across-the-board increase in agricultural productivity. The model was re-run several times with trade elasticities that were half or twice the original values. Output and land use results changed signi® cantly only in the cash crop sector, and even there the changes were modest. Results in other sectors differed little. Under a 3% agricultural productivity increase, cash crop output increases by 5.1% when the original (central) values of all trade elasticities are used. In contrast, cash crop output increases by 5.0% when trade elasticities are halved and by 5.8% when they are doubled. Land use in the cash crop sector, which increases by 3.0% when the original (central) trade elasticity values are used, increases by 2.8% when all trade elasticties are halved and by 4.0% when they are doubled. The model also utilizes factor substitution elasticities and consumption substitution elasticities, and these could affect model results as well. We do not conduct sensitivity analysis on these elasticities, however, since by choosing the Cobb-Douglas functional form, the elasticities of substitution in production and consumption are arbitrarily set at one. While unitary elasticitiesÐ implying major responses by producers and consumers to price changesÐ may be unrealistically high for some developing countries, we believe they are reasonable for Uganda, which has implemented impressive market reforms over the past decade. IV. CONCLUSIONS Among agricultural development specialists as well as environmentalists intent on conserving biodiverse habitats, there is a general consensus that ® nding better ways to produce crops and livestock helps to curb

248

Land Economics

deforestation in Africa, Asia, and Latin America. Although quite a lot of evidence exists to support this point of view, the relationship between agricultural productivity growth and habitat conservation turns out to be more complex and ambiguous. Having carried out a thorough review of the existing literature, Kaimowitz and Angelsen (1998) conclude that there are various circumstances under which agricultural development actually accelerates deforestation. The study presented in this paper corroborates the optimistic view that economic progress in the countryside and habitat conservation tend to be complementary while simultaneously validating some of what is said by those with a more skeptical perspective. An especially encouraging result is that across-the-board productivity gains in the agricultural economy lead to simultaneous increases in cash crops sold to foreigners, food items consumed domestically, and the area covered with natural vegetation. Almost certainly, the additional welfare created by this balanced approach to agricultural development far exceeds what can be gained by stressing technological improvement in a particular sector, be it export or food crops. Our ® ndings con® rm that productivity gains can, indeed, cause forest loss. Raising productivity in the export sector, in which demand elasticity is high, can easily lead to an expansion of cropped area, perhaps at the expense of natural habitats. By contrast, our ® ndings suggest that the improvement in living standards resulting from increased yields and lower prices of food crops causes households to alter consumption patterns. Rather than using all additional purchasing power to buy more food, they increase their spending on other things, such as energy and manufactured items. Land use expands in the sectors that produce the latter goods and services. It is also reasonable to conclude that raising agricultural productivity is a more ef® cacious approach to habitat conservation than using ® scal instruments to resolve the problem of neglected resource values. The limited effectiveness of fuelwood taxation ought not to disappoint the Ugandan government. Soon after coming to power, the National Resistance Movement (NRM) imposed

May 2001

a charcoal tax, which can be interpreted as an attempt to force consumers to internalize the user costs associated with forest resource depletion. However, the tax proved to be highly unpopular, provoking public demonstrations that led to its being withdrawn. Our ® ndings suggest that this will not greatly accelerate deforestation in Uganda. As can be said of any study, the research described in this article has its limitations. Some, like neglect of the stochastic nature of agricultural income, have been mentioned already. Also, data limitations are bound to be severe in an impoverished nation like Uganda, where little funding is available for collecting the sort of information needed to elaborate a CGE model, to estimate the replacement cost of forest resources, and so forth. Uganda emerged from economic chaos barely ten years ago, which means that only now are opportunities to estimate demand elasticities and other model parameters beginning to present themselves. In future research, it would be appropriate to use a less aggregated model, one that lends itself to the analysis of speci® c changes in policies and institutions. Another alternative, which should be considered as the necessary data become available, would be to elaborate a dynamic model, which can be used to examine the path-dependency of changes from one economic equilibrium to another. Nevertheless, studies of the kind we have carried out are useful because they yield important ® ndings that could not be obtained in a partial equilibrium analytical framework. By assessing economy-wide impacts of changes in policy and technology, we hope to have shed light on effective strategies for achieving economic progress and habitat conservation in the countryside. Certainly, our ® ndings add to the empirical base for distinguishing between the circumstances under which agricultural development arrests deforestation and the conditions under which it does not. References Armington, P. 1969. ``A Theory of Demand for Products Distinguished by Place of Production.’ ’ IMF Staff Papers 16 (1): 159± 78.

77(2)

Bashaasha, Kraybill, and Southgate: Policy and Land Use in Uganda

Barbier, E., and J. Burgess. 1997. ``The Economics of Tropical Forest Land Use.’ ’ Land Economics 73 (May): 174± 95. Bashaasha, B. 1998. ``Public Policy and Rural Land Use in Uganda.’ ’ Ph.D. diss., Ohio State University. Devarajan, S., J. Lewis, and S. Robinson. 1991. ``From Stylized to Applied Models: Building Multisector CGE Models for Policy Analysis.’ ’ Working Paper No. 616. Department of Agricultural and Resource Economics, University of California, Berkeley. Energy for Sustainable Development (ESD). 1996. ``A Study of Woody Biomass Derived Energy Supplies in Uganda’ ’ (Report to MNR). Kaimowitz, D., and A. Angelsen. 1998. Economic Models of Tropical Deforestation: A Review. Bogor, Indonesia: Center for International Forestry Research. Ministry of Finance and Economic Planning

249

(MFEP). 1993. Background to the Budget, 1993± 94. Entebbe, Uganda. Ministry of Natural Resources (MNR). 1994. State of the Environment Report for Uganda. Kampala, Uganda. Ohio State University (OSU). 1993. ``Manpower for Agricultural Development Project.’ ’ (Report to USAID). Columbus, Ohio. Salter, W. 1959. ``Internal and External Balance: The Role of Prices and Expenditure Effects.’ ’ Economic Record 35 (Aug.): 226± 38. Southgate, D. 1998. Tropical Forest Conservation: An Economic Assessment of the Alternatives in Latin America. New York: Oxford University Press. Tyler, G., andO. Akinboade. 1992. ``Structural Adjustment and Poverty: A Computable General Equilibrium Model of the Kenyan Economy.’ ’ Oxford Agrarian Studies 20 (1): 283± 89. World Bank. 1993. ``Uganda: Agriculture’ ’ (Country Study). Washington, D.C.: World Bank.

How Do National Markets and Price Policies Affect Land Use at the Forest Margin? Evidence from the Philippines Ian Coxhead, Agnes Rola, and Kwansoo Kim ABSTRACT. Agricultural growth in uplands of tropical developing countries is associated with deforestation, land degradation, and diminished watershed function. Using time-series price data from an upland Philippine watershed, we examine market integration and quantify product market links through which policy and macroeconomic shocksÐ including instability from the Asian ® nancial crisis of 1997± 98Ð are transmitted to farm gate prices. If market-driven incentives dominate farmers’ decisions, then our results indicate the desirability of using a broader range of policy instruments to address upland problems, and the need for upland projects to devote increased attention to national-level information dissemination and policy advocacy. ( JEL Q11, Q23)

I. INTRODUCTION Poor farmers in developing countries are the primary managers of an increasingly scarce natural resource, productive agricultural land. Their decisions, while they may be privately optimal, often con¯ ict with social goals of resource conservation. This is clearly true when farmers intensify production on soils that are easily eroded, and when agricultural expansion takes place through the conversion of forests and other permanent cover to seasonal crops. The empirical literature on tropical deforestation and land degradation is rich with studies of resource use by households whose actions are constrained by poverty, market failures, and risk aversion (e.g., Anderson and Thampapillai 1990; Southgate 1988; Shively 1997). The literature typically locates such immediate motivational factors Land Economics · May 2001 · 77 (2): 250± 267 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

within a broader context of absence or nonenforcement of property rights (resulting in open-access forest lands and tenure insecurity on farmed lands), and population pressure. These are identi® ed as providing the enabling environment for forest clearing and unsustainable patterns of agricultural land use by upland farmers (for an excellent survey of technical and economic issues within this tradition see Pingali 1997). There are also a number of analytical models exploring the in¯ uence of broader economic forces like price policies and wage trends on elements of the upland agricultural decision set, such as soil conservation (Barbier 1990; Barrett 1991) and deforestation (Angelsen 1999). At the broadest level are general equilibrium papers in which intersectoral linkages, through factor markets, product markets and trade, are seen to in¯ uence upland decisions (Lopez and Niklitschek 1991; Deacon 1995; Coxhead and Jayasuriya 1994, 1995). Looking across all the types of models one ® nds a wide array of assumptions about the economic links between upland economies and the national economies in which they are located (Angelsen 1999, in particular, explores many variations). The choice of marThe authors are, respectively, associate professor, Department of Agricultural and Applied Economics, University of Wisconsin± Madison; professor, Institute for Strategic Planning and Policy Studies, University of the Philippines± Los BanÄos College, Laguna, Philippines; and research associate, Department of Agricultural and Applied Economics, University of Wisconsin± Madison. The authors thank two anonymous referees and Sisira Jayasuriya for helpful comments on earlier drafts. For data collection we are grateful to Isidra Balansag and the Philippine Bureau of Agricultural Statistics. Financial support for this research was provided by the University of Wisconsin Graduate School, and by USAID through the SANREM CRSP.

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

ket assumptions conditions the behavior of a model and thus the policy conclusions that are drawn from it. As an example, general equilibrium approaches to deforestation, by allowing for intersectoral labor mobility, conveys the idea that upland population ``pressure’ ’ is a response to economic incentives, rather than an exogenous determinant of actions as in some of the other models. Similarly, there is a great deal of variation in the assumptions commonly made about product markets. Given the importance of a correct speci® cation of market structure and pricing, there are surprisingly few studies that bring empirical evidence to bear on the market and policy aspects of upland agricultural resource use decisions. The goal of this paper is to encourage a move in that direction, as a complement to the existing body of household-level analyses. It is our thesis that the design of upland projects directed at in¯ uencing smallholders’ land conversion and land use decisions in the direction of ``sustainability’ ’ could be greatly improved by a better integration of information on market- and sector-level incentives with information on household-level decisions and constraints. Perhaps because of a lack of data and empirical analysis, project solutions to deforestation and agricultural land degradation in developing countries focus mainly on direct interventions through technology transfer, institutional innovations and other household-level actions. The role of policy, (and especially its less direct manifestations through intersectoral product and factor markets) is generally given little emphasis.1 The obverse of this problem is a general neglect at the policy level of the intersectoral and environmental impacts of trade and agricultural pricing policies.2 Both forms of myopia may have restricted the domain of possible project solutions to upland environmental problems, and indeed may have increased the probability that projects will fail because of con¯ icting messages contained in the direct and intersectoral signals from economic policies. To illustrate this point, consider an upland economy producing two goods, one produced using a land-intensive technology and

251

the other using a labor-intensive technology.3 De® ne the relative price of the latter to the former as P, and price of the lowland good relative to the upland land-intensive good by Q. There are thus three product markets; two for upland goods and one for an aggregate lowland good. Assume that upland production uses land and labor, and that in order to be brought into production, land must ® rst be cleared of forest, an activity that uses labor. Pro® t-maximizing upland producers will thus allocate labor to forest clearing and farming, and to one crop or the other, in response to P. If there are links to the lowland economy, then Q will also play a role in these decisions. Suppose ® rst that all three goods are freely traded with the rest of the world at given world prices, and that the world price of Q increases. Whether this change has any effect in uplands, and if so in what manner, will depend on interregional labor markets. If labor is immobile between regions, the increase in Q will have no effect. If it is mobile, the increase will raise labor productivity in lowlands relative to uplands and induce outmigration. Since labor is needed for forest 1 In the Philippine case, a recent set of national government guidelines for watershed management (PCARRD 1999) makes only incidental reference to markets as in¯ uences on farmer behavior, and none to policies other than those which have direct effects on land useÐ zoning, tenure laws, and similar. 2 The following passage from a former undersecretary for policy and planning in the Philippine Department of Agriculture illustrates:

Policymakers in the Philippines tend to examine economic problems from the perspective of individual consumers and ® rms, and thus, generate and propose actions and measures focused on directly supporting these entities. In no way have economic policies been evaluated on the basis of their environmental impacts. In rare cases, farmer interests are accounted for. For instance, price controls [on rice and corn] were defended on the basis of their effects on the consumers of staple commodities and the costs of raw materials to enterprises. Rarely were the adverse effects on supply responses as well as the welfare of producersÐ particularly of farmers and ® shermenÐ considered. (Tolentino 1995). 3 For formal developments of the model in this section, see Lopez and Niklitschek 1991; Deacon 1995; and Coxhead and Jayasuriya 2000.

252

Land Economics

clearing as well as farming, deforestation must decline. However, now suppose that the labor-intensive good in uplands is not traded, so its price depends on domestic demand and supply; thus P 5 P(Q). Now an increase in Q will have different effects on uplands depending on whether the goods are substitutes in consumption as well as supply side factors and income effects. Similarly, imagine that Q is constant but that technical progress occurs in one upland crop. If products are freely traded at world prices, upland labor productivity will rise and both the share of land and the total area planted to the crop experiencing technical progress will increase; deforestation will rise in this case. On the other hand, if demand for the crop is downward-sloping (whether due to local or national non-tradability), then technical progress will alter P and the total area of upland, as well as the share planted to the labor-intensive crop, could rise or fall. These simple illustrations highlight the sensitivity of deforestation and upland land use outcomes to market conditions. Under one set of assumptions, technical progress in an upland crop is predicted to increase deforestation; under another, deforestation could fall. Conversely, a national policy innovation that alters P or Q (or both) has the potential to induce changes in deforestation and land use even when the policy measure is not directly related to agriculture. This is so even when all goods’ prices are exogenous, if labor is mobile between regions. Lastly, when upland farmers are risk-averse, the entire argument can be restated (with modi® cations as appropriate) using price variances as well as levels. In the rest of the paper we focus on a Philippine case study. We ® rst provide a brief survey of major macroeconomic and policy trends and their possible effects on upland resource use decisions. While we have information about macroeconomic and economywide phenomena, and about upland farmers’ decision-making processes, we know little about the nature and strength of market links between the two. We then use econometric analysis to examine linkages between national and farm-gate prices on the basis of re-

May 2001

cent data from a typical upland watershed in the southern island of Mindanao. II. GROWTH, POLICIES, AND UPLAND RESOURCE USE IN THE PHILIPPINES The pace of aggregate economic growth in the Philippines has accelerated in recent years, but the degree of dependence on agriculture and natural resources remains high by Southeast Asian regional standards. This is a function of earlier decades of slow growth and rapid population increase, which maintained a high level of dependence on agriculture. It can thus be argued that the persistence of pressure on forest and upland agricultural land resources, is in part a consequence of poor macroeconomic performance.4 In the early postwar years migration to heavily forested frontier areas in the Philippines was of® cially encouraged as a means of alleviating economic and political pressures generated by increasing population and stagnating technology in the country’ s ricegrowing heartlands. In subsequent decades, continued spontaneous internal migration has been fostered by low rates of non-agricultural labor absorption, as well as a series of labor-saving technical changes in lowland irrigated agriculture (Jayasuriya and Shand 1986), in the face of sustained high rates of overall labor force growth. The resulting increases in landlessness and unemployment stimulated searches for open-access resources from which incomes, however tenuous, could be earned (property rights to uncultivated lands in the Philippine uplands are poorly de® ned and dif® cult to enforce). The outcome was a trebling of upland population between 1950 and 1985, from 5.8 million to 17.5 million, and annual growth rates of upland cropped area of greater than 7% over the same period (M. Cruz et al. 1992). The evidence that macroeconomic instability and 4 In Thailand, rapid economic growth and especially the expansion of labor-intensive manufacturing industries, was the major contributor to the stabilizing of agricultural land area during the ``boom’ ’ years 1986± 1996, through outmigration from marginal upland and rural areas (Coxhead and Plangpraphan 1999).

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

growth (or the lack of it) in non-agricultural sectors were major forces driving migration and upland land use decisions is compelling, if circumstantial (Cruz and Repetto 1992). There is a strong suggestion that microeconomic and trade policies also promoted forest conversion and intensi® cation in upland agriculture. In commercially-oriented upland agricultureÐ or even simply where labor is mobile into or out of upland areasÐ agricultural price policy can exert a signi® cant, though not immediately observable, in¯ uence on natural resource management. In the Philippines there is evidence of a pervasive policy bias in favor of cropsÐ such as corn and temperate vegetablesÐ whose cultivation is most strongly associated with upland agricultural land degradation, soil erosion, and related water pollution. This commodity bias emanates mainly from national-level economic policies, some of them unrelated to agriculture; it has been complemented in the past by the allocation of agricultural research resources; and it appears not to be offset by policy measures in favor of more environment-friendly cultivation techniques. Throughout the postwar era, successive Philippine governments have pursued selfsuf® ciency in grains, along with cheap consumer cereals prices, as key components of food security and income redistribution strategies. Philippine cereal yields are low by Asian standards, and with relatively little spending on agricultural infrastructure and technology, yields have not risen as quickly as in comparable countries. Consequently, grain output growth in uplands has been due primarily to area expansion. Given the political importance of self-suf® ciency, grain imports are tightly circumscribed, and this in turn has maintained domestic producer prices at levels well above the domestic-currency equivalents of world prices.5 Vegetable production has also received substantial policy support. Import bans imposed in 1950 on fresh potato, cabbage, and other horticultural crops (and reiterated in legislation as recently as 1993) were repealed and replaced by tariffs only in 1996 (see below). Demand for these non-traditional foods

253

grows with per capita income and urbanization. Since supply growth is limited by trade restrictions and climatic constraints, their prices have tended to rise more rapidly than the general price level, and certainly more rapidly than prices of most exportable crops and staple grains. For potato, the ban raises Philippine farm gate prices to nearly double the imputed c.i.f. (landed) wholesale price of imports, if they were permitted (Coxhead 1997). The Agricultural Tarif® cation Act of 1996 brought Philippine agricultural policy into compliance with the Uruguay Round of the GATT. Quantitative restrictions on corn and vegetables were replaced by tariffs, and minimum access volumes (MAVs) were speci® ed for each product. The MAV is the volume of a product that is allowed to be imported at a lower rate of duty than the maximum bound rate under the GATT. For the period to 2004, in-quota corn tariffs (those applying to MAV imports, which themselves cover roughly 50% of annual imports) remain at 35%. Out-quota tariff rates for corn, set at 100% in 1996 are scheduled to fall to 65% in 2000 (similar changes apply to vegetables). These reforms, although they constitute important steps in the direction of more open trade, ensure that upland farmers will continue to bene® t from protection at signi® cantly higher rates than most other sectors for the foreseeable future. Trade and price policy biases are also re¯ ected in the allocation of agricultural research funds. Most important among these for uplands are corn programs. A number of provinces (including Bukidnon, from which our primary data are drawn) have been designated as ``key production areas (KPAs)’ ’ for corn in the Philippine government’ s Grain Production Enhancement Program (GPEP). 5 The nominal protective rate (NPR, a measure of the amount by which domestic prices exceed landed import prices) for corn has generally been much higher than for any other major agricultural product, especially after the mid-1970s when corn self-suf® ciency was made a policy goal. The NPR averaged 18% in 1970± 1974, but rose to 42% by 1983± 1986, and to 62% by 1990 (Coxhead 2000); it has since remained in this range.

254

Land Economics

Farmers in KPA areas are eligible for subsidies and supports directed at increasing corn production, and are the ® rst bene® ciaries of research and development directed at increasing corn yields (Philippine Department of Agriculture 1994). Similarly, temperate climate vegetable crops are also the targets of disproportionate research resource allocations (Coxhead 1997). This brief review of Philippine growth strategy and policies has indicated a number of channels through which decisions concerning use of upland forest and farm land are likely to be in¯ uenced. In the longer term, a successful development strategy would have raised lowland and non-farm labor productivity faster than in uplands and diminished the economy’ s susceptibility to destabilizing macroeconomic shocks; all these should have reduced net migration to uplands and by extension, pressures on forest and land resources. Trade policy liberalization would in general have promoted growth of export-oriented non-agricultural sectors and might have preserved the pro® tability of some upland perennial export crops, such as coffee, relative to annual crops, and this in turn might have caused some redirection of input subsidy schemes and R&D resources away from import-competing crops and towards more promising sectors. Moreover, this review makes it clear that there are many potential policy changes at the macroeconomic level or in trade and agriculture sector policy that could affect upland resource allocation. On the basis of this evidence any project directed at in¯ uencing upland resource allocation toward a ``sustainable’ ’ path should at least be cognizant of this broader setting, if not actively involved in trying to alter it. The evidence we have reviewed, however, is strictly circumstantial. Questions remain as to the strength and nature of linkages between uplands and the national economy, and it is in this ® eld of inquiry, as previously noted, that speci® c data and evidence are lacking. The gap creates room for competing hypotheses about the upland economy, and these in turn imply different diagnoses of upland environmental problems and their solutions. In the next part of the paper we de-

May 2001

scribe the site from which primary data have been drawn in an attempt to ® ll this gap. The Study Site

The research site is located in Lantapan municipality, Bukidnon province, in northern Mindanao. Bukidnon is landlocked; its center consists of tablelands which descend towards the coast to the north, and in all other directions climb into some of the Philippines’ highest mountain ranges. Lantapan is located in the upper Manupali river valley, 130 km from Cagayan de Oro, the closest major city and port. The landscape climbs from river ¯ ats (500± 600m) through a rolling middle section (600± 1100m) to high-altitude, steeply sloped mountainsides (1100m± 2200m). The northern boundary of the municipality is the boundary of a major forest reserve, the Mount Kitanglad Range Nature Park. Most households are poor by Philippine standards, and rainfed agriculture dominates the local economy.6 Low-lying ¯ atlands are devoted to rice and sugar cane and corn-sugarcane systems dominate rolling mid-altitude areas. At higher elevations, corn is the predominant crop, planted alongside coffee and temperate-climate cropsÐ beans, tomatoes, cabbages and potatoes. The latter two crops require cool nighttime temperatures and so are generally grown above 1000m in ® elds adjacent to, and even within the park boundary. In both spontaneous migration and of® cial programs since the 1950s, Bukidnon was a major destination and watersheds like the Upper Manupali were choice locations. Sparsely settled at Independence in 1946, population growth rates peaked at 10% per year in the 1950s, with most of the increase due to in-migration from economically depressed areas of the central and northern 6 In 1988, 71% of provincial employment was in agriculture, 5% in industry, and 23% in services, and agriculture provided the primary income source for 68% of households. Expenditures on food, fuel and clothing accounted for 60% of householdbudgets (NSO 1990). Farm sizes are small by upland standards: the modal farm size class (1± 3 ha) contains 46% of farms, and 75% of all farms are smaller than 5 ha.

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

Philippines (NSO 1990). In the decade from 1970 to 1980, Lantapan’ s population increased at an average annual rate of 4.6% (NSO 1990). Since 1980, the annual population growth rate has averaged 4%, far higher than the Philippine average of 2.4%. Commercial agriculture has expanded along with population. Internal migrants introduced commercial cultivation of potato, cabbage, and other vegetables in the 1950s. More recently infrastructural improvements, coupled with increasing demand for vegetables and feed (yellow) corn, has ensured that commercial agriculture in the province continues to adapt and thrive. Corn and vegetable production have ¯ ourished; falling transport costs have helped these become primarily commercial crops, exported from to the national economy where formerly they had been little traded outside the province. Is There an Environmental Problem?

In Lantapan, agricultural expansion has occurred substantially at the expense of perennial crops, including forest (Figure 1). Other things equal, the replacement of perennial land uses with short-season and annual crops on sloping lands is associated with rapid increases in soil erosion and land degradation. Field measurements and experiments with the cultivation of corn and vegetables under a range of management regimes in Lantapan con® rm rapid erosion and soil nutrient and organic matter depletion (Midmore et al. 2001). In spite of these negative effects of the spread of annual crops, few farmers display deep knowledge of soil degradation relationships (ibid.). Land fallowing and crop rotation is rare and usually undertaken only when yields decline to the point of economic losses in the current season. Although soil erosion and land degradation problems appear to be widespread, few farmers report signi® cant investments in soil-conserving structures or technologies. Failure to adopt soil conservation measures is correlated with tenure insecurity; ® elds held in private title are more likely to be fallowed regularly, to have tree crops, and to have perennial grasses planted on boundaries than are ® elds operated by tenants. Farmers are

255

more likely to practice contour plowing on owned than on rented land (Coxhead 1995; Midmore et al. 2001). Agricultural intensi® cation without adequate soil management has deleterious effects both on-site and off-site. Intensive cultivation of annual crops in general, and the increased use of fertilizer, pesticides, and other chemicals on vegetable crops in particular, are likely to degrade water quality and may create health problems for farm families and those living downstream. Lantapanbased water quality monitoring reveals both qualitative and quantitative evidence of such problems. Perceptions of pesticide residues have made some residents reluctant to water animals in streams during, or after heavy rain. Measures of total suspended solids (TSS) across sub-watersheds are considerably higher where agricultural cultivation is more widespread, in spite of lower average slope, and seasonal TSS peaks coincide with months of intensive land preparation activity. Many of the more noticeable changes in water quality and seasonal ¯ ows have occurred ``well within human memory’ ’ (Deutsch et al. 2001). Finally, the unchecked expansion of agriculture into the national park poses a potential threat to the biological integrity of the remaining forest. In the early postwar years, forest encroachment was driven mainly by commercial logging, but in the past two decades the expansion of small corn and vegetable farms has been the primary impetus, with decisive contributions from road development and the lack of established property rights in land (Cairns 1995). Concerns arising from forest removal and degradation include such speci® c phenomena as loss of watershed function (especially with clearing in the headwaters of creeks), changes in the quantity and seasonal distribution of water ¯ ow in springs and rivers, loss of wildlife habitat, and reduced availability of forestbased foods and raw materialsÐ as well as more general, and less easily quanti® ed, phenomena such as biodiversity loss. In summary, evidence on environmental problems in the watershed provides emphatic support for two arguments. First, the natural resource base of the watershed is undergoing

256

Land Economics

FIGURE 1 Land Use Changes, Municipality of Lantapan, 1973± 1994 Source: Li, 1994, Tables 5.9 and 6.12

May 2001

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

degradation of a nature and at a rate without modern precedent. Second, much if not most of the degradation can be attributed directly or indirectly to the spread of intensive agricultural systems based on corn and vegetables. In this setting, the designers of forest conservation and sustainable agriculture projects debate both the root causes of deforestation and land degradation, and the means by which they should be addressed. III. ANALYSIS OF MARKETS, PRICES, AND LAND USE DECISIONS Our research focuses on factors in¯ uencing land use in the middle and upper watershed areas, on relatively steep and easily eroded valley sides, and at the forest margins. The major crops grown are corn (both for feed and for human consumption) and vegetablesÐ especially cabbage, beans and potato. In the analysis that follows we concentrate on corn, as by far the most important crop, in terms both of land use and of net farm incomes, within the study site. Nationally, too, the area planted to corn is second only to rice, and corn accounts for, by far, the greatest part of upland agricultural land use. An initial survey of the Lantapan site had characterized agriculture in the upper watershed as ``subsistence’ ’ or ``semi-subsistence’ ’ (Bellows 1993). However, our data reveal clear commercial motivations for almost all farmers.7 More than 50% of corn production is destined for market, and vegetable crops such as cabbage, potato, and bean strictly for sale, with home consumption accounting for less than 10% of production in each case (Coxhead 1995). An econometric analysis of land use decisions by upland farmers in a comparable Philippine location indicates that their land allocations respond to relative prices, and to price variability, in statistically signi® cant ways (Shively 1998). A similar exercise using Lantapan land use data (Coxhead, Shively, and Shuai 1999) reveals that farmers’ decisions on total land area farmed and its allocation to crops are in¯ uenced in statistically signi® cant ways by household resource availability, physical and institutional

257

constraints, and the variances of expected revenues, the latter indicating risk-averse behavior.8 The results with respect to planted area response to relative prices are somewhat weaker; although coef® cients have the expected signs, they do not meet standard tests of statistical signi® cance. A question remains as to the relative importance of markets, as well as of national policies operating through them, as conditioning in¯ uences over farmers’ decisions. If prices or their variability are important determinants of land use decisions, what are their determinants? Market Integration and Price Causation

As argued earlier, understanding the nature of market links between uplands and the rest of the economy is critical to the ef® ciency of project and policy design. If markets within the study site were isolated from, or only weakly associated with regional markets (the ``semi-subsistence’ ’ hypothesis), we would expect to see seasonal or even longer-term divergence between trends in local and regional prices. Further, we would be unable to see evidence that local prices are driven by national prices. The tests of market integration and the direction of causation are important for both economic and environmental reasons. Under current production technologies corn, potato, cabbage, and other intensive crops in Lantapan generate annual erosion and soil nutrient losses far in excess of natural regeneration rates. Remoteness and poor quality of infrastructure are frequently taken to indicate that market links to the rest of the economy are tenuous at best. This, if true, would have two important implications for policy and project design. It would mean that agricultural prices and trade policiesÐ standard instruments for 7 Data on production, input use, land use and sales for major crops, were collected annually from a sample of 120 farms in four rounds between 1994 to 1998 (for full details see Coxhead 1995 and Rola and Coxhead 1997). 8 In the Philippines, corn prices are stabilized through policy interventions, and the results of this exercise con® rm that price stabilization encourages riskaverse farmers to increase corn area.

258

Land Economics

in¯ uencing agricultural resource allocation in lowlandsÐ could be expected to have little or no effect in uplands. By extension, the most effective instruments for promoting sustainable agriculture in uplands would be direct interventions such as technology transfer, extension, and education. Alternatively, if markets are integrated but farm-gate prices are most signi® cantly in¯ uenced by local production, then supply and price in upland agriculture will tend to move in opposite directions. If an increase in local supply drives prices down, then the pro® tmaximizing level of local output will be lower than if prices were unaffected. In this case, the price-reducing effects of local adoption of supply-increasing innovations such as new technologies or more ef® cient management practices might be expected to act as a ``natural brake’ ’ on the expansion of agriculture at the forest margin. However, these effects (even if they were observed) are likely to obtain only in the short run, since integration with the larger market will likely neutralize local effects in the longer run. Unfortunately, there is no widely accepted statistical test of market integration, only tests of relationships between prices that (if con® rmed) can be said to be consistent with integration. Theory tells us that if two markets are linked through trade, then under normal circumstances, differences in prices net of margins between the two markets create opportunities for arbitrage. Goods will ¯ ow between the two marketsÐ trade will occurÐ until the price difference is eliminated. Statistically, if the prices in the tested markets are non-stationary (that is, that they are trending over time rather than merely following a random walk) then the markets are integrated if their price series are cointegrated, meaning that there is a (single) stationary long-run relationship between them. We investigated the time series properties of the price data on yellow corn, white corn, potatoes, and cabbage in Lantapan and Agora and in each case found the series to be stationary.9 Therefore, we cannot conduct a statistical test of long-run market integration. However, our observation con® rms that trade between Lantapan and Agora is regular, seasonally consistent, and consists of high vol-

May 2001

umes; a statistical ® nding of no integration would be a very great surprise. Studies using aggregate data have indicated clearly that Philippine grain markets are integrated across regions and provinces (Mendoza and Rosegrant 1995; Silvapulle and Jayasuriya 1994). Examining the short-run dynamics of the price series permits tests of the hypotheses that upland farmers are price-takers and that national market and policy signals affect local prices. Our econometric method proceeds as follows. We ® t the data to a set of regression equations, each of which has the price of a crop in one market as the dependent variable, and its own lagged values, as well as the current and lagged values of the prices of the same crop in other markets, as explanatory variables. Hypothesis tests on the coef® cient estimates of these equations provide information about the direction of causation. As an example, for two markets A and B, when a price change in market A is shown to precede price changes in market B, we describe the price in A as ``Granger-causing’ ’ that in B. In our study, con® rmation that the local price Granger-causes the regional price would provide support for the ``natural brake’ ’ idea referred to above, that expanded production of a crop within the watershed will cause its price to fall, at least in the short run. Conversely, con® rmation that the regional price causes the Lantapan price would indicate a need to focus on agricultural price and trade policies as longer-run in¯ uences over farmers’ land use and crop production decisions.10 9 The test for stationarity is conducted with a Dickey-Fuller test of the null hypothesis that each price exhibits a unit root. For example, under an AR(2) representation of yellow corn prices (including seasonal dummy variables), the ADF test statistics for this hypothesis are 24.818 for Agora and 25.307 for Lantapan. At the 5% signi® cance level, the critical value for the test is 22.88, so we reject the null hypothesis. We obtain similar results for the other products; these results are robust with respect to different lag speci® cations. We conclude that these price series are stationary. 10 Both Granger-causality and the test of transmission of shocks (impulse response function) are founded on the vector autoregression (VAR) speci® cation of a price series.

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

259

FIGURE 2 Weekly P rice of Yellow (Feed) Corn, October 1994± December 1999 (peso/kg)

The test of causation is also a test of a suf® cient condition for short-run market integration, so long as at least one causal relationship is con® rmed. It is, however, important to note that strictly speaking, our method provides what is best described as circumstantial evidence on integration and causation. The conclusion of ``causation’ ’ is reached by observing temporal precedence, but no economic mechanism of causation can be spelled out. We apply these tests to weekly corn, potato, and cabbage prices in Lantapan and the main regional market. Crop price data were collected weekly from traders at several points in Lantapan, from provincial centers and from the main regional wholesale market (``Agora’ ’ ) in Cagayan de Oro, the regional capital and port. Much of the produce sold in the Agora market is shipped directly to Manila, the national capital and central market, either for processing or for sale; accordingly, Agora prices track the benchmark Manila prices. In this analysis, we concentrate on the Lantapan-Agora market relationship (the data series are summarized in Figures 2± 4).

To account for the time series properties of the data we employ a vector auto-regression (VAR) model (Sims 1980). The VAR approach to time series analysis is controversial. As Cooley and Leroy (1985) have pointed out, the VAR is ``atheoretical’ ’ in the sense that it embodies no explicit economic theory. However, when restrictions in the VAR model, in terms of choices of variables and lag lengths, are weaker than the restrictions imposed on structural models, the VAR approach can provide a foundation for testing hypotheses based on a priori reasoning (Backus 1986). In our investigation of price relationships, we use both economic and econometric tools to choose variables and lag lengths. We thus view the VAR approach as a complement to the structural models implied by theory. Speci® cally in the case of Lantapan, the quality of transport infrastructure, high frequency of public and private travel, and the distance (130 km, or at most 5 hours) to the major market all suggest that price signals can be exchanged, and arbitrage occur, well within the two-week interval implied by a two-period lag structure. The structural equations of the VAR

260

Land Economics

FIGURE 3 Weekly P rice of Cabbage, October 1994± December 1999 (pesos/kg)

FIGURE 4 Weekly P rice of P otato, October 1994± December 1999 (pesos/kg)

May 2001

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

model (with 2-period lags, suppressing cropspeci® c subscripts) are: PLt 5 a 1 PAt 1 b 11 PLt21 1 b 12 PAt21 1 g 11 PLt22 1 g 12 PAt22 1 n PLt PAt 5 a 2 PLt 1 b 21 PLt21 1 b 22 PAt21 1 g 21 PLt22 1 g22 PAt22 1 n PAt , [1]

where PLt and PAt are prices in Lantapan and in the Agora regional market respectively, and n PLt and n Pat are error terms that we assumed are serially and mutually uncorrelated. Eliminating current-period variables from the right-hand sides of these equations yields a reduced form: PLt 5 f 11 PLt21 1 f 12 PAt21 1 f 13 PLt22 1 f 14 PAt22 1 e 1t PAt 5 f 21 PAt21 1 f 22 PLt21 1 f 23 PAt22 1 f 24 PLt22 1 e 2t , [2]

in which e 1t and e 2t are unobservable variables which are the serially uncorrelated innovations in the PL and PA processes.

261

Granger causality tests utilize test statistics computed from the VARs. A variable mt is said to fail to Granger-cause another variable yt relative to an information set consisting of past values of mt and yt if EÃ[yt | yt21 ,mt21 ,yt22 ,mt22 , . . .] 5 EÃ[yt | yt21 ,yt22 , . . .]

[3]

where EÃ denotes a linear projection of the dependent variable. In our example, this means that PA does not Granger-cause PL relative to an information set consisting of past values of PA and PL if (and only if ) the estimates of f 12 and f14 are equal to zero. In practice, an F-test can be used to test the null that one variable does not Granger-cause another. The results of these F-tests are summarized in Table 1. All markets display some form of causation, and so we conclude that local and regional markets are integrated for all crops in the study. For yellow corn and

TABLE 1 Summary of Results of Granger Causality Tests for Corn and Vegetable P rices Testa

R2

DWb

F (N; d.f.)

P value c

Comments

Yellow Corn

Agora ® Lantapan Lantapan ® Agora

0.75 0.86

1.97 2.04

3.22 (182;2,176) 0.91 (182;2,176)

0.042 0.403

One-way causation

White Corn

Agora ® Lantapan Lantapan ® Agora

0.89 0.95

1.95 1.96

8.25 (162;2,156) 0.39 (162;2,156)

0.004 0.680

One-way causation

Avg. Potato

Agora ® Lantapan Lantapan ® Agora

0.81 0.84

1.95 2.08

6.61 (157;2,151) 7.17 (157;2,151)

0.002 0.001

Two-way causation

Cabbage

Agora ® Lantapan Lantapan ® Agora

0.86 0.68

1.97 1.96

2.88 (170;2,164) 5.60 (170;2,164)

0.005 0.004

Two-way causation

Avg. Potatod

Agora ® Lantapan Lantapan ® Agora

0.75 0.83

2.05 2.12

13.8 (83;2,76) 0.77 (83;2,76)

0.001 0.470

One-way causation

Cabbage

Agora ® Lantapan Lantapan ® Agora

0.61 0.56

1.90 1.99

3.36 (41;2,35) 0.34 (41;2,35)

0.046 0.710

One-way causation

Crop Weekly data

Monthly data

a Arrows indicate the direction of causation being tested, so for example ``Agora ® Lantapan’ ’ indicates a test that Agora price Granger causes Lantapan price. b Durbin-Watson statistic. c P , 0.01 indicates rejection of the null hypothesis (no causation) at 1% signi® cance level; 0.01 , P , 0.05 indicates rejection at 5%; 0.05 , P , 0.1 indicates rejection at 10%. d Biweekly data for average prices of large and medium potatoes.

262

Land Economics

white corn, the direction of causation runs from wholesale market to farm gate. Corn prices in the watershed are driven entirely by prices in provincial and national markets. For potato, weekly data indicate two-way causation: farm gate prices are in¯ uenced by wholesale prices, but a local supply shock in Lantapan may also have a short-run effect in wholesale markets. Using biweekly data, however, we ® nd a strong one-way relationship between Lantapan and Agora prices, with causality running from the latter to the former. For cabbage, the weekly data show a strong in¯ uence of Lantapan prices on wholesale prices, but monthly data show that when very short-term ¯ uctuations are smoothed out, cabbage prices are determined in the regional market and not within the watershed. To summarize, our results indicate that markets for the major crops grown in the watershed are integrated in the short run with broader regional markets. They also provide strong evidence for all crops that an expansion of supply within the watershed will have no measurable in¯ uence on prices in wholesale markets, beyond a period of one or two weeks for vegetable crops. Rather, the evidence is that farmers in the watershed are price takers in regional and national markets. Market and Policy Linkages in Lantapan

If markets are integrated as we have argued, and given that short-run causality runs only from regional to local market, what can we conclude about the implications of national policies for upland land use in Lantapan? For the reasons indicated earlier, we cannot as yet quantify the effects of changes in the trade policy regimes that underpin domestic market conditions for both corn and vegetables. For vegetables, import bans that prevailed until 1996 have been replaced with tariffs at prohibitive rates; in effect, there has been no trade policy change. For corn, in spite of the shift from quantitative restrictions to the MAV system with tariffs after 1996, announced trade policy changes are being introduced very gradually and are not scheduled to be completed before 2004.

May 2001

However, our ® nding that upland farmers are price-takers in regional markets makes it clear that any meaningful policy changes, were they to occur, would have direct effects on farmgate prices in the uplands. Of potentially greater interest is the observation that revenue instability, the phenomenon that risk-averse farmers strive to avoid, has intersectoral as well as local sources, even in a market (such as corn) which is subject to price stabilization. Our data span the recent economic crisis that engulfed Southeast Asian countries, beginning when the Thai currency collapsed in July 1997. While the crisis took different forms in each affected economy, there were three elements common to all. There was a sharp drop in overall economic growth, and there were sudden, unexpected and repeated re-evaluations of exchange rates that had previously been effectively pegged to the U.S. dollar. As a result there was a big increase in uncertainty among producers within the affected countries about ® nal demand and prices, input prices, and even availability of key inputs such as credit. Since trade policy renders Philippine corn prices largely independent of world prices in the short run, were upland markets affected by the macroeconomic instability re¯ ected in the exchange rate? We used information about exchange rate variability to de® ne the endpoints of the Philippine economic crisis.11 During the period August 1997 to November 1998, the daily peso-dollar rate ¯ uctuated wildly, whereas before and after this episode, the mean daily change was a fraction of 1% (Figure 5). We use this criterion to divide our data into ``precrisis,’ ’ ``crisis,’ ’ and ``post-crisis’ ’ periods; as Table 2 shows, the price variance of yellow corn, the major crop in Lantapan, increased substantially during the crisis, even through the mean price did not. We are then able to make a preliminary identi® cation of the ef11 Although the values of the exchange rate are of direct interest in their own right, here we are using exchange rate ¯ uctuations as a proxy for a more general set of macroeconomic conditions. In an open economy, exchange rate depreciation (as occurred during the early part of the crisis) serves as a proxy for (unobservable) in¯ ationary expectations; exchange rate variability is then a proxy for general price instability.

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

263

FIGURE 5 Daily Fluctuations of the P hilippine P eso Against the U.S. Dollar

fects of macroeconomic instability on the relationship between farm gate prices and those in national markets. We do this by calculating impulse response functions, which record the dynamic response of one data series to a onetime shock (``impulse’ ’ ) in another (see TABLE 2 Moments of Yellow Corn P rices Before, During and After the Exchange Rate Crisis Item

Pre-crisis

Crisis

Post-crisis

Lantapan mean price (Peso/kg) Agora mean price (Pesos/ kg)

5.57

5.85

5.19

6.15

6.25

5.23

Lantapan price variance Agora price variance

0.627

0.774

0.462

0.526

1.277

0.221

Exchange rate (Peso/USD) Exchange rate variance

25.9 0.468

38.7 16.892

38.9 0.817

Note: Periods are as de® ned in text and illustrated in Figure 5.

Greene 1993). For example, the dynamic response of a shock in Agora on the Lantapan price can be captured by ¶PLt1j /¶v PAt.12 The impulse response measures are thus computed from the same VAR model used earlier to test market relationships, only with the data divided into sub-periods as noted. The dynamic response of Lantapan corn prices to a shock in the Agora regional market price is shown for a two-period lag model in Figure 6. The ``impulse’ ’ is a one-peso per kilogram price shock, so the ® gures on the vertical axis of the graph are pesos per kilogram in the Lantapan market (the mean preshock corn price was about 6 pesos/kg, so this represents a shock of about 16%). In the pre-crisis and post-crisis periods, a shock in the Agora price yields a maximum rise in local prices of about 3% (0.2 pesos). The impulse response peaks 3 weeks after the shock and drops very sharply to a negligible amount by the 5th week after the shock. During the crisis, the peak is much larger (6%), and is sustained over several months. Comparing responses during the crisis period and 12 This algebraic derivation involves successive substitution (Greene 1993).

Land Economics

264

May 2001

FIGURE 6 Impulse Response Function: Yellow Corn (Two-P eriod Lag Structure)

in the earlier and later periods, we see that in the post-crisis era the signal from the leading Agora price to the Lantapan price is very much more ``noisy’ ’ than in the prior period. The price dynamics indicate that during the crisis, a temporary disturbance in the Agora series induces a larger and longer-lived response in farm-gate prices. While very preliminary in nature, the impulse response analysis suggests that the effects of macroeconomic instability ® nd their way into the behavior of prices that guide farming decisions even in areas far from the main regions and sectors of economic activity.13 The economic signals upon which upland farmers make resource allocation decisions are not independent of conditions in national markets and in the macroeconomy. More rigorous investigations of these relationships, for corn and for other crops, will become feasible as more data from the postcrises era become available. IV. CONCLUSIONS Commodity market development, along with policy biases, has contributed to de-

forestation and the adoption and spread of relatively erosive crops, produced using relatively land-degrading technologies, in the upland Philippine watershed of our study. The environmental ill-effects of these crops could be minimized by adoption of appropriate technologies for example, to reduce erosion and preserve soil quality. However, only a few farmers in the study site have adopted effective soil conservation measures, and while this is clearly related to tenure insecurity, there is also evidence that among all farmers, the choice of annual commercial crops, and the failure to adopt soilconserving technologies, has economic as well as institutional roots. If market-driven incentives dominate in farmers’ decisions,

13 The exact effects of price instability on land use by Lantapan farmers cannot currently be determined with any greater precision than is provided in this statement. One reason is that our land use response estimates in Table 4 are based only on pre-crisis data and may not be stable once post-crisis data are incorporated. The analysis of the effects of the crisis in Philippine upland agriculture is the subject of ongoing research as new data become available.

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

there is a case for broadening the range of policy instruments brought to bear on the upland environmental problem; moreover, project design may be improved by a different balance of local action and national-level information dissemination and policy advocacy. We have demonstrated that in spite of remoteness, the farmers in our study area produce for markets that are integrated in the national system. Supply shocks from the site have no effect on prices in broader markets: farmers are price-takers in these markets. National markets transmit both price information and the effects of macroeconomic instability. While empirical tests of the effects of trade policies on prices await substantive policy changes, it is nevertheless clear that agricultural markets convey the effects of trade policies to the farm gate, even in upland agriculture. Trade liberalization can therefore be expected to reduce the farm-gate prices of corn and vegetables, the two most environmentally damaging crops currently grown in Lantapan and many similar Philippine watersheds. Finally, anecdotal evidence of the importance of macroeconomic trends in driving upland migration and land use patterns is provided some additional contemporary support by our ® nding that the stability of market price relationships is a function of price stability in the overall Philippine economy. During the recent economic crisis, we ® nd that instability at the macroeconomic level (as re¯ ected in daily exchange rate movements) was associated with a noisier signal from wholesale to farm gate prices. Future research on the links between deforestation and agricultural expansion should bene® t from this exposure of the importance of markets and prices in a typical frontier area of a tropical developing country. A combination of project-speci® c and more general policy measures is called for if the former are to succeed in changing farmers’ actions, and if the latter are not to discourage environmentally sustainable strategies. At a policy level this research, if supported by counterpart studies from other sites, should provoke a reconsiderationÐ and indeed a substantial

265

broadeningÐ of the set of policy instruments available to in¯ uence upland agricultural and forest land allocations. References Anderson, J. R., and J. Thampapillai. 1990. ``Soil Conservation in Developing Countries: Project and Policy Intervention.’ ’ Agriculture and Rural Development Department, Policy and Research Series, Paper No. 8. Washington, D.C.: World Bank. Angelsen, Arild. 1999. ``Agricultural Expansion and Deforestation: Modelling the Impact of Population, Market Forces and Property Rights.’ ’ Journal of Developmental Economics 58:185± 218. Backus, D. 1986. ``The Canadian-U.S. Exchange Rate: Evidence from a Vector Autoregression.’ ’ Revue of Economic Statistics 68:628± 37. Barbier, E. B. 1990. ``The Farm-Level Economics of Soil Conservation: the Uplands of Java.’ ’ Land Economics 66 (May): 198± 211. Barrett, S. 1991. ``Optimal Soil Conservation and the Reform of Agricultural Pricing Policies.’ ’ Journal of Developmental Economics 36 (2): 167± 87. Bellows, B., ed. 1993. ``A Participatory Landscape-Lifescape Analysis of the Manupali Watershed in Bukidnon, Philippines: Characterization of the Landscape and Identi® cation of Research Priorities for the SANREM CRSP.’ ’ Athens, Ga. Mimeo. Cairns, M. 1995. ``Ancestral Domain and National Park Protection: Mutually Supportive Paradigms? A Case Study of the Mt. Kitanglad Range National Park, Bukidnon, Philippines.’ ’ Paper presented at a workshop on Buffer Zone Management and Agroforestry, Central Mindanao University, Musuan, Philippines, August. Mimeo. Cooley, T. F., and S. F. Leroy. 1985. ``Atheoretical Macroeconomics.’ ’ Journal of Monetary Economics 16:283± 308. Coxhead, I. 1995. ``The Agricultural Economy of Lantapan Municipality, Bukidnon, Philippines: Results of a Baseline Survey.’ ’ Sanrem Social Science Group Working Paper No. 95/ 1. Madison, Wis. Mimeo. Ð Ð Ð . ``Induced Innovation and Land Degradation in Developing Country Agriculture.’ ’ Australian Journal of Agricultural and Resource Economics 43 (2): 305± 32.

Ð

266 Ð

Land Economics

Ð . 2000. The Consequences of Philippine Food Self-Suf® ciency Policies for Economic Welfare and Agricultural Land Degradation.’ ’ World Development 28 (1): 111± 28. Coxhead, I., and Jiraporn Plangpraphan. 1999. ``Economic Boom, Financial Bust, and the Decline of Thai Agriculture: Was Growth in the 1990s Too Fast?’ ’ Chulalongkorn Journal of Economics 11 (1): 76± 96. Coxhead, I., and S. K. Jayasuriya. 1994. ``Technical Change in Agriculture and Land Degradation in Developing Countries: A General Equilibrium Analysis.’ ’ Land Econonomics 70 (Feb.): 20± 37. Ð Ð Ð . 1995. ``Trade and Tax Policy Reform and the Environment: The Economics of Soil Erosion in Developing Countries.’ ’ American Journal of Agricultural Economics 77 (3): 631± 44. Ð Ð Ð . 2000. ``Models of Deforestation and Land Degradation in Open Developing Economies: A Consolidation.’ ’ Staff Paper Series No. 430. Department of Agricultural and Applied Economics, University of Wisconsin± Madison. Coxhead, I., and G. Shively. 1998. ``Measuring the Environmental Impacts of Economic Change: The Case of Land Degradation in Philippine Agriculture.’ ’ Journal of Agricultural Economics and Development 26 (1/2): 61± 92. Coxhead, I., G. Shively, and X. Shuai. In press. ``Development Policies, Resource Constraints and Agricultural Expansion on the Philippine Land Frontier.’ ’ Environment and Development Economics. Cruz, W., and R. Repetto. 1992. The Environmental Effects of Stabilization and Structural Adjustment Programs: The Philippines Case. Washington, D.C.: World Resources Institute. Cruz, Maria C., C. A. Meyer, R. Repetto, and R. Woodward. 1992. Population Growth, Poverty, and Environmental Stress: Case Studies from the Philippines and Costa Rica. Washington, D.C.: World Resources Institute. Deacon, R. 1995. ``Assessing the Relationship between Government Policy and Deforestation.’ ’ Journal of Environmental and Economics Management 28 (1): 1± 18. Deutsch, W. G., A. L. Busby, J. L. Orprecio, E. Cequina, and J. Bago. 2001. ``CommunityBased Water Quality Indicators and Public Policy in the Rural Philippines.’ ’ In Seeking Sustainability: Challenges of Agricultural Development and Environmental Management in a Philippine Watershed, ed. I. Coxhead and G.

May 2001

Buenavista. Los BanÄos, Philippines: Philippine Council for Agriculture, Forestry, and Natural Resources Research and Development (PCARRD). Greene, W. H. 1993. Econometric Analysis. Englewood Cliffs, N.J.: Prentice-Hall. Jayasuriya, S., and R. T. Shand. 1986. ``Technical Change and Labor Absorption in Asian Agriculture: Some Emerging Trends.’ ’ World Development 14 (3): 415± 28. Li, Bin. 1994. Impact Assessment of Land Use Changes in a Watershed Area Using Remote Sensing and GIS: A Case Study of Manupali Watershed, the Philippines. M.S. thesis, Asian Institute of Technology, Bangkok. Lopez, R., and M. Niklitschek. 1991. ``Dual Economic Growth in Poor Tropical Areas.’ ’ Journal of Developmental Economics 36:189± 211. Mendoza, M. S., and M. W. Rosegrant. 1995. Pricing Behavior in Philippine Corn Markets: Implications for Market Ef® ciency. Research Report 101. Washington, D .C.: International Food Policy Research Institute. Midmore, D., L. Ramos, W. Hargrove, D. Poudel, T. Nissen, F. Agragan, and G. Betono. 2001. ``Stabilizing Commercial Vegetable Production in the Manupali Watershed, the Philippines.’ ’ In Seeking Sustainability: Challenges of Agricultural Development and Environmental Management in a Philippine Watershed, ed. I. Coxhead and G. Buenavista. Los BanÄos, Philippines: Philippine Council for Agriculture, Forestry, and Natural Resources Research and Development (PCARRD). Philippine National Statistics Of® ce (NSO). 1990. Bukidnon Provincial Pro® le. Manila: NSO. Philippine Council for Agricultural and Rural Resources Research and Development (PCCARRD). 1999. Guidelines for Watershed Management and Development in the Philippines. Los Banos, Laguna: PCARRD. Philippine Department of Agriculture. 1994. Grain Production Enhancement Program. Quezon City: Philippine Department of Agriculture. Pingali, P. 1997. ``Agriculture-Environment Interactions in the Southeast Asian Humid Tropics.’ ’ In Sustainability, Growth, and Poverty Alleviation: A Policy and Agroecological Perspective, ed. S. Vosti and T. Reardon. Baltimore: Johns Hopkins University Press for the International Food Policy Research Institute. Rola, A. C., and I. Coxhead. 1997. ``The Agricultural Economy of an Upland Community: A

77(2)

Coxhead, Rola, and Kim: National Markets and Price Policies

Revisit to Lantapan, Bukidnon, Philippines.’ ’ University of the Philippines± Los BanÄos. Mimeo. Shively, G. E. 1997. ``Consumption Risk, Farm Characteristics, and Soil Conservation Adoption among Low-Income Farmers in the Philippines.’ ’ Agricultural Economics 17 (2): 165± 77. Shively, G. E. 1998. ``Economic Policies and the Environment: The Case of Tree Planting on Low-Income Farms in the Philippines.’ ’ Environment and Development Economics 3 (1): 83± 104. Silvapulle P., and S. Jayasuriya. 1994. ``Testing

267

for Philippines Rice Market Integration: A Multiple Cointegration Approach.’ ’ Journal of Agricultural Economics 45 (3): 369± 80. Sims, C. 1980. ``Macroeconomics and Reality.’ ’ Econometrica 48 (1): 1± 48. Southgate, D. 1988. ``The Economics of Land Degradation in the Third World.’ ’ Working Paper No. 2. Washington, D.C.: World Bank Environment Department. Tolentino, V. B. J. 1995. ``Intention vs. Implementation of Philippine Economic Reforms under Aquino, 1986± 1992.’ ’ VRF Series No. 240. Tokyo: Institute of Developing Economies.

Agricultural Change, Rural Labor Markets, and Forest Clearing: An Illustrative Case from the Philippines Gerald E. Shively ABSTRACT. This paper studies the links between agricultural employment and upland activities at a tropical forest margin. A model of lowland agricultural production is combined with a model of labor allocation on a representative upland farm to show how labor productivity, agricultural wages, and the returns from upland activities determine rates of forest clearing. Farm level data from the Philippines demonstrate how agricultural intensi® cationÐ in the form of lowland irrigation developmentÐ led to an increase in labor demand, an increase in employment of upland inhabitants, and small but statistically signi® cant reductions in rates of forest clearing. (JEL Q15, Q23)

I. INTRODUCTION Inadequate labor absorption is an important economic and environmental policy problem in many developing countries. In the Philippines, approximately 4.4 million jobs must be generated each year to employ additions to the labor force, two-thirds of which come from rural areas. An important outcome of rapid population growth in many frontier areas of the Philippines has been the expansion of agriculture into marginal and environmentally sensitive upland areas (Cruz, et al. 1992; Western 1988). Historically, deforestation rates in the Philippines have been high.1 High rates of forest clearing in the uplands are driven, in part, by the efforts of low-income farmers to secure subsistence.2 Finding ways to increase agricultural capacity and rural incomes without jeopardizing remaining forest resources has emerged as an important policy goal throughout the Philippines, and especially in Palawan (Sandalo 1996). Agricultural intensi® cation in lowland areasÐ for example, through innovations such Land Economics · May 2001 · 77 (2): 268± 284 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

as irrigation developmentÐ is one path by which agricultural capacity and rural incomes can be enhanced. But from an environmental perspective, attempts to raise rural incomes through agricultural improvement can work at cross-purposes. Technologies that increase the returns to agriculture can reduce the need for subsistence-driven land clearing, but by raising incomes and the returns to agricultural activities they can also provide incentives to convert forest to farmland or other uses (Kaimowitz and Angelsen 1998). Forest clearing is driven by numerous factors beyond timber trade. In many areas of The author is an associate professor in the Department of Agricultural Economics at Purdue University. Ed Barbier, Ian Coxhead, Sisira Jayasuriya, Will Masters, and two anonymous reviewers provided helpful comments and suggestions on an earlier draft. Elmer Martinez and Richard Yao provided valuable research and ® eld assistance. Funding was provided by grants from the Ford Foundation and the SANREM CRSP under USAID award RC710-006/ 5912994. 1 The terms deforestation and forest clearing are not synonymous and their use is problematic. When pressure on forests is low, forest clearing, which may take the form of cyclic clearing of trees followed by forest regeneration after fallowing, need not lead to deforestation. Deforestation generally implies permanent loss of forest cover, which may or may not occur in response to short-cycle forest clearing by smallholders. In this paper, the term forest clearing is preferred, and is used to describe a set of activities undertaken by upland households that includes expansion of agricultural land and extraction of forest products such as charcoal, fuelwood, or building materials that may temporarily or permanently alter the character of a forest. 2 During the past three decades, Philippine economic policies also have played an important role in promoting the expansion of agriculture at forest margins. These policies have consisted mainly of market interventions directed at supporting and stabilizing farm prices and trade interventions directed at reducing dependence on imports and defending the livelihood of upland farmers. Corn producers in particularÐ mainly upland farmersÐ have received considerable encouragement in the form of import restrictions and domestic price supports (Coxhead 1997).

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

Central America and the Amazon, farmers seeking expansion of their livestock holdings are an important part of the process of forest clearing (Faris 1999; Mattos and Uhl 1994; Yanggen and Reardon 2001). Rising prices of shrimp for export combined with favorable economic conditions for their production have provided the catalyst for mangrove destruction in many areas (Nickerson 1999; Parks and Bonifaz 1995). And policy-induced expansion of plantation and cash crops has been implicated in forest destruction throughout Africa, Asia, and Latin America (e.g. Angelsen 1995; Theile and Wiebelt 1993; Barbier and Burgess 1996). In most settings, incentives for clearing forest are determined, in part, by the relative returns to labor directed at cutting trees. In a recent review of the literature on tropical deforestation, Kaimowitz and Angelsen (1998) conclude that an inverse relationship exists between rural wages and deforestation rates. This is because in many areas the forest constitutes an important potential source of new land and livelihood for those who are not otherwise gainfully employed. Although not an employer in the literal sense, the forest is the ® rst destination of many settlers and the ® nal destination of others, and incentives to clear forest are strongly in¯ uenced by costs of access and returns to alternative activities. Where rural labor markets exist, employment opportunities in¯ uence choice of activity and time allocation. As a result, the rural labor market provides a useful lens through which to examine the link between economic development in the lowland economy and incentives for forest clearing in upland areas. Although rural employment obviously includes employment both inside and outside of agriculture, the focus in this paper is on the agricultural labor market and the employment effects of agricultural intensi® cation. The argument explored in the paper is the following: economic policies that intensify lowland agriculture have the potential to encourage labor absorption and help slow local rates of forest clearing. But such policies do not guarantee that employment gains will be sustained or that forest clearing will stop. Part of the reason behind this cautionary

269

view is that, from a theoretical perspective, the overall impact of agricultural intensi® cation on rural agricultural employment is ambiguous. Some forms of intensi® cation, such as irrigation development, can increase labor demand by facilitating multiple cropping and thereby increasing the annual effective area under cultivation. But in the case of irrigation it is equally clear that technical progress can reduce overall labor demand if it is biased against labor. A number of researchers have observed that, while irrigation may not have a built-in bias against labor, farmers who have access to irrigation also tend to adopt labor-saving methods such as mechanized production or chemical-based weed control (e.g., Lingard 1994; Castillo, Gascon, and Jayasuriya 1983; Kikuchi and Hayami 1983). As an example, Coxhead and Jayasuriya (1986) describe a case from the Philippines where employment in irrigated farming declined, even though real wages were falling. The goal in this paper is both conceptual and empirical. The investigation focuses on understanding how a change in the technology of lowland agricultural production alters incentives for activities at the forest margin. Two theoretical models are presented to link representative lowland and upland households. The subsequent empirical analysis tests the hypothesis that irrigation development in lowland agriculture reduces rates of forest clearing in adjacent upland areas. Data from a frontier farming area of the Philippines are used to trace the impact of lowland irrigation development to changes in the range and intensity of activities undertaken by upland households. The patterns of activity observed at these sites underscore the importance of the rural labor market as a mechanism in¯ uencing environmental outcomes in remote and environmentally sensitive areas. Results show irrigation development led to an overall increase in labor use on lowland farms vis-aÁ-vis rainfed conditions. Although labor demand per hectare was lower on irrigated lowland farms than on neighboring rainfed farms, and was also lower on irrigated lowland farms than it had been when the same parcels were not irrigated, increases in cropping intensity were suf® cient to com-

270

Land Economics

pensate for labor shedding. Annual agricultural employment under irrigated conditions was higher in absolute terms than under rainfed conditions. In addition, the number of upland workers hired to work on lowland farms increased following irrigation. Upland residents who gained employment on lowland farms reduced their rates of forest clearing by small but statistically signi® cant amounts. Whether the observed gains in employment and the associated reductions in rates of forest clearing can be sustained remains unclear. In addition to endogenous forces at work in the local economy, external factorsÐ such as continued population growth, rural to rural migration, recovery from the Asian economic crisis and from El NinÄo events, and shifts in economic policy all may in¯ uence incentives for agricultural expansion and forest degradation over time. Furthermore, results from simulations based on data collected on irrigated farms also suggest that, ceteris paribus, if lowland farmers adopted a more ef® cient strategy for production on irrigated farms, the result would be a net loss of jobs and possibly greater forest pressure in the uplands. The simulation results therefore cast a somewhat cautionary shadow over the otherwise upbeat conclusions of the paper and underscore the fact that, in general, technical change in lowland agricultural production has ambiguous implications for changes in forest cover. II. A MODEL OF LOWLAND TECHNICAL PROGRESS AND UPLAND ACTIVITY The theoretical model that underlies the empirical section of the paper seeks to integrate in the most simple way possible lowland and upland farm households. The structure of the model is guided both by a desire for parsimony and by a set of stylized empirical facts drawn from the sample of farms studied below. Lowland households are viewed as agricultural in their orientation. Upland households, because of their limited agricultural capacity, are recognized as engaging in a range of income-generating activities. These include low-input and shifting

May 2001

agriculture in the uplands, forest clearing, exploitation of upland forest resources, and sales of labor in the lowlands when opportunities arise. Although the model of lowland-upland interaction does not constitute a formal general equilibrium analysis, the spirit of the undertaking is consistent with the view that rates of forest clearingÐ and reductions in those ratesÐ can be traced to incentives generated in the larger economy within which frontier farming takes place (Coxhead and Jayasuriya 1994). As noted above, increases in rural employment may originate outside agriculture. But in the context of the empirical example presented below, no upland residents were found to be working outside the agricultural sector. So the primary phenomena of interest here are the impacts of irrigation development in lowland agriculture on labor demand, and the willingness of upland farmers to shift their labor from upland activities (where returns to farming and clearing forest are relatively low) to lowland farms (where increases in agricultural productivity and income generate demand for labor). The analysis begins with lowland farms and then traces policy and technology changes from lowland farms to the forest margin, via direct and indirect in¯ uences on upland households. Technical Change and Lowland Agricultural Production

In the case of lowland farms the conceptual goal is to study the impact of irrigation development on resource allocation and commodity demand. The analysis builds on a standard model of household production, which integrates producer, consumer, and worker decisions (e.g., Singh, Squire, and Strauss 1986). For convenience, production and consumption decisions are assumed to be separable. This implies a recursive solution to the household’ s problem. For the current study this approach is justi® ed by empirical facts: the lowland households studied below face exogenously determined prices; they utilize credit and have access to markets for all inputs and products; and they are, at least in years without aberrant rainfall patterns, net sellers of food.

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

To proceed, let us assume the lowland household maximizes utility, which is de® ned as consumption of an agricultural good (c a ), a composite ``forest product’ ’ consisting of charcoal, fuelwood, building materials, etc. (c f ), and leisure (c l ). Prices of these goods are p a, p f , and w, where the latter re¯ ects the opportunity cost of leisure time. Let x de® ne a k 3 1 vector of agricultural inputs and let p x de® ne the corresponding vector of factor prices. Then the household’ s problem is to maximize utility subject to a full income constraint and a time constraint. The problem is de® ned as: Max u(c a , c f , c l )

[1]

271

k 1 1 equations for factor demands: xi 5 xi (p a ,p x ,w;q),

i 5 l to k,

l 5 l(p a ,p x,w;q),

[8] [9]

and three equations for commodity demands: c j 5 c j (p a ,p f ,w,y), j 5 a, f, l.

[10]

Several points regarding this system of equations are worth highlighting. First, consider the impact of changes in technology on household income and commodity demand. By equation [6] we can write this impact as:

c a, cf, cl

¶y

subject to:

¶q

p* 1 wl 5 p f cf 1 wc l 1 p xx

[2]

c l 1 l s 5 l,

[3]

where p* denotes maximum agricultural pro® t, l s represents the lowland household’ s supply of labor to agricultural production and l represents the household’ s labor endowment. The household ® rst solves the production problem, namely maximization of pro® t: p 5 p a q a 2 p xx 2 wl,

[4]

subject to the technology of production: g(q a ,x,l;q) 5 0,

[5]

where q a represents agricultural output, l represents total labor use (which includes household and hired labor, if any), and q represents the technology of production. Via standard analytical techniques one can assess household response to price and technology changes. De® ning full income as: y 5 p a q a 2 p xx 2 wl 1 wl,

[6]

the reduced form of the household problem includes one equation for agricultural supply: q a 5 q a (p a,p x,w;q),

[7]

5 pa

¶qa ¶q

2

^p i

¶xi xi

¶q

2w

¶l . ¶q

[11]

The expression in equation [11] will be positive so long as technical change is factor augmenting and all inputs are normal. Note further that if x excludes the ® xed-factor land, and if the impact of technical progress is such that it raises revenues by a suf® ciently large amount, then the impact of technical progress on the incomes of land-owning households will be ``strong’ ’ in the sense that ¶y/¶q will be large. This income gain, the magnitude of which is primarily an empirical matter, ® gures prominently in the household’ s labor response to technical change because it in¯ uences consumption of leisure. This issue is addressed below. As always, relative shifts in factor shares as a result of the technical change will depend on relative factor prices and the extent to which the technology exhibits factor bias. These too are empirical issues. Regarding commodity demands, the impact of technical change on agricultural goods and forest products can be described by: ¶c i ¶q

5

¶c i ¶y , i 5 a, f. ¶y ¶q

[12]

The impact of an income-increasing shift in technology will be positive for normal goods and negative for inferior goods. This result

272

Land Economics

provides the ® rst of two important insights regarding the impact of lowland technical change on activities at the forest margin, namely that by raising lowland farm incomes, technical progress in the lowland sector unambiguously increases demand for forest products with positive income elasticities and decreases demand for forest products with negative income elasticities. Whether an increase in demand for forest products results in deforestation depends on many factors. Some forms of forest use (such as extraction of minor forest products) are benign and sustainable and therefore do not necessarily place pressure on the forest. Other types of forest use, including timber extraction and permanent conversion of forest to agricultural land may directly lead to deforestation. Further complicating the analysis of demand-driven impacts on the forest is that the incentive for upland households to provide forest products rests upon the relative returns to factors used in their production. Many forest products generate extremely low returns, and thus the extent to which an increase in the price of forest products increases their supply depends on the opportunity cost of labor among upland residents. A secondÐ and from the perspective of this paperÐ more important insight into how lowland technical change might affect activities at the forest margin comes from examining the impact of technical progress on the agricultural wage and on patterns of lowland labor demand. To study this, it is necessary to consider two interacting forces: (1) the extent to which technical change increases or decreases demand for labor on lowland farms, that is, ¶l/¶q, and (2) the extent to which technical change increases demand for consumption of leisure by lowland household members, that is, ¶cl /¶q. In general, a rising wage tends to reduce incomes on farms that purchase labor, and therefore discourages consumption of leisure. However, if technical progress increases the returns to land owned by a household, this increase in income may outweigh the reduction in income associated with a higher wage bill and lead to greater consumption of leisure. To see

May 2001

this, one can differentiate the leisure demand equation with respect to the technology parameter. The impact of technical progress on the demand for leisure is: ¶cl ¶q

5

¶c l ¶y ¶c l ¶w 1 , ¶y ¶q ¶w ¶q

[13]

where the ® rst term on the right hand side of equation [13] is an income effect (leisure increasing) and the second term is a price effect (leisure reducing). To gain some insight into the net effect of these opposing forces, ® rst differentiate the demand for leisure with respect to the wage and decompose the result via the Slutsky equation. This yields: ¶c l ¶w

5

¶c l ¶w

)

2 (l 2 l s ) u5u

¶cl ¶y

,

[14]

where the term (l ± l s ) is positive for households that are ``net buyers’ ’ of labor and is negative for households that are ``net sellers’ ’ of labor (Sadoulet and de Janvry 1995). To the former, wages are a cost of production; to the latter, wages represent income. Now use [14] and the result for ¶y/¶q from [11], to rewrite [13] as:

1

¶c l ¶c l ¶q pq 5 2 ¶q ¶y ¶q 1

1 )

¶w ¶c l ¶q ¶w

^p i

¶xi xi

¶q

2w

2 (l 2 l s ) u5u

2

¶c l ¶y

¶l ¶q .

2 [15]

The ® rst term on the right hand side of equation [15] is the change in consumption of leisure that results from a change in agricultural income. This is positive. Its magnitude will depend on factor payments and the amount by which output and revenue increase as a result of technical progress. If agricultural returns, net of variable input costs, accrue to land, then land-owning households might enjoy a relatively large increase in income as a result of technical progress. Indeed, this is a major motivating force behind public invest-

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

ments in agriculture and land-titling and agrarian reform programs.3 The second term on the right hand side of [15] is the change in consumption of leisure associated with an increase in the agricultural wage. This is unambiguously negative for farms that hire labor: income and substitution effects both discourage consumption of leisure.4 Thus whether lowland technical progress increases rural employment depends largely on three forces: (1) the factor requirements of the technology; (2) the change in income on lowland farms; and (3) the propensity of lowland farm households to consume leisure. When a technology uses labor intensively, increases income, and promotes consumption of leisure, then technical change will increase local demand for labor. In all other instances, the net impact of technical progress on labor demand depends on the empirical signs and magnitudes of these effects. To summarize the gist of the argument thus far, consider an initial equilibrium that is characterized by an existing technology used by a representative lowland farm. Keeping with the empirical example presented below, this existing technology is taken to be rainfed rice production. The lowland farm may rely solely on family labor, orÐ as is more likelyÐ it may use a combination of family, shared, and hired labor. This lowland farm uses hired labor up to the point where the value of the marginal product of hired labor equals the wage.5 Starting from this initial equilibrium, suppose an innovation takes place in the lowland agricultural sector, for example development of an irrigation system to provide water storage and delivery. If this innovation raises the value of labor in production, it will tend to boost labor use. Furthermore, if irrigation facilitates multiple cropping during a calendar year, annual labor demand may rise. This may occur because more labor is used during a single cropping season, because labor is used during times of the year that it was formerly not used, or both. In other words, the technical change may increase effective labor demand, that is, the total amount of labor used on a hectare of land in a calendar year. As empirical results reported below illustrate, this distinction is

273

important. It is clearly possible for irrigation to be labor saving, and to thereby reduce perhectare amounts of labor used in a particular cropping season, but to nevertheless increase the amount of labor used in a calendar year by increasing cropping intensity. Note further that, even if technical change reduces effective labor demand, it could increase the consumption of leisure in a lowland household if it increases income by a suf® cient amount. Higher consumption of leisure will lead the lowland household to substitute hired labor for family labor. Greater demand for labor will in turn raise the wage rate because a higher wage is required to draw new workers away from alternative activities. This logic leads to the second main implication of the model. By raising demand for leisure and reducing the supply of household labor used in agricultural production, technical progress in the lowland sector may raise demand for labor and thereby put upward pressure on the agricultural wage. If lowland technical progress increases lowland incomes by a large amount, workers will be hired to replace family labor. To the extent newly hired workers would otherwise engage in activities at the forest margin, the local labor market provides a conduit through which improvements in lowland agriculture may in¯ uence rates of forest clearing. 3 Consistent with this view, the econometric results presented below show that production and agricultural revenue were higher on owner-occupied parcels of land than on rented parcels. Evidence also suggests owners were more likely to hire workers than were renters. 4 For a farm household that sells labor, the pure income effect of a higher wage (the second term on the right hand side of equation [14] can in some instances dominate the substitution effect resulting from a higher price of leisure. This is because when the level of consumption of leisure is low, the marginal utility associated with consuming an additional unit of leisure is suf® ciently high that it outweighs utility gains from consumption of other goods. In the sample studied in this paper all of the lowland households were either self-suf® cient or net purchasers of labor. 5 In many cases hired labor may be an imperfect substitute for family labor, especially if supervision is dif® cult or costly. This can dampen incentives to replace family labor with hired labor. For an example from a different location in the Philippines, see Coxhead, Shively, and Shuai (in press).

Land Economics

274

Upland Farms and Labor Allocation Decisions

The analysis now shifts to the uplands. To fully understand how lowland technical progress in¯ uences rates of forest degradation it is necessary to build a conceptual link between the lowland and upland sectors. In this section, arguments regarding labor allocation are again developed somewhat formally in order to highlight the mechanism that leads upland households to reallocate labor. Consider the labor allocation decision of a representative upland household. For sake of simplicity, and in keeping with the relatively rudimentary and low-input nature of the upland economy, it is assumed that (1) labor is the only variable input allocated by upland household; (2) the pool of labor available in the household is homogenous; and (3) labor is allocated to maximize economic returns. The upland household can engage in three income-generating activities: upland agriculture, forest exploitation (including land clearing for agriculture), and off-farm work. For convenience, the upland agricultural product may be thought of as essentially identical to the product of lowland production (though it need not be). In terms of notation, upper case letters are used for the upland household; otherwise de® nitions remain as above. Agriculture is identi® ed by the subscript a, forest use is identi® ed by subscript f, and off farm work is identi® ed by subscript o. Returns to upland activities are determined by the price of output associated with the activity and the level of labor effort devoted to the activity. No other inputs are used. The upland production functions, G(La ) for agriculture and F(Lf ) for forestry are assumed to be concave. Labor is a necessary input, and the production functions exhibit diminishing returns to use of labor. When working on a lowland farm, an upland worker receives the wage w. This wage is set competitively, in the local labor market, and depends, via the implicit function de® ned by equation [9], on the technology of lowland production.6 The lowland technology is again represented by q. De® nedin this way, the uplandhousehold’ s income-generation problem is to maximize: p 5 paG(La ) 1 pF F(Lf ) 1 w(q)Lo ,

[16]

May 2001

subject to: La 1 Lf 1 Lo 5 L.

[17]

As stated, the upland household’ s problem assumes production and consumption decisions are separable. This assumption, of course, is somewhat problematic in the case of upland households, many of which cannot readily participate in markets. Nevertheless, because upland households participate in the local labor market, it is logical to assume that labor allocation decisions are guided by the opportunity cost of labor, as governed by the lowland agricultural wage. It is important to recognize, however, that supply responsiveness in upland households may not be frictionless due to failures in closely related product or factor markets. The focus at this stage is on the relationship between technical progress in lowland agriculture and the amount of labor devoted to the forest-degrading activity; that is, the sign and magnitude of ¶Lf /¶q. If an upland household engages in all activities, then a household maximum occurs where the value of the marginal product of labor is equal for each activity:7 6 Wages at the ® eld site depart from those prevailing elsewhere in the Philippines. Wages are determined locally due to remoteness of the area and relatively low population density in surrounding areas. Agricultural wages depend on the nature of tasks performed, the perception of worker productivity (upland workers tend to be paid slightly less for the same task than lowland workers), and the season in which they are performed. Planting and harvest periods in the lowlands tend to coincide with those in the uplands. As a result, the wage adjusts throughout the year in response to the opportunity cost of labor. To the extent irrigation facilitates cropping during the dry season, some of the gains in employment documented for upland workers arise from reductions in the planting of the dry season crop (corn), rather than the wet season crop (rice). 7 Not all households engage in all activities, of course. If a household specializes (either by choice or due to resource constraints), an appropriate modi® cation of equation [18] may be required to account for inequalities. The deviation of shadow prices from market prices will depend, in general, on transaction costs, risk aversion, and the covariance of risks across activities (for a discussion, see Sadoulet and de Janvry 1995). A household’s apparent failure to equate marginal returns may actually re¯ ect attempts to equate shadow values. This will especially be the case if production and consumption decisions are made jointly.

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

pa

¶G(La) ¶La

5 pF

¶F(Lf ) ¶Lf

5 w(q).

[18]

Despite the limitations acknowledged above, the logic behind equation [18] provides a useful framework for investigation. For example, provided work is available, a household that reports no off-farm employment (i.e., Lo 5 0) but some amount of upland agricultural production (i.e., La . 0), reveals that the available wage is less than the expected return to labor allocated on the farm (i.e., w , ¶G/¶La ). Similarly, a household that clears forest to establish new agricultural areas (Lf . 0) may be responding to a low rate of return to farming an existing parcel of land. To proceed, consider an initial equilibrium characterized by an existing technology in the lowlands and a representative upland household engaging in each upland activity. Suppose innovation takes place in the lowlands. This can be represented by a shift in the lowland technology parameter from q to q¢. Let us assume that the change raises lowland labor demand and increases the agricultural wage. This higher wage produces a temporary disequilibrium in the upland household’ s optimal labor allocation pattern since pa

¶G(La) ¶La

5 pf

¶F(Lf ) ¶Lf

, w(q¢).

[19]

An increase in the agricultural wage encourages the household to re-equate marginal returns to labor. If the production functions G(La ) and F(Lf ) are concave, a new equilibrium can be obtained by reducing levels of La and Lf (so that ¶G/¶La and ¶F/¶Lf rise). In theory, the amount of labor allocated to all alternative activities will fall in response to an increase in the wage. The amount by which these allocations fall depends on the curvature of the respective production functions. In general, whether upland forestdegrading activities decline in response to a technology shift in the lowlands depends on two factors: ® rst, the extent to which the technological change precipitates an increase in the wage (i.e. ¶w/¶q); and second, the de-

275

gree to which a change in the opportunity cost of upland laborÐ as re¯ ected in the wage rateÐ precipitates a reallocation of effort away from the forest margin (i.e. ¶Lf / ¶w). The impact of lowland technical progress on rates of upland forest clearing will therefore depend on how technical progress affects factor intensities and factor payments, as well as the magnitudes of income elasticities of demand for products provided by the upland sector. From a trade-theoretic perspective, the upland sector may be thought of as a small Hecksher-Ohlin economy. The impact of lowland technical progress on upland activity depends on both the direct impacts arising in the labor market and the indirect impacts arising in commodity markets. Growth in lowland production (as a result of irrigation) tends to pull labor out of upland production.8 This is simply a Rybczynski cost effect. But to the extent the growth in lowland agricultural production increases incomes throughout the lowland economy, technical progress will increase demand for upland products (and simultaneously reduce incentives for upland households to abandon forestdegrading activities). The bundle of upland products may include products with high-income elasticities (such as temperate-zone vegetables) and products with low-income elasticities (such as charcoal, fuelwood, or rough building materials). As Jayasuriya (2001) argues, as long as the complete bundle of forest products coming out of the upland sector is not highly income elastic, the labor-pull effect of lowland technical progress is likely to dominate the commodity demand effect. In other words, faster growth in the lowland economy, fueled by technical progress, will tend to reduce forest clearing by pulling labor resources out of the uplands. The remainder of the paper investigates this conjecture empirically, using data from lowland and upland farms to measure the impact 8 Rising wages also will tend to draw labor from neighboring lowland communities, or even induce migration from a distance, but only to the extent information about the wage premium is known, and only if the premium is suf® cient to compensate for transport and dislocation costs.

Land Economics

276

of irrigation on lowland labor demand, and the impact of labor market opportunities on the labor allocation decisions of households living along the forest margin. III. DATA The data for this study come from farms in two lowland communities and two upland communities in southern Palawan, in the Philippines. The data were collected in 1996 and 1997 and cover agricultural production during the periods June 1995 to May 1996 and June 1996 to May 1997, respectively. The lowland dataset includes information from 108 farms. Prior to irrigation, all of the lowland farms were rain fed. At the time of the second survey, the process of irrigating lowland farms was not yet complete; 53 farms were irrigated and 45 farms remained rain fed. Together, the set of lowland respondents constitutes a 35% sample of the two lowland communities. The upland dataset consists of information from 104 upland farms adjacent to the lowland study sites. The upland dataset constitutes a 30% sample of the two upland communities. All upland households lived on or near the forest margin. Previous research from the upland sites reveals strong positive correlation between levels of household poverty and the probability of forest use (Shively 1997). Although in many frontier areas of the tropics, especially in Latin America, forest clearing for livestock grazing is important (e.g., Faris 1999), livestock do not play an important role in upland production strategies at this site. Garcia et al. (1995) and Martinez and Shively (1998) provide additional details regarding the site. Columns 1 and 2 of Table 1 illustrate some of the differences observed between irrigated and rainfed lowland farms. Although lowland farms were broadly similar in terms of demographic features, average farm size differed signi® cantly by irrigation status. The average irrigated farm occupied 2.6 hectares and the average rainfed farm occupied 5.1 ha. This indicates that some rainfed farms were reduced in size following irrigation. With the exception of hired labor, all means for irrigated farms reported in Table 1 were signi® cantly different from means for rainfed

May 2001

farms (at a 90% con® dence level). As expected, average yield among irrigated farms (3,639 kgs/ha) was higher than among rainfed farms (3,200 kgs/ha). Furthermore, due to unreliable water supplies during the dry season prior to irrigation, rainfed farms tended to produce only one crop per year whilst irrigated farms produced two crops per year, on average. As a result, income per hectare was signi® cantly higher on irrigated farms (41,651 Pesos/hectare/year compared with 25,921 P/ha/yr on rainfed farms). Irrigated farms spent 80% more on pesticides than rainfed farms (1,656 P/ha/yr vs. 917 P/ ha/yr). However, they used less labor per hectare overall, and less family labor per hectare than rainfed farms (37 and 13 mandays per hectare compared with 43 and 20 man-days, respectively). The latter pattern indicates irrigation induced a reduction in overall labor use per hectare and a release of family labor. The ® nal columns of Table 1 contain data from the sample of upland farms. For upland households, data in Table 1 do not distinguish between pre-irrigation and postirrigation periods but rather between a household’ s employment status following irrigation. It is important to point out that, both before and after irrigation, all employment reported by upland residents originated on lowland farms located in the study area. As data in Table 1 indicate, following irrigation 83 households (80% of the upland sample) reported earnings from work on lowland farms. The average wage share of total income in the upland sample (including the imputed value of output consumed at home) was 0.10 prior to irrigation and 0.35 following irrigation. Households engaging in work on lowland farms had slightly smaller farms, on average, than those that did not (2.0 ha vs. 2.5 ha). They also had lower per capita incomes (3,244 P/yr vs. 4,586 P/yr). Employment on lowland farms certainly helped to augment income for households with limited agricultural capacity, but as Shively and Martinez (2001) point out, wage earnings were not suf® cient to close the income gap between upland household with and without employment. As a result of low agricultural capacity, upland households with wage in-

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

277

TABLE 1 Characteristics Farms in the Sample Uplanda Lowland

Farm size (hectares) Household size (members) Income per capitab (pesos/person, 1996) Income per hectareb (pesos/ha, 1996) Tenure security (% w/title) Rice yield (kgs/ha) Effective cropping (crops/yr) Fertilizer use (kgs/ha or %) Pesticide use (pesos/ha or %) Total labor use (days/ha) Family & shared labor (days/ha) Hired labor (days/ha) Forest clearing (% of households) Charcoal and fuelwood sales (% of households) Area of forest cleared (ha/yr, average) Number of farms

With off-farm work

Without offfarm work

Rainfed

Irrigated

4.2

2.5

2.0

2.5

4.8

5.8

4.9

4.8

25,364

22,604

3,224

4,586

25,921

41,651

6,783

7,302

78% 3,200

48% 3,639

1.2

42% 1,733

1.9

43% 1,833

1.0

1.0

180

157

29%

33%

917

1,656

14%

10%

43

37

Ð

Ð

20

13

Ð

Ð

23

25

Ð

Ð

Ð

16%

15%

Ð

18%

23%

Ð

Ð

Ð

Ð 45

0.18 53

83

0.10 21

a

For upland households, data in this table are pooled across years and do not distinguish between pre-irrigation and post-irrigation periods. b At the time of the survey $1 US 5 25 pesos.

come were still more likely to report activities with relatively low rates of return. As the ® nal rows of Table 1 indicate, this included a slightly greater likelihood of forest clearing (16% vs. 15%) and a higher probability of charcoal and fuelwood sales (27% vs. 23%). Upland households with off-farm employment also reported clearing larger areas of forest than those without off-farm work (0.18 hectare per year vs. 0.10 hectare per year).9 With the exception of some limited forays into cashew and mango production, none of the upland households reported engaging in production of crops that could be construed

as destined for export markets. Nevertheless, the decisions by upland households to plant these and other cash crops (such as corn) have historically responded to changes in relative prices (Shively 1998). It therefore 9 Based on rates of forest clearing reported by respondents and the estimated 30% sample frame, newly cleared area represented about 7% of all cropped area in the uplands. However, not all area cleared constitutes destruction of primary forest. Data from the site suggests that about 30% of newly planted area in 1996 had been virgin forest in the preceding year, 46% had been degraded forest and shrubland, and 24% had been grassland.

Land Economics

278

seems very likely that some of the observed rates of forest clearing are driven by factors determined outside the local economy. IV. RESULTS Patterns of Labor Use on Lowland Farms

Overall patterns of labor use under irrigated and rainfed conditions display some parallels. For example, owners tended to hire labor at a greater rate than renters, regardless of irrigation status or farm size. Similarly, per-hectare labor use (especially of family and shared labor) was greatest on small farms, and decreased with farm size, regardless of irrigation status. The latter re¯ ects higher rates of hand-tractor and chemical use on large farms. When employment is decomposed by source and farm size, patterns of labor use differ between irrigated and rainfed groups. For all farm sizes, levels and proportions of family and shared labor were lower under irrigated conditions than under rainfed conditions. In contrast, the use of hired upland labor was higher under irrigated conditions, regardless of farm size. The data thus very clearly show that during the shift from rainfed to irrigated operations, family labor was released from rice production and upland workers were hired as substitutes. Although many lowland workers continued to be hired to work on lowland farms following irrigation, the bulk of post-irrigation employ-

May 2001

ment gains accrued to upland residents. Martinez and Shively (1998) report that, in the study area, upland workers constituted 4% of total lowland labor prior to irrigation and 12% of total lowland labor after irrigation. The data on employment reveal no strong seasonal patterns, although there was a slight tendency for hiring to expand in the dry season. This may indicate a greater willingness by upland farmers to forgo corn production (a dry season cash crop) than rice production (a wet season food crop). Summary data regarding labor use and cropping intensity on lowland farms are presented in Table 2. Several important patterns are displayed in these data. As stated above, the amount of labor used under irrigated conditions in a given season was lower than that used under rainfed conditions (37 vs. 43 mandays per hectare, per cropping). That is, a decrease in labor demand per hectare accompanied the shift from rainfed to irrigated production. However, after one accounts for the higher cropping intensity on irrigated farms (1.9 crops per year vs. 1.3 under rainfed conditions), total annual labor use was approximately 36% higher with irrigation. These ® gures translate into total annual labor use of 70 days/ha on irrigated farms and 55 days/ha on rainfed farms. In other words, the shift from rainfed to irrigated production resulted in a 27% increase in total labor use per hectare. Data in Table 2 further show that family labor used in production

TABLE 2 Labor Use, Cropping Intensity, and Changes in Labor Use

Family labor (days/ha) Upland labor (days/ha) Total labor (days/ha) Cropping intensity (crops/yr) Effective demand (days/ha/yr)

Rainfed (observed)

Irrigated (observed)

Change from rainfed case a

18.2

12.7

230%

1.6

4.3

1269%

42.7

37.1

213%

1.3

1.9

147%

55.1

70.1

127%

a Data refer to observed means for the sample of farms without irrigation and the sample of farms with irrigation, irrespective of the period of observation.

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

279

TABLE 3 Forest Conversion Indicators Before and After Irrigation a Pre-irrigation Households reporting forest clearing (%) Average forest area cleared (hectares/year) Area in rice (hectares) Area in corn (hectares) Lowland employment (days/household/year) Average lowland agricultural wage (pesos/man-day) Agricultural wage income c (pesos/household/year)

Post-irrigation

18%

12% b

2.5

1.9 b

0.95

0.94

1.20

1.05

18

44b

45

75b

1,150

3,226 b

a ``Pre-irrigation’ ’ refers to the period prior to irrigation: all lowland farms were rain fed. ``Postirrigation’ ’ refers to the period after which lowland irrigation development had taken place. At the time of the survey some lowland farms remained rain fed. b Indicates means are signi® cantly different at a 95% con® dence level. c Refers only to households who had employment on lowland farms.

fell following irrigation. As noted previously, the main bene® ciaries of this shift were upland households: the annual number of days of upland labor hired per hectare rose 268% (albeit from a small base) during the transition to irrigated production. Implications for Patterns of Activity on Upland Farms

The data in Table 3 highlight reported activities and outcomes on upland farms before and after irrigation. Rates of forest clearing, average area cleared, and wage income all differ in the pre- and post-irrigation samples by statistically signi® cant amounts. Data point toward a small but signi® cant reduction in the proportion of households reporting forest clearing before and after irrigation. The proportion of households reporting that they cleared forest during the previous year fell from 18% before irrigation to 12% after irrigation. Also signi® cant is the change in reported area cleared. In the pre-irrigation sample the average area reported cleared (by those reporting land clearing) was 2.5 ha. In the post-irrigation sample the corresponding ® gure was 1.9 ha. Taken together, these statistics indicate that the annual rate of forest

clearing declined 48% between the preirrigation and post-irrigation periods.10 Modest changes in upland agricultural practices were also observed following irrigation. Although the reported area planted to rice (the staple upland crop) did not change, the average area planted to corn (a cash crop), fell slightly, from 1.2 ha to 1.1 ha. Wages from employment on lowland farms more likely served as a substitute for cash income from corn production, than as a substi-

10 It is dif® cult to ® nd a compelling competing explanation for the reduction in cleared area. No major expansion of cash crops took place immediately prior to irrigation and no fundamental change in economic policy occurred that strongly in¯ uenced crop prices over the period. Although the Philippines was hit by the Asian crisis and spontaneous currency devaluation beginning in mid-1997, it seems unlikely that this could have led to the reduction in forest clearing observed in the 1997 data. If anything, depreciation of the peso would have made export crops more lucrative, but the data show a reduction in forest clearing post-depreciation, not an increase. The only remaining possibility is that the reduction in rainfall associated with El NinÄo weather patterns might have discouraged agricultural activities during late 1996 and early 1997, and thereby reduced demand for new parcels of land. Follow-up data from the area will be needed to gauge the importance of these external effects, however.

280

Land Economics

tute for the staple crop.11 The sharp spike in wages that followed irrigation suggests upland labor supply may be constrained by the marginal value product of labor on upland ® elds. Although upland households exhibit some willingness to abandon their main cash crop (corn) in favor of wage labor, it seems likely that a much larger wage increase would be necessary to precipitate upland abandonment of the main upland staple crop (rice). In the long run, a key determinant of forest pressure in the area will be the way in which new income (both on lowland and upland farms) gets used (e.g., whether for chain saws, livestock, or fertilizer). Moreover, as Carter and Wiebe’ s (1990) analysis from Kenya demonstrates, the distribution of such gains may have important implications for not only forest pressure, but also for the trajectory of agricultural productivity and agrarian structure in the area. Overall welfare changes for upland households cannot be completely assessed with these data. However, data on days of employment and average wage income support the hypothesis that lowland irrigation development improved welfare for upland households. The number of upland households with wage employment rose following irrigation; the average number of days of employment on irrigated farms rose considerably; and the average reported daily wage was two-thirds higher after irrigation. These patterns corroborate local reports that the labor market experienced a ``boom’ ’ following irrigation. Aggregating across all upland households, data show that the aggregate amount of wage income reported by the sample of upland respondents increased nearly three-fold following irrigation. V. OPTIMAL LABOR USE AND FORECASTS OF POTENTIAL LABOR SHEDDING Although these empirical ® ndings are encouraging, one needs to exercise caution in concluding that lowland irrigation will unambiguously reduce forest clearing, even where scope remains for expanding the lowland area under irrigation. In particular, data presented above reveal only the initial, short-

May 2001

run impact of irrigation on patterns of labor demand on lowland farms. To investigate whether these employment gains might be sustained over time, the analysis now moves beyond discussion of observed patterns to ascertain whether the logic of pro® t maximization might eventually lead lowland farmers to use less labor-intensive combinations of inputs. The goal of this analysis is to predict whether logical changes in lowland farmer practices might reverse the bene® cial impacts of irrigation highlighted above. To get at these issues, simulation results based on the empirical data are used. The analysis follows a standard approach to measuring and forecasting factor demands. To begin, production functions were formulated and estimated econometrically. Regression results were then used to derive pro® t-maximizing input levels conditional on observed input and output prices. By estimating optimal labor use under irrigated conditions, and by comparing these results to input levels observed on rainfed and irrigated farms, it is possible to draw inferences regarding changes in labor demand and rates of forest clearing that could arise over time in response to irrigation. Table 4 contains regression results from ordinary least squares (OLS) and stochastic frontier (SF) production functions. In both regressions the logarithm of per hectare output serves as the dependent variable and the functional form used is Cobb-Douglas. Both regressions contain identical sets of independent variables, with the exception of two error parameters that appear in the SF model. The SF model was estimated under the assumption of an exponential error structure. The logic behind using these speci® cations is that observed labor use may not re¯ ect the most pro® table or ef® cient use of labor and that, over time, lowland farmers may adopt factor shares that more closely resemble the most pro® table production strategies. 11 Data reported by Shively (1998) show an inverse relationship between agricultural capacity and rates of perennial crop adoption in the area. These results suggest poverty alleviation might not only reduce rates of forest clearing in an upland community, but also encourage a transition to more sustainable perennial-based cropping patterns on upland farms.

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

281

TABLE 4 P roduction Function Results OLS

Stochastic Frontier

6.7513 (0.3997) 0.0988 (0.0621) 0.1309 (0.0648) 0.0386 (0.0215) 0.2629 (0.0534) 20.0259 (0.0111) 0.1519 (0.0574

7.1929 (0.3408) 0.0577 (0.0513) 0.1395 (0.0629) 0.0324 (0.0198) 0.2204 (0.0468) 20.0209 (0.0096) 0.0783 (0.0526)

Independent Variables Constant Log of labor (man-days per hectare) Log of Fertilizer (kgs per hectare) Log of Pesticide (Pesos per hectare) Season {0,1} Farm size (hectares) Tenure (0 5 rented; 1 5 owned) Error parameters q Ð

s2 Ð

R2

0.36

4.0158 (0.8817) 0.1313 (0.0328) Ð

Log-likelihood

25.76

23.38

Number of observations

105

105

Clearly, otherÐ and more ¯ exibleÐ formulations of the production function are possible. The goal here is merely to explore alternative views of labor demand that are supported by the data. As the results in Table 4 indicate, all parameter estimates in the regressions have the expected signs and most are statistically signi® cant at the 90% con® dence level or above. Labor, fertilizer, and pesticide all contribute positively to yields. The wet season is associated with higher yields. The negative sign on farm size suggests smaller irrigated farms in the sample were either more ef® cient in their production or occupied more productive land. Owner-occupied farms reported signi® cantly higher yields than rented farms. The source of the difference, however, remains unobserved. The regressions exhibit strongly diminishing returns to input use. An important difference between the OLS and SF results arises with respect to the relative

importance of labor and fertilizer: the marginal impact of labor is substantially lower in the frontier speci® cation, while the marginal impact of fertilizer is somewhat higher. Based on a likelihood test suggested by Pollak and Wales (1991), the SF model is preferred to the OLS model on purely statistical grounds.12 12 The asymptotic criterion for model selection is a choice between nonnested hypotheses. Let H1 represent the hypothesis that the OLS model is correct and let H2 represent the hypothesis that the SF model is correct. Let n1 5 the number of parameters in the OLS model and let n2 5 the number of parameters in the SF model. Finally, let L1 5 the log-likelihood value of the OLS model and let L2 5 the log-likelihood value in the SF model. The test proposed by Pollak and Wales (1991) is based on a likelihooddominance criterion (LDC) and advises the following grounds for model selection:

LDC prefers H1 if L2 2 L1 , [C(n 2 1 1)] ± C(n 1 1 1)] /2 LDC prefers H2 if L2 2 L1 . [C(n 2 2 n 1 1 1)± C(1)] /2

Land Economics

282

May 2001

TABLE 5 P redicted Changes in Labor Demand a Effective Demand (days/ha/yr) Rainfed (observed) Irrigated (observed) Predited (OLS) Predicted (SF) Predicted (SF, full utilization) a

Change from Rainfed (%)

55.1

Ð

70.1

127%

62.4

113%

43.5

221%

46.0

216%

Results are based on a representative three-hectare, owner-occupied farm.

Regression results reported in Table 4 were used to derive pro® t-maximizing input levels, conditional on input and output prices, to determine input levels that would constitute pro® t-maximizing optima for a representative irrigated farm. Derived input levels are reported in Table 5, along with levels observed on rainfed and irrigated farms. In interpreting results in Table 5, it is important to recall data from Table 2 showing that, for this sample of lowland farms, the shift from rainfed to irrigated production led to a 13% reduction in labor demand per hectare, but that an increase in cropping intensity boosted effective (annual) labor demand per hectare by 27%. In contrast, results in Table 5 suggest that, on irrigated farms, observed levels of labor use exceeded levels derived from the OLS and SF models. In other words, irrigated farms appear to have been operating below pro® t maximizing levels, and the shift from observed to ``optimal’ ’ factor proportions could be expected to generate additional labor shedding.13 How great might the reduction in labor use be? In the case of the OLS parameter estimates, effective labor demand falls to 62.4 days/ha/yr. This still represents a 13% increase over the rainfed baseline. But in the case of the SF parameter estimates, effective labor demand is predicted to fall to 43.5 days/ha/yearÐ a 21% reduction in overall labor demand from the rainfed baseline. In other words, econometric evidence shows that if lowland farmers adopted a

more ef® cient strategy for production on irrigated farms than was observed, the result would be a net loss of jobs. As the ® nal row of Table 5 indicates, full utilization of irrigation in the dry season tends to offset these per hectare declines, though not entirely. VI. CONCLUSION This study examined whether technical progress in lowland agriculture led to increases in agricultural productivity and wages, and whether these changes, in turn, led to greater employment opportunities for low-income households living near the forest margin. The analysis showed that, compared with outcomes for rainfed conditions, irrigation development in lowland agriculture increased the probability of employment for upland residents, more than doubled the number of days of employment for those where C(n) is the critical value of the chi-square distribution with n degrees of freedom at some ® xed signi® cance level. In the case of the models reported in Table 4, relevant data are n 1 5 7, n 2 5 9, L1 5 25.76, L2 5 23.38. Since L2 2 L1 5 2.38 exceeds the 95% critical bound [C(n 2 2 n 1 1 1) 2C(1)] /2 5 1.99, the test indicates a statistical preference for the SF model. 13 As noted in the text, the regression results from Table 4 were used to parameterize a pro® t function for irrigated operations. One might question whether pro® t maximization accurately represents the goals of farmers. Martinez and Shively (1998) argue the observed increase in labor use accompanying irrigation could re¯ ect other considerations by farmers, including risk aversion.

77(2)

Shively: Agricultural Change, Rural Labor Markets, and Forest Clearing

working on lowland irrigated farms, and increased the wage income of farms in the upland sample by a factor of three. Importantly, these short run changes coincided with reallocations of time away from forest clearing and hillside farmingÐ especially of annual cash cropsÐ in the uplands. This ® nding is important because it con® rms that under the right agronomic and demographic conditions, lowland agricultural intensi® cationÐ whether brought about through irrigation development or other meansÐ can reduce pressure on forests and improve environmental outcomes in marginal upland areas. Despite these fairly encouraging ® ndings, the ® nal section of the paper suggests caution in their interpretation. While irrigated farms exhibited fairly intensive use of fertilizers and pesticides, additional reductions in labor use could occur. Simulations based on econometric results indicate labor use would likely fall under pro® t maximization, perhaps by as much as 21% compared with rainfed production. This would be the case, for example, if lowland farmers adopted factor proportions consistent with pro® t maximization at the production frontier. It is also worth bearing in mind that the upland area studied in this paper is physically adjacent to the lowland area. As a result, generalizing these ® ndings to locations where larger distances separate lowland and upland areas might be problematic: the opportunity cost of travel for upland households could discourage reallocation of labor from upland to lowland activities. Finally, it is important to point out that, because factor substitution is driven in part by factor costs, government policies that reduce costs for pesticides or machinery without taking account of the favorable environmental impacts of more labor-intensive production could seriously undermine the bene® cial impacts of rural employment reported here. In contrast, policies that encourage the use of rural laborÐ either explicitly, or implicitly through improvements in labor productivityÐ could prove to be powerful tools to reduce rates of forest clearing. Given the bene® cial economic, environmental, and distributional impacts associated with drawing labor away from the forest margin, such policies could likely be justi® ed on grounds

283

of ef® ciency, environmental protection, and equityÐ a win-win-win situation. Space limitations preclude further investigation of the potential impacts of price and policy changes on observed patterns of employment and forest clearing, but based on empirical evidence from other frontier areas, these are likely to be extremely important conditioning factors over time, and warrant continued attention by researchers and policymakers.

References Angelsen, A. 1995. ``Shifting Cultivation and `Deforestation’ : A Study from Indonesia.’ ’ World Development 23 (10): 1713± 29. Barbier, E. B., and J. C. Burgess. 1996. ``Economic Analysis of Deforestation in Mexico.’ ’ Environment and Development Economics 1 (2): 203± 40. Carter, M. R., and K. D. Wiebe. 1990. ``Access to Capital and Its Impact on Agrarian Structure and Productivity in Kenya.’ ’ American Journal of Agricultural Economics 72 (5): 1146± 50. Castillo, L., F. Gascon, and S. Jayasuriya. 1983. ``Off-farm Employment of Farm Households in Laguna, Phlippines.’ ’ In Off-Farm Employment in the Development of Rural Asia, Volume 2, ed. R.T. Shand. Papers presented at a conference held in Chiang Mai; Thailand, 23 to 26 August 1983. Canberra: Australian National University, National Centre for Development Studies, 133± 46. Coxhead, I. 1997. ``Induced Innovation and Land Degradation in Developing Countries.’ ’ Australian Journal of Agricultural and Resource Economics 41 (3): 305± 32. Coxhead, I., and S. Jayasuriya. 1986. ``LaborShedding with Falling Real Wages in Philippine Agriculture.’ ’ Asian Survey 26 (10): 1056± 66. Ð Ð Ð . 1994. ``Technical Change in Agriculture and Land Degradation in Developing Countries: A General Equilibrium Analysis.’ ’ Land Economics 70 (Feb.): 20± 37. Coxhead, I., G. Shively, and X. Shuai. In press. ``Development Policies, Resource Constraints, and Agricultural Expansion on the Philippine Land Frontier.’ ’ Environment and Development Economics. Cruz, M. C., C. A. Meyer, R. Repetto, and R. Woodward. 1992. Population Growth, Poverty, and Environmental Stress: Frontier Mi-

284

Land Economics

gration in the Philippines and Costa Rica. Washington, D.C.: World Resources Institute. Faris, R. 1999. ``Deforestation and Land Use on the Evolving Frontier: An Empirical Assessment.’ ’ HIID Development Discussion Paper No. 678. Cambridge: Harvard Institute for International Development. Garcia, H. N. M. et al. 1995. Soil Conservation in an Upland Farming System in Palawan: A Socio-Economic Survey. SEARCA-UQ Upland Research Project Survey Report 4. Los BanÄos: Southeast Asian Regional Center for Graduate Study and Research in Agriculture. Jayasuriya, S. 2001. ``Agriculture and Deforestation in Tropical Asia: An Analytical Framework. In Agriculture Technologies and Tropical Deformation, ed. A. Angelson and D. Kaimowitz. Wallingford: CAB International. Kaimowitz, D., and A. Angelsen, 1998. Economic Models of Tropical Deforestation: A Review. Bogor: Center for International Forestry Research (CIFOR). Kikuchi, M., and Y. Hayami. 1983. ``New Rice Technology, Intrarural Migration, and Institutional Innovation in the Philippines.’ ’ Population and Development Review 9 (2): 247± 57. Lingard, J. 1994. ``Farm Mechanization and Rural Development in the Philippines.’ ’ In Case Studies in Economic Development, ed. T. Lloyd and O. Morrisey. New York: St. Martin’ s Press; London: Macmillan Press. Martinez, E., and G. Shively. 1998. ``Irrigation, Employment, and the Environment in Southern Palawan.’ ’ Journal of Agricultural Economics and Development 26 (1± 2): 112± 35. Mattos, M. M., and C. Uhl. 1994. ``Economic and Ecological Perspectives on Ranching in the Eastern Amazon.’ ’ World Development 22 (2): 145± 58. Nickerson, D. J. 1999. ``Trade-offs of Mangrove Area Development in the Philippines.’ ’ Ecological Economics 28 (2): 279± 98. Parks, P. J., and M. Bonifaz. 1995. ``Nonsustainable Use of Renewable Resources: Mangrove Deforestation and Mariculture in Ecuador.’ ’ In Property Rights in a Social and Ecological

May 2001

Context: Case Studies and Design Applications, ed. S. Hanna and M. Munasinghe. Stockholm: Beijer International Institute of Ecological Economics. Pollak, R. A., and T. J. Wales. 1991. ``The Likelihood Dominance Criterion.’ ’ Journal of Econometrics 47: 227± 42. Sadoulet, E., and A. de Janvry. 1995. Quantitative Development Policy Analysis. Baltimore: Johns Hopkins University Press. Sandalo, R.M. 1996. ``Sustainable Development and the Environmental Plan for Palawan.’ ’ In Palawan at the Crossroads, ed. J. F. Eder and J. O. Fernandez. Manila: Ateneo de Manila University Press. Singh, I., L. Squire, and J. Strauss. 1986. Agricultural Household Models. Baltimore: Johns Hopkins University Press. Shively, G. E. 1997. ``Poverty, Technology, and Wildlife Hunting in Palawan.’ ’ Environmental Conservation 24 (1): 57± 63. Ð Ð Ð . 1998. ``Economic Policies and the Environment: the Case of Tree Planting on LowIncome Farms in the Philippines.’ ’ Environment and Development Economics 3 (1): 83± 104. Shively, G. E., and E. Martinez. 2001. ``Deforestation, Irrigation, Employment, and Cautious Optimism in Southern Palawan, the Philippines.’ ’ In Agricultural Technologies and Tropical Deforestation, ed. A. Angelsen and D. Kaimowitz. Wallingford: CAB International. Theil, R., and M. Wiebelt. 1994. ``Policies to Reduce Tropical Deforestation and Degradation: A Computable General Equilibrium Analysis for Cameroon.’ ’ Quarterly Journal of International Agriculture 33:162± 78. Western, S. 1988. ``Carrying Capacity, Population Growth, and Sustainable Development: A Case Study from the Philippines.’ ’ Journal of Environmental Management 27:347± 67. Yanggen, D., and T. Reardon. 2001. ``Kudzu Improved Fallows in the Peruvian Amazon.’ ’ In Agricultural Technologies and Tropical Deforestation, ed. A. Angelson and D. Kaimowitz. Wallingford: CAB International.

Playing Games in the Forest: State-Local Con¯ icts of Land Appropriation Arild Angelsen ABSTRACT. This paper explores possible strategic interactions between the state and local community in games of tropical forestland appropriation. Three typical cases are discussed, corresponding to a development over time of increased resource competition and market integration. The local response to more state deforestation depends on the costs, market, and behavioral assumptions, and less on the structure of the game (Cournot or Stackelberg). The state fuels local deforestation by providing infrastructure (roads) which reduces the net costs of agricultural expansion, or when markets are imperfect and local behavior determined by survival needs. The game structure is, however, important for total deforestation. ( JEL O13, Q15, Q23)

I. INTRODUCTION Property rights to tropical forests are normally not well de® ned or enforced, indeed, many frontier areas are de facto open access. Moreover, forest clearing and conversion to other uses often establish or strengthen land rights. This situation creates incentives for strategic behavior and ``land races,’ ’ and forest might be cleared beyond the point where the current net bene® ts are zero (the private property solution) for at least three reasons. First, forest is converted up to the point where the net present value of current and future net bene® ts is zero. Even though net pro® t might be negative the ® rst years, technological progress, new roads, etc. will make it pro® table in the future, and farmers must clear forest now to prevent others from claiming the land (Angelsen 1995, 1999). Second, forest is cleared to capture an expected pro® t through later sale, in a context where land prices are not solely a function of productive capacity (speculative land markets, ``rational bubbles’ ’ (Clark et al. 1993), Land Economics · May 2001 · 77(2): 285± 299 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

or Alan Greenspan’ s ``irrational exuberance’ ’ ). Third, in situations where few agents compete for forestland, and deforestation by one agent is costly to the other, there may be incentives to ``squeeze the others’ ’ by clearing more themselves. The latter phenomenon is the topic of this paper. We frame the analysis as a con¯ ict between two agents. One might put different labels on the two (coalitions of) players to make them correspond to various real-life situations. A classical con¯ ict in the history of tropical deforestation has been between the state or large commercial users supported by the state, and local, traditional user groups (e.g., Bromley and Chapagain 1984; Colchester and Lohmann 1993; Colchester 1994). The paper formalizes the interaction between the state and a local community in forestland appropriation by applying noncooperative game theoretic models. Two key questions are addressed. First, how does structure of the game (Cournot or Stackelberg) in¯ uence overall deforestation? Second, under which circumstances does more state deforestation stimulate (replace) local deforestation? There is a substantial literature using game theory to study tropical resource management problems, mainly applying binary choice models (e.g., Ostrom, Gardner, and Walker 1994; Baland and Platteau 1996). The Cournot game presented in this paper can be considered a continuous choice version of the prisoner’ s dilemma (PD) game The author is associate professor, Department of Economics and Social Sciences, Agricultural UniverÊ s, Norway, and associate scientist at sity of Norway, A the Center for International Forestry Research (CIFOR), Bogor, Indonesia. Thanks to Olvar Bergland, RoÈgnvaldur Hannesson, Stein Holden, David Kaimowitz, Karl O. Moene, Ottar Mñ stad, Anna Maria Oltorp, Karl R. Pedersen, Ussif Rashid Sumaila, Arne Wiig, and two referees for comments to draft versions of the paper.

286

Land Economics

discussed in this literature. The continuous choice model offers, however, a richer approach as one can study agents’ response to exogenous changes and the equilibrium outcome with non-cooperation and Pareto inef® ciency both before and after the change. An example of this approach applied to global carbon emissions is Sandler (1996), who uses the game theoretic framework of joint production and public goods developed by Cornes and Sandler (1996). Similarly, Alston, Libecap, and Mueller (1999) develop a model in which squatters and landowners in Brazil play a game to get property rights to land, leading to or escalating rural violence (and deforestation). While the models developed in this paper share many features with the above ones, they also differ in several respects. First, our primary focus is on deforestation.1 Second, we combine household models of agricultural economicsÐ in particular those dealing with imperfect markets and subsistenceoriented behaviorÐ with game theoretic models. This has implications for the qualitative results. Third, we note the asymmetry between the players in the forest race, in particular, the mixed effect of state deforestation on the local community: it increases land scarcity and reduces forest income, but also provides infrastructure which reduces the costs of agricultural expansion. The outline of the paper is as follows. Section 2 presents the basic elements of the models. Three different cases or games are discussed in the following sections. Section 3 focuses on a poor, isolated local community. The interaction with the state is studied as a static game with simultaneous moves (Cournot).2 Section 4 discusses a situation with higher forestland scarcity and a localled land race, that is, the local community is the leader in a Stackelberg game. Section 5 analyzes a third case with intense resource scarcity and competition, a community integrated into the regional/national economy, and the state as the Stackelberg leader. Section 6 compares the different cases, answer the two questions raised above, and discusses the possibilities for cooperation in forest management and the impact of unilateral con-

May 2001

servation efforts. The ® nal section concludes. The three cases might correspond to a possible development over time in terms of (1) increased resource scarcity; (2) increased integration of the local community in the regional/national economy; and (3) more aggressive behavior by one of the players. This suggests that at early stages state deforestation works in tandem with local deforestation. At later stages state deforestation tends to replace and squeeze local deforestation. II. PRELIMINARIES The model developed can be seen as a joint product model with a private good and a public good (or ``bad’ ’ ) (Cornes and Sandler 1996; Sandler 1996), an example of the ``tragedy of open access,’ ’ or negative externalities. We consider a given forest area (HT) which has three uses: it can be converted to agricultural land by the local community (HL), to plantations, logging or other large scale projects by the state (HS), or it can remain natural forest (HF )Ð the public good; HT 5 HL 1 HS 1 HF.

[1]

The state and local community each choose the level of its strategic variable, HS and HL, respectively. New forestland is allocated on a ® rst-come-® rst-served basis. Forest clearing might give more permanent land rights, in which case the income and cost variables should be interpreted as discounted values. Tenure insecurity could then be incorporated by reducing the discounted values (risk discounting). The local and state incomes are determined by the land area converted for their own use, as well as the remaining natural forest. Local forest bene® ts include non-timber 1 A comprehensive review by Kaimowitz and Angelsen (1998) of some 150 economic models of deforestation does not include any game theoretic models. 2 We have adopted the terminology of the duopoly literature (e.g., Shapiro 1989), and use the terms Cournot model or game for the Nash equilibrium in a oneshot game with simultaneous moves with quantities as the decision variable, and the Stackelberg model or game for the sequential moves or leader-follower game.

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

forest products and various protective functions, whereas the state bene® ts from natural forest are in the form of, for example, ecotourism and protective functions, as well as more intangible bene® ts such as existence values and a green image. The net income to the local community is given by; L 5 r(HL) 1 t(HF ) 2

#

HL

c(HT 2 HS 2 x, HS)dx 5 l(HL, HS).

287

crossderivative is not too large, the land scarcity effect will increase relative to the infrastructure effect as state deforestation increases: ¶c 2/¶ 2HS 5 c 11 2 c 22 2 2c 12 . 0. Thus in forest abundant situations we can expect the infrastructure effect to dominate. The state revenue is determined in a similar manner as local income, except that local forest clearing does not have any cost reducing effects through provision of infrastructure;

[2]

0

r(HL) is the gross income of forest clearing for agricultural production, for simplicity assumed to be a function of land area only. t(HF ) gives the net forest income as a function of total forest area. We assume decreasing returns (t1, r 1 . 0; t11, r 11 , 0). The last element in [2] gives the aggregate cost of agricultural production, which we interpret as labor. Without loss of generality, we set the real wage rate equal to one. c(HF, HS) is the marginal costs (labor units required) of land expansion. A larger primary forest area will reduce the costs as new land is more easily available (c 1 , 0, c 11 . 0). Further, state forest clearing has a cost reducing effect on the marginal costs of expansion, as it provides infrastructure, particularly roads (c 2 , 0, c 22 . 0). State appropriation therefore has contradictory effects on the costs of agricultural land expansion: the land scarcity effect raises the costs, whereas the infrastructure effect reduces them. The net effect depends both on the type of land use, and the stage in the deforestation process. State land uses which involve forest clearing and permanent conversion of the land, for example, permanent agriculture, plantations, hydropower and infrastructure developments can be expected to have relatively strong land scarcity effects. On the other hand, loggers are interested in the large trees, not the land, and we might expect the infrastructure effect to dominate (c2 2 c1 , 0). As observed throughout Asia, migrating farmers often follow in the wheel tracks and clear logged forest. Further, from the assumptions made, and assuming the

S 5 v(HS ) 1 g(HF ) 2

#

HS

h(HT 2 HL 2 y)dy 5 s(HS, HL). [3]

0

v() is the gross revenue from forest appropriation by the state, whereas g() is the state’ s bene® ts from primary forest. The bene® t functions are strictly concave (g 1, v 1 . 0; g 11, v 11 , 0). The marginal cost of forest appropriation, h(HF ), is lower the larger the area of virgin forest, but this effect is diminishing (h 1 , 0, h11 . 0).3 We identify three critical assumptions in the models. First, the strength of the land scarcity effect vs. the infrastructure effect in the local cost function, which we just have argued depends on the forest abundance (level of state deforestation) and the type of forest use by the state. Second, how are the markets functioning? As known from the agricultural household literature (e.g., Singh, Squire, and Strauss 1986), the response of farm households depends critically on the market assumptions, in particular, whether an off-farm labor mar3 We have assumed a rather narrow objective function for the state: only own income counts (a predatory state). A more general formulation is that the state gives a certain weight (g) to local income relative to state income (a developmental state), either because of altruism or due to the need for popular support and get reelected: Max W 5 S 1 gL. The response curve derived in the next section will still be downward sloping, but the location and slope will change, for example, if the land scarcity effect dominates, the curve will move downwards. Although including local income in the state’ s objective function in some situations would provide a more realistic description, we maintain g 5 0 since the qualitative results only depend on the slope of the response curve.

288

Land Economics

ket exists or not. When we assume a perfect labor market, the opportunity costs of labor can be taken as exogenous and the model becomes recursive: the production decisions can be separated from the consumption decisions and studied as a pro® t-maximizing problem. If some prices are not market-determined, the production and consumption decisions must be solved simultaneously (family labor supply 5 farm labor demand). The local behavior is studied as a utility maximizing problem, that is, a problem of balancing the utility of consumption (income) and the disutility of labor (the Chayanovian model).4 Third, what is the structure of the game? We discuss the Nash equilibria in both the Cournot game (a static game with simultaneous moves, the zero conjecture or independent adjustment equilibrium), and the sequential leader-follower or Stackelberg game with either the state or the local community as the leader. The follower observes the leader’ s choice and chooses the optimal strategy in a similar manner as in the Cournot game. The leader, choosing ® rst, anticipates the response of the follower, and includes the follower’ s response in the optimization problem. Varying the three assumptions above yields a dozen different cases (2 3 2 3 3), but we will be focusing on three typical cases. III. CASE 1: POOR, ISOLATED LOCAL COMMUNITY In the ® rst case we consider the interaction between state and local deforestation in the context of a poor, isolated local community. This case could describe the situation for many tribal communities. Their livelihood, based on forest income from hunting, gathering and extensive forms of agriculture such as long-fallow shifting cultivation, is being undermined as the area of natural forest declines through state appropriation. Examples of this situation are found in the Amazon and Southeast Asia, for example, Colchester and Lohmann (1993). We do not make any a priori assumptions about whether the land scarcity or infrastructure effect will dominate, as both situations

May 2001

could be empirically relevant. We assume imperfect markets, and therefore use the utility-maximizing approach. Finally, we assume a Cournot game, although we realize the dif® culties to determine a priori the game formulation that most realistically describes the situation. The State’ s Response Curve

The objective of the state is to maximize income as given in [3]. The state will then choose the amount of land for plantations, logging, etc., such that the marginal gross income from forest conversion is equal to the marginal costs in terms of reduced forest bene® ts (opportunity costs) and the direct costs of forest clearing;5 s 1 5 v 1 2 g 1 2 h(HF ) 5 0.

[4]

The optimal amount of land clearing by one agent is a function of the amount appropriated by the other. We de® ne the optimal level of HS as a function of the local community’ s choice, that is, the response or bestreply function for the state; HS* 5 HS(HL).

[5]

To explore the characteristics of the response function, we differentiate (4) to obtain; s 12 g 11 1 h 1 dHS* 52 52 , 0. L dH s 11 v 11 1 g 11 1 h 1

[6]

The response curve of the state is downward sloping in an HL 2 HS diagram for two reasons, cf. Figure 1. More local forest clearing implies that the remaining forest becomes more valuable, that is, the net marginal bene® ts of virgin forest (g 11 ) and the opportunity costs of conversion increase. 4 The distinction between pro® t and utility maximizing local behavior relates also to the time horizon for the analysis; the small, open economy assumption is relatively more relevant for long-term analysis when migration is an option and the opportunity costs of labour therefore can be taken as exogenously given. 5 It follows from the assumptions made that s 11 , 0.

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

289

FIGURE 1 The Response Curve and Indifference/ Iso-P rofit Curves for the Local Community (Solid Lines) and the State (Stacked Lines) in the P oor, Isolated Community Case

Further, the marginal costs of forest conversion will be higher as the remaining forest is less suitable or accessible (h1 ). The iso-pro® t curves for the state are de® ned by setting S 5 S. The shape of the curves is found by total differentiation of [3]; s1 v 1 2 g 1 2 h(×) dH 52 52 . HS dHS s2 h 1 dy 2 g1 1

3

U 5 U r(HL) 1 t(HF ),

#

HL

0

L

#

is increasing in gross income and decreasing in labor input;

[7]

0

Whereas the response curve shows the optimal response to changes in the other player’ s choice, the iso-pro® t curves simply show the change necessary to maintain the same income. s 2 is always negative, whereas s1 is positive for small values of HS, zero in optimum, cf. [4], and negative for larger values. Thus the state’ s iso-pro® t curves is inverted C-shaped in an HL-HS diagram, see Figure 1. The Local Response Curve

Following the Chayanovian approach (e.g., Nakajima 1986), the problem of the local community is to maximize utility, which

4

c(HT 2 HS 2 x, HS )dx 5 u(HL, HS ). [8]

The optimal amount of agricultural land is determined by; u 1 5 0 Û r 1 2 t1 2 zc(HT 2 HS 2 HL* ) 5 0; U2 z;2 . [9] U1

Net marginal income from forest conversion (r1 2 t1) should in optimum equal the marginal labor requirement for land expansion multiplied by the shadow wage rate or virtual price of labor (z). The use of virtual prices facilitates the comparative statics. The substitution effect is found by keeping z constant, whereas the income effect is determined by the change in z (Nakajima 1986; Angelsen 1999).

Land Economics

290

Eq. [9] implicitly de® nes the optimal local deforestation (HL*) as a function of HS, or the local response function; HL* 5 HL(HS).

[10]

The local response to increased state deforestation (which is the inverse of the slope of the response curve in Figure 1) is; u 12 t11 2 z(c 2 2 c1 ) 2 c(×)zH S dHL* 52 52 _ 0; S dH u 11 r 11 1 t11 1 zc 1 2 c(×)zH L zH S ;

¶z ¶z , zH L ; . [11] S dH dHL

The denominator in [11] is negative, corresponding to the second order conditions for maximum (u 11 , 0). The response of the local community to higher HS (sign of u 12 ) is ambiguous. As a ® rst step to analyze the effect, z is kept constant, corresponding to the perfect market case. There are three effects to consider. First, more state deforestation implies higher net marginal bene® ts of virgin forest (t11), that is, the opportunity costs of agricultural conversion increases. Second, the marginal costs of land expansion increases as the remaining forest is less suitable or accessible for agriculture (c 1 ). Third, state clearing provides infrastructure which reduces the costs of land expansion (c 2 ). As shown above, the infrastructure effect will be relatively larger to the land scarcity effect the lower the level of HS, whereas the impact on the ® rst effect cannot be determined from the assumptions made. However, we can conclude that if the infrastructure effect is suf® ciently strong the expression (t11 2 z(c 2 2 c 1 )) in [11] will be positive. In the second step, we consider the effect of changes in the shadow wage (z), which gives the income effect. The impact of more state deforestation on z is through its effect on income and labour; both higher income and more labour input increase z. First, higher HS affects the total costs as shown by the ® rst element in the numerator. For example, if the land scarcity effect is strong (c 2 2 c 1 . 0), more state deforestation implies higher labour input and therefore higher z. Second, higher HS will reduce the income by

May 2001

lowering the primary forest area, which reduces z. Adding these two effects of higher HS, the income effect is clearly negative when there is a strong infrastructure effect: lower z pulls towards more local forest conversion as the opportunity cost of using labour in agriculture is lowered. The income effect is ambiguous when the land scarcity effect dominates. Summarizing the sign of the four elements in [11], the local response curve is upward sloping (u 12 . 0) when the infrastructure effect, the income effect, or both are strong.6 Angelsen (1999) argues that the income effect is likely to be strong among poor farm households (our case), which is to say that subsistence needs dominate their responses. To illustrate this point, assume an extreme version of the utility maximizing approach in which the households have lexicographic preferences: they aim for a subsistence income at minimum labour input (``full belly’ ’ preferences). The optimization problem becomes very simple: the community gets a basic income from natural forest, t(HF ), and then clears as much forest for agriculture as required to reach the subsistence income: r(HL) 1 t(HF ) 5 I min. This also implicitly de® nes the response curve of the local community, which is upward sloping with the inverse of the slope being t1/(r 1 2 t1) . 0, cf. [11]. More state deforestation reduces local forest income, and this has to be compensated for by expanding agricultural land area. If the marginal bene® ts from forest products are large relative to the bene® ts from agricultural land, state forest clearing has a signi® cant impact on local agricultural expansion. Returning to our original model, the community’ s indifference curves are de® ned by setting U 5 U, and the curvature is found by differentiation of [8]; u1 dHS 52 5 L dH u2

r 1 2 t1 2 zc(×) t1 1 z

#

0

HL

.

[12]

(c 2 2 c 1 )dx

6 Sandler (1996) and Alston, Liebcap, and Mueller (1999) also discuss positively sloped response curves, but the causes of the upward sloping differ among the papers.

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

u 1 goes from being positive to negative as HL increases, and is zero in optimum. The shape of the indifference curves depends on the sign of u 2, which may be either positive or negative. When the infrastructure effect is small, u 2 , 0. The local indifference curves are then inverted U-shaped, which we shall assume in the following.7 Cournot Equilibrium

The Cournot or Nash-Cournot equilibrium is found where the two response curves intersect (A) in Figure 1. This is the only point with consistency for both players between their own optimal level of forest clearing and the level chosen by the other. The necessary condition for a stable equilibrium is (e.g., Shapiro 1989, 386); u 11 s 11 2 u 12 s 12 . 0.

[13]

As seen from [11] and [6], we have u 11 , s 11, s 12 , 0; u 11 , u 12 ; s 11 , s 12, which ensure that the condition is met. Graphically, [13] says that the local response curve, when moving south, must intersect with the state’ s response curve from above. Consider an exogenous shift in the state’ s response curve, as indicated by the dotted line of Figure 1. For any given value of HL, the state wants to appropriate more land than before. This could be due to, for example, higher prices of plantation products, technological progress, or less value attached to virgin forest. The local response will be more forest clearing, and the new equilibrium is in point B. State deforestation fuels local deforestation in this case; it reduces local forest income which must be compensated for by expanding the local agricultural area. If state deforestation in addition provides infrastructure such that the cost of agricultural expansion is reduced, this gives an additional incentive for local land expansion. An illustration of the empirical relevance of this case is given in a review of local studies on poverty and tropical forest degradation by Kates and Haarmann (1992). They identify two major sources of displacement of in-

291

digenous hunter-gatherers or poor farmers; one is by (state-sponsored) commercial activities, the other by spontaneous in-migrants or government planned resettlement programmes. This leads to degradation of forest resources on which the traditional users depend, and subsistence requirements force them to expand their agricultural activities into new forest areas (point B in Figure 1). IV. CASE 2: INCREASED FORESTLAND COMPETITION; LOCAL-LED LAND RACE When forestland scarcity and competition increases, one possibility is that we move from a Cournot game to a Stackelberg game with the local community as the leader. This game would then describe a race for primary forest where the local community is the ``aggressive’ ’ player, and clear forest in order to squeeze the state. Why is it reasonable to assume the local community to be a Stackelberg leader? Besides the need to test the implications of different game assumptions, one could argue that the local community has greater ¯ exibility than the state in adjusting its forest clearing, for example, because the state’ s decisions must move through a bureaucracy, and often require heavy capital investments. Further, local communities may have some knowledge of the state’ s plans and therefore be in a position to engage in proactive deforestation. We make no a priori assumptions about the local economy, and discuss the imperfect market (utility-maximizing) case which is the most general one. The Stackelberg game with a local leader appears to be most reasonable when state deforestation is costly to the local community, hence we assume the land scarcity effect to dominate the infrastructure 7 Note that the conditions for inverted U-shaped indifference curves (u 2 , 0) are not the same as the condition for a downward sloping response curve (u 12 , 0), although they are related. In particular, the sign of u 12 is in¯ uenced by the relative strength of the income effect, whereas u 2 is not. It is possible with both u 12 , 0 and u 2 . 0, or u 12 . 0 and u 2 , 0. To simplify the presentation, we shall in the following assume a upward sloping response curve due to strong income effects and that the indifference curves are inverted U-shaped.

Land Economics

292

effect. We discuss, however, examples in section 6 where this assumption may not hold. Local Behavior and the Stackelberg Equilibrium

The problem for the local community as a leader is to maximize utility as given in [8], subject to the response function for the state as given in [5]. The optimal level of local forest clearing must satisfy the following condition; u1 1 u2

3

dHS* 5 r 1 2 t1 2 zc(×) dHL

1 2 t1 2 z

#

0

HL

4

[14]

dHS* (c 2 2 c 1 )dx 5 0. dHL

The ® rst part of the expression is similar to the Cournot case, cf. [9]. In addition, the local community takes into account the state’ s response to local forest clearing, dHS*/dHL , 0. We have assumed state deforestation to be costly to the local community (u 2 , 0). Thus local forest clearing has

May 2001

become less costly on the margin because it reduces state deforestation, which both increases forest income and reduces the costs of agricultural expansion. The Stackelberg equilibrium is presented in Figure 2. The local community’ s preference direction is south, and the equilibrium is given in point B where the local indifference curve tangents the state’ s response curve. Compared to a Cournot equilibrium (A), the local community will clear more and the state less forest. The community uses its strategic position as the leader to ``squeeze the state’ ’ and prevent it from converting as much forest as it would have done in a Cournot game. A Stackelberg game with the local community as the leader gives more overall deforestation compared to a Cournot game. Point A is inside the 45° line, on which the sum of state and local deforestation is the same as in point B. The local community gets higher utility, whereas the state’ s pro® t will be lower in B compared to A. Note that the above results do not depend on the slope of the local response curve.

FIGURE 2 Local Community as the Leader in a Stackelberg Game

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

Local Response to Higher State Deforestation

An exogenous shift in state deforestation is illustrated by an upward shift in the state’ s response function, but the effect is not readily seen graphically. Differentiation of [14] gives; dHS* ¶(dHS* /dHL) u u2 1 22 dHL ¶HS dHL* 52 . dHS dHS* ¶(dHS* /dHL) u 11 1 u u 1 21 2 dHL ¶HL [15] u 12 1

We assume the denominator to be negative (second order conditions for maximum). The numerator consists of three terms. The ® rst term [u12], which gives the Cournot response can be either negative or positive, cf. Case 1. It is negative if both the infrastructure and the income effects are small. We are now considering the case when the infrastructure effect is small, hence the sign depends on the strength of the income effect relative to the substitution effect. The second term relates to the change in the local costs of higher state clearing. More state clearing implies that on the margin state clearing is more costly to the local community. However, as higher local forest clearing reduces state clearing, this effect will push in the direction of higher local deforestation. The gain from squeezing the state is higher.8 The third term relates to the changes in the slope of the state’ s response curve, and we cannot say much a priori on the sign of this effect. Overall, one could expect the ® rst and most direct effect to dominate over the second and third. Two major conclusions emerge from this case. First, compared to a Cournot game there will be more local and less state deforestation, and more overall deforestation. The local community gains and the state loses compared to a Cournot game. Second, the local response to more state deforestation is similar to Case 1. If the income effect is strong, farmers compensate for the reduced forest income by increasing agricultural land area. If the income effect is weak, or we are in the perfect market case where only substi-

293

tution effects apply, more state deforestation implies less local deforestation because local land expansion has become more costly. Empirical Relevance

The local community uses its position as the leader to squeeze the state, as expressed by the difference between point B and A in Figure 2. Such a local-led land race that results from a change in the local strategy (from Cournot to Stackelberg leader) has been observed empirically. The development since the mid-1980s in the Seberida district in Sumatra could be interpreted as such a shift (Angelsen 1995). The local villages were not just passively adapting to forest appropriation by the state (logging, transmigration, andÐ more recentlyÐ oil palm plantations), but they played strategically in the way that they cleared and cultivated more forestland that otherwise could have been appropriated by the state. The Stackelberg game with the local community as the leader could also describe con¯ icts arising when the governmentÐ against the local willÐ decides that a forest area should be protected. Before actual implementation and enforcement of, say, a national park or forest reserve, the local community might claim rights by deforesting the area (which in addition makes it less attractive as a protected area). The rationale behind the strategy is easily understood when we also take into account the option value of natural and unclaimed forest (a reservoir for future agricultural land). Such a case is reported in PRODISA (1996), where the government of Bolivia declared a forest reserve in Santa Rosa and San Carlos municipalities. A group of colonists moved in, and wanted to prove their rights to the land by cutting trees and converting forest to agriculture. Another illustrative case is presented in Alston, Liebcap, and Mueller (1999). The two agents are landless peasants (squatters) and large landowners in rural Brazil. The 8 The picture is, in fact, more complicated if we also consider that z will change. With suf® ciently large income effects this result might change.

Land Economics

294

Constitution states that land must ful® ll its ``social function’ ’ to remain secure, and land reforms intend to reallocate land from unproductive latifundia to the landless. Thus peasant invasion on farmsÐ normally by converting forest to farmlandÐ ® nds its justi® cation, although also landowners have their legal backing. The result is both violent con¯ icts and a series of court trials. Furthermore, this institutional setting provides strong incentives for deforestation. The landowners cut the trees to avoid invasion, as forest is considered ``unproductive’ ’ and not ful® lling its ``social function.’ ’ Similarly, the peasants cut the trees to demonstrate their investment in the land and thereby strengthening their claims.9 V. CASE 3: FIERCE LAND COMPETITION; THE STATE AS THE LEADER

Local Behavior

Given local income maximization, the optimal agricultural land area is determined by; l 1 5 r 1 2 t1 2 c(HF, HS ) 5 0.

l 12 t11 1 c 1 2 c 2 dHL* 52 52 , 0. S dH l 11 r 11 1 t11 1 c 1

[16]

The inverse of the slope of the response curve is;

[17]

The local response curve is downward sloping as we have assumed that the land scarcity effect dominates the infrastructure effect (l 12 , 0). State Behavior and the Stackelberg Equilibrium

The objective of the state is to choose its level of HS such that the revenue given in [3] is maximized, taking into account the response of the local community as given in [17]. The revenue is maximized when; s1 1 s 2 1

In the third case the competition for forestland is strong, and the local economy is well integrated into the regional/national economy. Compared to the two previous cases, this might describe the situation at a later stage in the economic development of a region. This game could therefore be used to illustrate the interaction between the state and local communities in more central areas of Southeast Asia, where there is relatively little forest left and farmers are well integrated into national markets. Related to our main assumptions, this situation implies that the land scarcity effect dominates the infrastructure effect in the local cost function, and that we can assume a perfect labor market and study the local adaptation as a pro® t-maximizing problem. The game played is a Stackelberg game with the state as the leader.

May 2001

dHL* 5 v1 2 g1 2 h(HF ) dHS

3

dHL* 2 g1 1 dHS

#

0

HS

4

h 1 dy 5 0.

[18]

This is a modi® ed version of the optimality condition in the Cournot game, cf. [4]. The state now takes into account the negative effect on local forest clearance when deciding its own level of deforestation. Reduced local deforestation always increases the net bene® ts to the state (s 2 , 0). Figure 3 illustrates this case. The Stackelberg equilibrium will be where the isopro® t curve is tangent to the response curve of the local community (B). Compared to the Cournot solution (A), the present game gives more forest clearing by the state and less by the local community. The intuition behind these results is straightforward, and parallel to Case 2. Forest conversion by the state is less costly to the state because it knows that local deforestation will be reduced. The state uses its strategic position to squeeze the local community. The distance between A and B (measured on the x-axis) gives the optimal 9 In the introduction we noted three possible reasons as to why farmers might clear land beyond what a secure private property solution suggests. In empirical work, it might be dif® cult to distinguish between the different causes. All cases discussed here probably also have a strong element of the ® rst reason mentioned, that is, land is cleared to establish or strengthen land rights before others arrive to claim the land.

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

295

FIGURE 3 State as the Leader in a Stackelberg Game

``squeeze.’ ’ Total deforestation will increase as point A is inside the 45° line. Thus compared with the Cournot equilibrium this Stackelberg game contributes to further forest exploitation. State income is higher and local income lower compared to the Cournot equilibrium. As is generally known from the oligopoly and game theoretic literature, there is a ® rstmover advantage in games where quantities are the decision variables: the leader could have chosen the Cournot quantity if it were to yield higher income. The last-mover disadvantage is seen both by studying the isopro® t curves, and from the fact that both HL and HF are smaller in the Stackelberg model. A positive exogenous shift in the revenue function of the state (iso-pro® t curves shift upwards) will obviously make the local community reduce its forest clearing. Total deforestation will, however, increase as we move northwest along the local response curve (the slope of the local response curve is greater than minus one). National governments commonly view forest clearing by local farmers as the real problem of deforestation, often justi® ed by

the low yield of shifting cultivation systems. It is sometimes referred to as ``unplanned’ ’ deforestation, in contrast to the ``planned’ ’ and desirable deforestation by the state. This view is re¯ ected in the assumptions underlying the state’ s objective function. The model therefore provides an explanation of the commonly observed ``aggressive’ ’ behavior of the state in forest conversion. By being the leader, the state not only increases its own forest clearing and income, but it will also reduce what is considered the real environmental problemÐ local deforestation. Dove (1987) and Dauvergne (1994) discuss how the (old) Indonesian government made this distinction, and this might be used to understand the state-support for large-scale logging, transmigration, and plantation projects. VI. DISCUSSION We have focused on three typical cases. By varying the three key assumptions (cf. Section 2), we get 12 different game situations. Table 1 summarizes for each game the answers to the two main questions asked.

Land Economics

296

May 2001

TABLE 1 The Main Results of Different Local-State Games Market (and behavioral) Assumptions and Cost Structur e for the Local Community

Type of Gam e 1. Simultaneous moves (Cournot)

Sequential moves (Stackelberg) 2. Local community as the leader

3. State as the leader

A. Income max. (small, open economy), or utility max. (autarky) when substitution effects dominate income effects

i. Land scarcity effect dominates

I: n.a. II: decrease

I: higher II: decrease (Case 2)

I: higher II: decrease (Case 3)

ii. Infrastructure effect dominates

I: n.a. II: increase

I: lower II: increase

I: lower II: increase

B. Local utility maximization (autarky) when income effects dominate substitution effects

i. Land scarcity effect dominates

I: n.a. II: increase (Case 1)

I: higher (same) II: increase (Case 2)

I: lower II: increase

ii. Infrastructure effect dominates

I: n.a. II: increase (Case 1)

I: lower (samea ) II: increase

I: lower II: increase

Notes: I 5 The total level of deforestation in the Stackelberg games compared to the Cournot game. II 5 The effect of higher forest clearing by the state on local deforestation. a The level of deforestation will be the same in the ``full belly’ ’ case.

Which Games Promote Deforestation?

In the perfect market case with small infrastructure effects (downward sloping local response curve), a Stackelberg game with either player as leader gives more overall deforestation than a Cournot game: the ``squeezing effort’ ’ by the leader is larger than the ``squeeze’ ’ of the follower. How robust is this result, or, in other words, will Stackelberg games always lead to more deforestation? Consider ® rst, games where the state is the leader. When the local response curve is upward sloping, forest appropriation has become more costly to the state when it takes into account the local response. This provides an incentive for the state to reduce own deforestation, and this Stackelberg game gives less overall deforestation. When the local community is the leader, we found in Case 2 that irrespective of whether income or substitution effects dominate, the Stackelberg equilibrium yields more deforestation than the Cournot equilibrium. This assumes that the land scarcity effect dominates the infrastructure effect.

If the infrastructure effect dominates, however, the conclusion is reversed. The local community will reduce its own clearing to promote state clearing, which is bene® cial to them. Although this might appear as an odd situation, there are many examples where communities have no interest to keep the state out because it is useful to them as a provider of infrastructure. A village headman in East Kalimantan, Indonesia, told me, in November 1998, that the village would very much welcome coal mining in the nearby forest as it would provide a new road to the village. In Para, Brazil, many farmers welcome loggers as they provide roads and market access, while the competition for land is limited (Emilio Moran, pers. com. March 2000). In conclusion, even though we found in Cases 2 and 3 that the Stackelberg equilibrium gives more deforestation than the Cournot equilibrium, this result is sensitive to the other assumptions. In particular, when the local response curve is upward sloping, a rational state should as a leader reduce its forest clearing, and the result is less deforestation. What is worse from a forest conservation

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

view, the state or the local community being the leader in a Stackelberg game? This depends, inter alia, on the slope of the response curves. If the local response curve due to strong income effects is upward slopingÐ or downward sloping but relatively steepÐ then a game with a local leader leads to more deforestation compared to a game with the state as the leader. When Does State Deforestation Fuel Local Deforestation?

The second major question is in which situations increased forest appropriation by the state will stimulate local forest clearing. The answer is quite simple: if the local response curve is downward sloping, more state deforestation reduces local deforestation. In other words, when (1) the infrastructure effect is small, and (2) the income effect is small (or non-existing as in the perfect market situation), higher state deforestation will to some extent replace local deforestation. If the infrastructure and/or the income effect are strong, the result is reversed. These results hold in all three game structures. The local level of deforestation is also affected by the game played, as discussed above, and the state might in¯ uence the structure of the game. Moving from Cournot to a game with the local community as the Stackelberg leader will increase local deforestation if the infrastructure effect is weak, while it will decrease if the infrastructure effect is strong. Moving from a Cournot game to a Stackelberg game with the state as the leader will always imply less local deforestation, irrespective of the slope of the local response curve. Given the advantages of being the Stackelberg leader, we might get a situation of Stackelberg warfare. Although this calls for a truly dynamic (and more complex) model, and raises new issues such as ``credible threats,’ ’ we might gain some insight from the present model. In the case with a downward sloping local response curve, both players would choose a level of forest clearing higher than the Cournot equilibrium, trying to squeeze the other and hoping that it will be the follower, cf. Figures 2 and 3. The result

297

is more overall deforestation than in both the Cournot equilibrium and either of the two Stackelberg equilibria. This outcome is, however, not a stable equilibrium: both agents will be better off by moving back on their response curve. Cooperation on Forest Conservation

The models of this paper also illustrate the potential and problems of cooperation in forest management and conservation. The shaded eye-shaped area in Figure 3 gives the set of combinations of HL and HS where both agents have at least as high income as in point B, that is, the area for Pareto improvements. The issues of community participation, co-management or state-local partnership in forest management can be viewed as attempts to establish a cooperative solution within the shaded area. Both agents reduce their conversion of forest to contribute to the preservation of the public goodÐ primary forest. The contract curve will be the line where the iso-pro® t curves of the two agents tangent each other. Chopra, Kadekodi, and Murty (1990) discuss possible state-local contracts in forest management through the application of a bargaining model. The well-known problem is that even if both parties would gain from being inside the shaded area compared to B, both would also have an incentive to defect an agreement. Related to the binary choice game literature, the choice between B and any point within the shaded area can be considered a prisoner’ s dilemma game. The models can also be used to explain why there are limited incentives for each party in unilateral actions for forest conservation. Reduced state appropriation will result in more local conversion in Case 3, or more generally in games with small infrastructure and income effects. Particularly, if the slope of the response curve for the local community is close to minus one, reduced state deforestation will be offset by an almost equal increase in local deforestation. From [11] we see that this will be the case when the infrastructure effect is small, and the gross marginal bene® ts of forest conversion are relatively constant. The latter applies to

Land Economics

298

situations where the products are sold in a large market, and there are few constraints on the labour input, for example, through migration. In this situation, the forest conservation effect of unilateral actions by the state will be negligible. As a corollary to the above result, conservation efforts by the state will be particularly effective in the cases where state appropriation fuels local deforestation. Unilateral conservation efforts by the local community will always be met by more state deforestation. VII. CONCLUSION This paper has presented a simple game theoretic analysis on how games of forest appropriation might escalate the problem of tropical deforestation, and how state actions will impact on local communities’ forest conversion. In real life situations, there are more than just two (coalitions of ) players, and the assumption of uni® ed agents might not hold. The state consists of various ministries with different agendas (e.g., forestry vs. environmental), and at different geographical levels (federal governments might be more sensitive to environmental issues than state/provincial authorities). Different large-scale commercial users might also compete for land (and the backing of the state). Also within the local community, there might be groups with different interests. Non-cooperation within one or both coalitions would be an interesting path for further research.10 In spite of the simplifying assumptions, there are some lessons to be drawn from the model. We have given particular attention to the assumptions made about the local costs of land expansion and the degree of market integration, which are critical for many of the results. We found in Cases 2 and 3 that Stackelberg games with either the local community or the state as the leader yield more deforestation than Cournot games. The leader uses its position to squeeze the other agent, and the net result is more deforestation. Thus the kind of strategic behavior that arise in Stackelberg games is bad from a forest conservation viewpoint.

May 2001

This outcome is, however, sensitive to assumptions made about the market and cost structure. If the infrastructure effects dominate, Stackelberg games with either the state or the community as the leader yields less deforestation. The second question raised was under which circumstances higher forest appropriation by the state also will advance local deforestation. Again, the answer is sensitive to the underlying assumptions. If neither the infrastructure nor the income effect dominates, we get conclusions similar to a conventional duopoly game: higher state appropriation will squeeze the local community. These results are robust with respect to the assumptions made about the structure of the game. If, however, one or both of these effects are strong, then state deforestation fuels local deforestation. The case with strong infrastructure effect has received some attention in the literature on tropical deforestation. It is generally argued that logging, plantations, and other large-scale projects provide infrastructure, particularly roads, which gives farmers easier access to primary forest. This phenomenon is sometimes referred to as the ``logging-shifting cultivation tandem,’ ’ common in many Asian countries. The other possibility for a state-induced local deforestation is when the need to survival determines the local response (strong income effects), which was discussed in relation to Case 1. Overall, the sensitivity of the results to market and cost assumptions gives a warning about making general conclusions on how ``games in the forest’ ’ affect total deforestation, and how state deforestation in¯ uences local behavior. We might, however, tentatively suggest that at early stages in the deforestation process, with isolated communities and forestland abundance, state deforestation works in tandem with local deforestation. At later stages, with more ® erce land competition and marketintegrated communities, state deforestation tends to replace and squeeze local deforestation.

10

This is discussed by Sandler (1996).

77(2)

Angelsen: State-Local Con¯ icts of Land Appropriation

References Alston, Lee J., Gary D. Libecap, and Bernardo Mueller. 1999. ``A Model of Rural Con¯ ict: Violence and Land Reform Policy in Brazil.’ ’ Environment and Development Economics 4 (2): 135± 60. Angelsen, Arild. 1995. ``Shifting Cultivation and ``Deforestation’ ’ . A Study from Indonesia.’ ’ World Development 23 (Oct.): 1713± 29. Ð Ð Ð . 1999. ``Agricultural Expansion and Deforestation: Modelling the Impact of Population, Market Forces and Property Rights.’ ’ Journal of Development Economics 58 (Jan.): 185± 218. Baland, Jean-Marie, and Jean-Philippe Platteau. 1996. Halting Degradation of Natural Resources: Is There a Role for Local Communities. Oxford: Oxford University Press. Bromley, Daniel W., and Devendra P. Chapagain. 1984. ``The Village against the Center: Resource Depletion in South Asia.’ ’ American Journal of Agricultural Economics (Dec.): 868± 73. Chopra, Kanchan, Gopal K. Kadekodi, M. N. Murty. 1990. Participatory Development: People and Common Property Resources. New Dehli, Newbury Park, London: Sage Publications. Colchester, Marcus. 1994. ``Sustaining the Forest: The Community-based Approach in South and South-East Asia.’ ’ In Development and the Environment: Sustaining People and Nature, ed. Dharam Ghai. London: Blackwell and UNRISD. Colchester, M., and L. Lohmann, eds. 1993. The Struggle for Land and the Fate of the Forest. Penang, Malaysia: The Rainforest Movement, with the Ecologist and Zed Press. Cornes, Richard, and Todd Sandler. 1996. The Theory of Externalities, Public Goods, and Club Goods. 2nd ed. New York: Cambridge University Press. Dauvergne, Peter. 1994. ``The Politics of Defor-

299

estation in Indonesia.’ ’ Paci® c Affairs 66 (4): 497± 518. Dove, Michael R. 1987. ``The Perception of Peasant Land Rights in Indonesian Development: Causes and Implications.’ ’ In Land, Trees and Tenure, ed. J. B. Raintree. Nairobi and Madison, Wis.: International Centre for Research on Agroforestry (ICRAF) and Land Tenure Center. Kaimowitz, David, and Arild Angelsen. 1998. Economic Models of Tropical Deforestation: A Review. Bogor, Indonesia: Center for International Forestry Research (CIFOR). Kates, R., and V. Haarmann. 1992. ``Where the Poor Live: Are the Assumptions Correct?’ ’ Environment 34:4± 28. Moran, Emilio. 2000. Personal communication. March. Nakajima, Chihiro. 1986. Subjective Equilibrium Theory of the Farm Household. Amsterdam: Elsevier Publications. Ostrom, Elinor, Roy Gardner, and James Walker. 1994. Rules, Games, & Common-Pool Resources. Ann Arbor: University of Michigan Press. PRODISA (Prefectura del Departamento de Santa Cruz Programa de Desarrollo Micro-reginal de las Provincias Ichilo y Sara). 1996. Plan de Ordenamiento Territorial para la Unidad Agroforestal del PLUS (AF) al norte de los municipios de Santa Rosa y San Carlos. Santa Cruz, Bolivia: PRODISA. Sandler, Todd. 1996. ``A Game-Theoretic Analysis of Carbon Emissions.’ ’ In The Political Economy of Environmental Protection: Analysis and Evidence, ed. Roger Congleton. Ann Arbor: University of Michigan Press. Shapiro, Carl. 1989. ``Theories of Oligopoly Behaviour.’ ’ In Handbook of Industrial Organization, Vol. 1, ed. R. Schmalensee and R. D. Willig. Amsterdam: Elsevier Publications. Singh, I., L. Squire, and J. Strauss, eds. 1986. Agricultural Household Models. Extensions, Applications, and Policy. Baltimore: Johns Hopkins University Press.

Royalty Systems, Government Revenues, and Forest Condition: An Application from Malaysia Gregory S. Amacher, Richard J. Brazee, and Meindert Witvliet ABSTRACT. Royalty structure has been linked to deforestation through non-sustainable harvesting and high grading. Yet royalties are an important rent generation mechanism for governments. In this paper, a dynamic model of government policy choice is used to compare different royalty systems with respect to government revenue generation and high grading. Empirical analyses of the various royalty systems is then undertaken for forest concessions in Malaysia. One unique aspect of our study is an examination of how different royalty systems impact harvesting of highand low-valued timber species, government rent capture, and concessionaire pro® ts. The results should help with future royalty design. (JEL Q23, Q28)

I. INTRODUCTION Royalties are an important focus in the ongoing debate regarding deforestation and tropical forests. It is often argued there is a link between royalty structure and non-sustainable forestry (Vincent 1990; Hyde and Sedjo 1992). This line of reasoning ® nds that most royalty systems contribute to degradation of forest conditions through incentives for logging practices such as high grading (FAO 1993, Grut, Gray, and Egli 1991, Hyde, Amacher, and Magrath 1996, and Barbier and Burgess 1995). High grading may occur when royalties are either too low or are not differentiated enough (Repetto and Gillis 1988; Bushbacher 1990). High grading and poor logging practices have been estimated to have degraded over 3 million hectares of forest land in Malaysia, particularly in the state of Sabah.1 Of course, royalties applied to forest concessions are often viewed differently by home governments relative to the rest of the Land Economics · May 2001 · 77 (2): 300± 313 ISSN 0023-7639 Ó 2001 by the Board of Regents of the University of Wisconsin System

world. Royalties provide for domestic revenues and growth, especially since many tropical countries have vast forests and comparative advantages to exploit through wood export. In Lao PDR, royalties from timber concessions have contributed over 60% of the Agricultural ministries budget in recent years (Hyde and Kuuluvainen 1996). Royalty revenues in Malaysia (all states) totaled RM 2.0 billion in 1995. Not surprisingly, a wide variety of royalty systems now exist in tropical forest-owning countries, from those that target income or volumes to those that target speci® c species. Also not surprisingly, many countries are seeking to make available larger land areas for concessions and associated royalty revenues. In fact, the National Forest Policy of 1978 (revised in 1992) of Malaysia speci® cally lists increasing the forest sector’ s contribution to national income as a goal of the program (FAO 1997). Despite the importance of royalties, there is little rigorous work where different systems are compared with respect to forest condition and government revenues. We do not know which royalty systems allow governments to capture the most rent, or which ones lead to unsustainable harvesting and degraded forest conditions. Although high grading has been discussed frequently, there is little empirical work that examines how selective harvesting of high-valued (high quality) species depends on royalty structure. SeThe authors are, respectively, associate professor, College of Natural Resources, Virginia Tech; associate professor, Department of Natural Resource and Environmental Sciences, University of Illinois; and economist for the Ministry of Economic Affairs, The Netherlands. The authors thank two anonymous referees for useful comments. 1 In fact, Repetto and Gillis (1988) argue that Sabah, Malaysia’ s royalty system provides incentives to highgrade species having high prices.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

lective harvesting is the most common logging practice in Malaysia as well as many other countries. Our purpose is to examine these issues for timber concessions in Sabah, Malaysia. Royalty design is now a major issue in Malaysia, as the country is currently in a transition toward increased forest processing in Peninsular Malaysia, Sabah, and Sarawak (Douglas 1993). Our data is species-speci® c, so we can study for the ® rst time how royalty structure affects changes in concessionaire rents, incentives to high grade the best-valued species, and the ability of the government to capture economic rents associated with concessions. The few existing related empirical studies of tropical countries have relied on data where tree species are grouped together into an aggregate measure. The results will reveal which royalty systems do the most to promote sustainable forestry through reductions in high grading. An important comparison is made between tradeoffs in government revenue and concessionaire rents, which accompany reduced harvesting pressure on high-valued species. The comparison is important for future forest land use in Malaysia or any other country that is seeking to expand forest processing and further develop forest sectors. Because previous work has focused on either the impacts of a single royalty system or has made use of data aggregated across species, our study provides a more complete picture for how royalties should be reformed or structured, and how royalties contribute to changes in economic rent associated with forest exploitation. II. REVENUES, FOREST CONDITION, AND ROYALTIES For any given royalty system and a government with the dual purpose of attracting concessions and raising revenues, the royalty choice problem can be written: T

Maxg(t)

#e 0

2rt

[aR(S(t),g(t),t)

1 (1 2 a)V*(S(t),g(t),t)],

[1]

301

where V*(.) is the indirect (maximized) pro® t function of a representative concessionaire at time t, R(.) is the revenue collection function of the government, r is the government’ s discount rate, S(t)is the stock (volume) of forest at time t, and a is a measure of the weight the government attaches to its revenues relative to concessionaire pro® ts.2 In [1], the government chooses the royalty rate, g(t), to maximize a weighted welfare function of revenue collections and pro® t of concessionaires. We assume for convenience there is one representative concessionaire. At each point in time, the ® rm’ s indirect pro® t function is de® ned, V*(S(t), g(t), t) 5 p l(t)h* l (S(t), g(t), t) 1 p h(t)h* h(S(t), g(t), t) 2c l(S(t),t) 2 c h(S(t), t) 2 G(t),

[2]

where h*l(.) is its optimal low-valued species harvests, h*h(.) is optimal high-valued species harvests, ph(t) and pl(t) are the exogenous world prices for high- and low-valued species, and ch(.) and cl(.) are the respective harvesting costs.3 The term G(t) represents a ``royalty payment function’ ’ speci® c to each royalty system; this is a function of harvesting, the type of royalty, and the royalty rate, G(t) ; G(g(t), h*l(.), h*h(.)) ; G(g(t), S(t),t). Given our representative ® rm assumption, the royalty payment function is equal to the government’ s revenue function in [1], G(t) 5 R(g(t),S(t),t). As we demonstrate later, the exact speci® cations of payment and government revenue functions depend on how the royalty is de® ned. 2 It is easy also to use the policymaker’ s problem to study several other cases, including ones where the policymaker does not care about the impact of high grading on the stock, and where the government only cares about the pro® ts of concessionaires that harvest timber or about government revenues. 3 We lose nothing qualitatively by only examining two types of species harvested or assuming a representative ® rm. In part, the species grouping is consistent with our data discussed later, and the notation of the model is simpli® ed somewhat. Allowing for many ® rms would only complicate notation but would not signi® cantly enhance results.

Land Economics

302 Forest Condition and High Grading

The total forest stock is a function of public forest growth less harvesting of high- and low- valued species, Ç (t) 5 G(S(t), H(t)) 2 hh (S(t), g(t), t) S 2 hl(S(t), g(t), t),

[3]

where G(S(t)) is concave growth function, and H(t) is a measure of forest condition at time t. The last two terms on the RHS de® ne total harvesting. High-valued species coincide with high-quality fast growing stock of good form. As harvesting of highvalued species increases, forest condition degrades and future growth of the stock (i.e., future harvesting) is reduced; the future forest will not contain the same proportion of high-valued species due to the ecology of regeneration and competition for site space. This exempli® es the ``high grading’ ’ problem (e.g., see Poore 1993, 50± 51). It is most prevalent when logging proceeds by single tree or group selection (like it does in Malaysia and many other tropical countries).4 Following this argument, we more completely de® ne forest condition by assuming a direct relationship between the stock and the level of harvesting for high-valued species, that is, H(t) 5 H(S(t), f(t)),

[4]

where f(t) 5 [hh(S(t), g(t))/hl(S(t), g(t), t)], Hs $ 0, Hss , 0, Hf , 0, Hf f , 0, and Hf s 5 0. From our discussion above, the change in forest condition over time is tied to movements in the stock driven by harvesting. To see this, differentiate [4] with respect to time and obtain, ¶H Ç ¶H ¶f(.) Ç ¶H(t) S1 S 5 ¶t ¶S(t) ¶f(.) ¶S(t) 1

¶H ¶f(.) ¶g(.) ¶H ¶f(.) 1 . ¶f(.) ¶g(.) ¶t ¶f(.) ¶t

[5]

May 2001

Equation [7] suggests that high grading can occur either through increases in the proportion of high-valued species harvested, or through increases in the total volume removed from the site. The ® rst two RHS terms depend on evolution of the stock through [3]. The last two terms depend on the time path of harvesting. In particular, the third RHS term reveals that royalties can be structured to control high grading and the resulting degradation of forest condition. Externalities and Concessionaire Harvests

The representative ® rm’ s optimal choices for harvesting, h*l(.) and h*h(.), de® ne its response to a royalty system or change in the system. We can derive these optimal choices for a given royalty payment function, G(t), by maximizing the direct pro® t functional over time, V(g(t), S(t), t), subject to conditions on Ç h(t) 5 evolution of the stock, that is, S h l l Ç Ç l g(t), t); S g(t), t); S 2h (S(t), (t) 5 2h (S(t), Ç h5S Ç (t); where a ``·’ ’ indicates a time 1S derivative. The total forest available to the ® rm is determined by the stock set aside for concessions by the home country government. An externality results because the ® rm takes the total stock of high- and low-valued species, Sh(.) and Sl(.), as given and harvests selectively from it over time. This is not socially ef® cient given that the concessionaire ® rm, and not the government, determines the mix of high- and low-valued species to harvest. The ® rm responds only to its own costs of harvesting. It does not internalize changes in future forest condition (equation [5]) brought about by its harvesting actions. By solving the ® rm’ s problem we can show that harvesting choices depend on world prices, the type of royalty, and the stock. As we will demonstrate, the government’ s policy objective in [1] depends only on elasticities of the ® rm’ s harvesting with respect to the royalties; it is not sensitive to 4 There are two ways to consider high-grading, either as removal of the best-formed trees, or removal of the highest-valued species within a stand. Typically the two are related.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

the explicit solution of harvesting paths (and this is why we do not present them here). Government Revenue Collections

In the equations above, the royalty payment function and government revenues are speci® ed for a given royalty system. For a differentiated advalorem royalty, high- and low-valued species are targeted with different royalty rates, and these rates are applied to volume (or value) harvested. The government’ s revenue function in [1] would be speci® cally de® ned as, Rda(S(t), g hda(t), g lda(t), t) 5 g lda(t)h l(S(t), g lda(t), g hda(t), t) 1 g hda(t)hh(S(t), g hda(t), g lda(t),t),

[6]

where Rda(.) denotes government revenues, and g lda(t) and g hda(t) represent the royalty rates applied to harvesting high and low quality species. For an undifferentiated advalorem royalty, g lda(t) 5 g hda(t)5 ga(t). The royalty revenue function in the undifferentiated case collapses to Rua(.) 5 ga(t)[hl(.) 1 hh(.)]. Now consider a differentiated uniform lump sum royalty, where a constant but species-speci® c royalty payment is made for harvesting rights. The corresponding government revenue function is, Rdl(S(t), g hu(t), g lu(t), t) 5 g lu(t) 1 g hu(t),

[7]

where g lu(t) and g hu(t) are lump sum payments applied to low- and high-valued species made from concessionaires to the government at each point in time. Notice equation [7] assumes each type of royalty payment depends on harvesting of speci® c species types. If the lump sum payment does not depend on species type, then the royalty would be of the undifferentiated lump sum variety. That is, g lu(t) 5 g hu(t) in [7]. The time path of government revenues depends on the ® rm’ s harvesting choices, the type of royalty system, and the corresponding royalty rates. For example, totally differ-

303

entiating the government’ s revenue function at one point in time under an undifferentiated advalorem royalty, ga, we have

5

¶Rua(S(t), ga(t), t) 5 dga(t) hl(S(t), ga(t), t) ¶ga(t) 1 hh(S(t), g ha(t), t) 1 ga(t) 1 ga(t)

¶hh(.) ¶ hl(.) 1 ga(t) ¶ga(t) ¶ga(t)

6

¶h l(.) ¶S(t) ¶hh(.) ¶S(t) 1 ga(t) . ¶S(t) ¶ga(t) ¶S(t) ¶ga(t)

[8]

Equation [8] describes how government revenue collections change as the royalty rate and harvesting paths change. In so far as harvesting and high grading degrade the forest, government revenue collections will also be affected through corresponding changes in the stock (last two terms in equation [8]). Thus, from the de® nition of H(t) in [5], the key element to understand about the royalty path is how harvesting of low- and high-valued species change as a result of the royalty’ s structure. To see this we convert [8] to elasticities, ¶Rua(S(t), ga(t), t) 5 dga(t){hl(.) 1 h h(.) 1 elgh l(.) ¶ga(t) 1 ehgh h(.) 1 eshhl(.) 1 eslh l(.)}, [9]

where esh and esl are elasticities of high- and low-valued species harvests with respect to the stock, and ehg and elg are the elasticities of high-valued and low-valued species harvesting with respect to the royalty level. It is clear from [9] that a government choosing policies to promote a sustainable forest condition will experience changes in revenue collections from changes in the distribution of species harvested, as well as harvesting changes induced by changes in royalty rates. A similar exercise could be undertaken for other types of royalties, and equations corresponding to [8] and [9] could be derived. The differentiated advalorem royalty leads to equations identical to [8] ± [9] except there are separate terms multiplying royalty pay-

Land Economics

304

ments for high- and low-valued species harvesting. Generally, we cannot say how the sign and magnitude of these elasticities vary across royalty types, but clearly they will differ according to how each royalty affects the pro® ts of concessionaires.

May 2001

¶aR(.) ¶(1 2 a)V*(.) 1 ¶S(t) ¶S(t) a 1

¶R(.) ¶V*(.) 1 (1 2 a) ¶ga ¶ga

32 ¶h¶G(.)(.) ¶h¶g (t)(.) 1 ¶h¶g (t)(.) 1 ¶h¶g (t)(.)4 h

The Government’ s Optimal Royalty Choice

¶V*(.) ¶R(.) a 1 (1 2 a) ¶ga ¶ga 1 l(t)

3¶h¶G(.)(.) ¶h¶g (t)(.) 2 ¶h¶g (t)(.) 2 ¶h¶g (t)(.)4 5 0, h

h

h

l

a

a

a

and

[10]

¶aR(.) ¶(1 2 a)V*(.) 1 ¶S(t) ¶S(t) 1 l(t)

3

4

¶G(.) ¶hh(.) ¶h l(.) 2 2 2 rl(t) 5 lÇ (t). ¶S(t) ¶S(t) ¶S(t)

h

l

a

a

a

¶h (.) ¶h (.) Ç 2 2 2 r4 5 l(t). 3¶G(.) ¶S(t) ¶S(t) ¶S(t) h

Given [3], [5], [8], and [9], we now examine the government’ s ef® cient royalty choice that balances high grading and revenue collection. For illustration we will use a uniform advalorem royalty system. In this case, equation [1] de® nes the objective functional, [3] is the equation of motion, g(t) is the control, and S(t) is the state. Constructing the current value Hamiltonian, obtaining the necessary conditions assuming an interior solution, and using the envelope theorem we obtain,5

h

l

[11]

Equation [11] implies that the rate of change in the optimal royalty rate should depend on changes in concessionaire pro® ts and government revenues (® rst two terms on LHS) adjusted for changes in the stock due to harvesting of high- and low-valued species. We now investigate their magnitudes in the next section. III. THE MALAYSIAN EXAMPLE The government’ s problem has no closed form solution, and so in this section we use empirical estimates to examine potential reform in royalty systems and tradeoffs to government revenue collections coinciding with improvements in forest condition (or, equivalently, decreases in high grading). This will be the empirical equivalent of equations [5]± [9]. The speci® cations for harvesting follow from the distinction between high- and lowvalued species. In reduced form we need to estimate, h h(t, S(t), g(t)) 5 f h(p(t), S(t), g(t); b h, mh)

Where l(t) is the costate variable associated with the equation of motion, and ¶G(.)/¶j indicates a total derivative of G(.), which is not expanded for simplicity. The ® rst two terms on the RHS in the ® rst necessary condition represent the marginal change in pro® ts from a change in the royalty. The third term measures the marginal change in government revenues, and the last term is an adjustment for changes in the stock due to forest condition changes induced by harvesting of highvalued species. Reducing the two equations to one by eliminating l(t) gives an interpretation of the path of the costate variable over time,

h l(t, S(t), g(t)) 5 f l(p(t), S(t), g(t); b l, ml),

[12]

where bh and bl are regression coef® cients, mh and ml are the errors, and p(t) represents a vector of species-speci® c world export prices 5 Note we assume for simplicity that the government does not explicitly control forest condition, so that H(t) is not present in [1]. This has no bearing on our empirical results or our interest in illustrating theoretical linkages between royalties, revenues, and forest condition. Forest condition is of course implicit in [1] through its importance to revenues and pro® ts, R(.) and V(.), which depend on harvesting paths over time. If the government could control these harvest paths then highgrading would not be an issue.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

(see eqn [2]). Concessionaires bid on timber harvests and determine, given prevailing prices, costs, and terms of the contracts, species and volumes to harvest from a site (Witvliet 1996). These harvesting functions therefore represent supply equations. As such, all arguments of the f k(.) functions are exogenous given the small country assumption and the fact that royalties are taken as given by the concessionaire ® rm. From the perspective of each ® rm, (export) prices are exogenous, as is the stock that the government sets aside for harvesting each period. Several econometric issues must be addressed when estimating [12]. The ® rst is the choice of estimator. Although we have data on species harvested over time, the model might not be entirely consistent with a time series cross section speci® cation, because data is not available by ® rm or by timber sale. This makes it dif® cult to de® ne a panel-speci® c error, which would be required for an error component speci® cation. Conversely, one could argue that world markets may result in shocks speci® c to each species group. If this is the case, then a panel-speci® c error term could indeed be a component of the overall error for harvest supply. Given that we have no way of telling a priori, we provide estimates based on both OLS and panel estimators. It is also possible that heteroskedasticity exists in each harvest supply function, and this will be corrected ex ante to estimation using White’ s method. The equations will also be checked and corrected for any possible autoregression that might be present.6 The data come from the Sabah Forestry Department and are reported by species harvested for 1990± 1995. In Sabah, there are an estimated 7.4 million hectares of land, of which over 60% are forest. Most are government-owned and classi® ed as production (3.07 million ha), protection (0.53 million ha), and of® cial conversion forests (3 million ha). A growing concern is unof® cial conversion of production forests to other non-sustainable uses such as shifting cultivation. Concessions occur exclusively on government-owned land, with the majority involving native forests, or ``Permanent Forest Estates.’ ’ These harvests are based on selection cutting which involves removal of speci® c

305

species in 25± 50 year cycles (FAO 1999).7 There are some diameter limit restrictions (50± 60cm DBH) on removals of a few dipterocarp species in production forests within the Permanent Forest Estate areas, but it has been claimed that loggers do not always follow prescribed harvest guidelines. As a result, the damage to forest condition is a growing problem (FAO 1999). The data cover the years 1991± 1995 on harvesting rates for ® fty types of species in Sabah’ s classi® ed production forests, where poor logging practices are common. Also included are royalty rates and payments, prices, and concessionaire revenues (Witvliet 1996). Royalty systems in Malaysia depend on the state. Sabah has a ``differentiated’ ’ lump sum payment system with ® ve categories of species groups based on prices. The speci® c royalty payment is calculated based on the total value of the sale less an average cost of harvesting. The difference between rates applied to the highest and lowest-valued species varies less than 1%, that is, it is effectively undifferentiated (Witvliet 1996).8 The export of unprocessed logs is currently under a partial ban in Sabah, primarily to support increased processing in Peninsular Malaysia and Sabah; as a result, the prices for raw logs do not re¯ ect market-clearing values. The export price for sawnwood of each species will be used instead. Sawnwood exports from Sabah and Malaysia are an extremely important indicator of world demand for wood, increasing over 300% since 1985; in 1995, sawnwood exports numbered RM3.6 million, nearly one-third of the country’ s total wood export base (FAO 1997). Conversely, 6 Our prices are species-speci® c. Thus, the use of substitute prices does not make sense given that data is not available by sale. This limits our choice of functional form. 7 Peninsular Malaysia is ruled under the National Forest Policy of 1978, but Sabah follows the older forest Policy of 1954. As such, the cutting cycle of selective harvests in Peninsular Malaysia is longer at 30± 55 years. Finally, harvests from private and public plantations are few, with only an estimated 112,700 hectares currently under management. 8 The neighboring state of Sarawak has a lump sum differentiated system with a difference in rates between high- and low-valued categories of 20± 60% (Witlvliet 1996).

Land Economics

306

Sabah’ s raw log exports have been very small and declining, numbering only 0.1 cubic meters in 1994 (or about10% of the total for Malaysia). Harvests mainly support over 30% of Malaysian sawnwood exports each year (FAO 1997), given that one-third of all Malaysia’ s sawmills are located in Sabah. Sabah’ s forest stock measures are not available for individual species. However, a measure of the land area was obtained from the Sabah Forestry Department and will be used as a proxy. The fact that tree species differ across a unit of land is accommodated by our data on harvesting of speci® c species. Tree species were grouped according to the export prices. Low-valued species include gagil, bawhutan, bayur, berangan, bintagor, binuang, assam, keranji, kandis, karai, kedondong, impas, perapat, kungur, laran, magas, obah, pisang, rengas, semangk, sepet, takalis, teluto, and terap. High-valued species include batumer, belain, durain, jelutong, jongkong, kapur, kayumala, kayupen, kembang, melapi, mengar, mengilan, merbau, nyatoh, obasul, and keruing. Obviously there are some borderline tree species which could ® t into either category. Thus, a RESET test will be used to compare models for these border categories; the results of these tests will indicate whether our division of species into two categories represents statistically signi® cant groupings (e.g., see Greene 1997).9

May 2001

Table 1 details important descriptive statistics. Considering the standard deviations, royalty payments have changed frequently over time and there is enough variation for their inclusion in the estimation. Harvesting of high-valued species is considerable, even though the royalty payments per cubic meter are higher. This may give some support to the argument that royalties may be too small or not differentiated enoughÐ we return to this below. It could also mean that volume recovery is higher for high-valued species. Prices for the this group are about 40% larger than prices for low-valued species. From a forest condition perspective, the larger harvest volumes for high-valued species may indeed mean either that high grading is occurring or that the forests covered by the concessions are very high quality in general. From [9], predictions about the impacts of different royalties can be determined by considering both the royalty rate and price elasticities from estimating [12]. Estimated price elasticities of supply can be used to determine how harvesting of high- and lowvalued species depend on the Advalorem royalties targeting the price of harvested spe9 In fact, we tried splitting the species into more than two categories according to quality, but the results were very similar. Thus, we err conservatively by using the two-group estimates in Table 2.

TABLE 1 Descriptive Statistics for Sabah Data (averages are reported for the period 1990± 1995, according to species groups) Variable Priceb Harvest Volume Royalty Paymentc Stock (Million Ha)d Number of Observations

High-Valued Speciesa

Low-Valued Species

926 (409.9) 281,665 (5.4 3 106) 152.65 (96.63) NA

565.25 (193.5) 39,462 (4.2 3 104) 80.1 (66.35) NA

745 (367.4) 159,756 (3.99 3 105) 116.72 (90.28) 4.44 (0.03)

149

151

300

All Species

Notes: Standard deviations are in parentheses. a See text for identi® cation of species groupings. b Harvest volumes and prices refer to cubic meters and Malaysian dollars per cubic meter, respectively. c Royalty is measured on per cubic meter basis, in Malaysian dollars d Stock varies on a per year basis only.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

cies groups. Harvest responses to lump sum royalties can be recovered from the elasticity of supply with respect to the existing Sabahimposed royalty payments. Recall this payment is effectively a reduction in concessionaire pro® ts, thus, it effectively measures the impact of a lump sum royalty payment on harvesting. Uniform advalorem and lump sum royalty elasticities will be recovered from estimating the supply function for all species combined (i.e., the royalty rate and price elasticities will then apply to all species in an undifferentiated manner). Differentiated advalorem and lump sum royalty elasticities will be determined by grouping species

307

according to value and estimating separate supply estimations. In this case, a separate response to policies for each harvest type will be recovered and used to test the impact of differentiated policies that target each species group differently. Estimation Results

Table 2 presents the estimation results. Log-log and linear speci® cations worked best for the OLS and ® xed effect (FE) estimators respectively (in fact, the panel models only converged under a linear speci® cation). The F tests and adjusted R-squares are satis-

TABLE 2 Estimated Elasticities for Harvesting of High- and Low-Valued Species, under Various Types of Royalties for OLS and FE Harvest Supply Regressions Elasticity of Harvest with Respect to

High-Valued Speciesa

Low-Valued species

All Species Combined

Model

OLS

FE

OLS

FE

OLS

Uniform Royalty Payment Advalorem Differentiated Royalty Price Elasticity

NA

NA

NA

NA

22.55* (236.48) NA

20.81** (21.60) NA

1.31* (12.65) 22.55* (236.48)

0.71 (1.40) 20.81** (1.60)

Advalorem Undifferentiated Royalty Stockb F-Statistic for regression Adjusted R2 Log Likelihood RESET Test Statisticd (Null Hypothesis: high- and low-valued species groupings are identical) Hausman Test Signi® cance Level Number of Observations

22.69* (223.58) 1.36* (8.77) NA 4.94 (0.99) 148.8* 0.50 2266.44 NA

NA 149

21.80* (3.49) 1.63* (3.11) NA DNCc

22.54* (229.70) 1.26* (10.43) NA 21.98 (20.48)

20.28 (20.91) 0.22 (0.74) NA DNC

FE

2.12 (0.98)

0.14 (0.97)

5.85*

222.84*

80.73*

405.85*

11.28*

0.46

0.60

0.57

0.58

0.46

22048.1 NA

0.10 149

2253.90 NA

NA 151

21734.5 NA

0.36 151

2529.1 25.08*

NA 300

24188.1 627.2*

0.00 300

Notes: t-statistics of estimates are in parentheses. Royalties and prices are measured on a per cubic meter basis. a All regressions corrected for AR(1) when present. b Land Area in millions of hectares. c DNC 5 Did not converge with variable included in model. d The RESET test addresses the null hypothesis that the regression coef® cients of harvest supply functions for high-and lowvalued species are identical versus the alternative hypothesis that the estimated coef® cients are distinct across groups. * Indicates variable was signi® cant at 1% level or better; ** indicates signi® cance at .15 or better.

308

Land Economics

fying. Although the stock variable is not statistically signi® cant, recall that stocking data does not exist per tree species and is based only on total area. The RESET test statistic is signi® cant at the 10% level, indicating support for rejection of the null hypothesis of equivalence between high- and low-valued tree species supply estimates. Elasticities are presented for harvest supply with respect to different types of royalties and for species groups. An interesting observation is that the harvest of high-valued species is more price and royalty elastic than the low-valued species for all estimators. In addition, harvesting seems to be more sensitive to pro® t reductions than to royalties speci® cally targeting price. Royalties that are too low encourage proportionately greater harvesting of high-valued tree species relative to low-valued ones (the difference could be as large as 5± 10% for every $100 decrease in the royalty payment). Following equations [3] ± [4], the end result will be reduced forest condition and lower future stocks, because the ratio of high- to low-valued tree species harvested (f(t) in [4] ± [5]) will increase. It is also clear from Table 2 that differentiating royalties can improve forest condition. That is, given that harvesting of high-valued trees is more price and royalty elastic than low-valued harvesting, a large enough royalty speci® cally targeting high-valued trees will provide greater forest condition improvements than an undifferentiated royalty. This will be achieved by either a shift toward harvesting of lower-valued species with lower royalty burdens, or from general reductions in harvesting for all species.10 However, differentiating royalties where harvests are more price elastic may be most effective in reducing high grading through reductions in hh(.), but they may decrease government revenues proportionately more (see the discussion surrounding equations [6]± [9]). If government revenues are adversely effected, then the royalty system may not be well accepted by the home country government. We explore magnitudes of these changes below. Improving Forest Condition

Tables 3 and 4 present changes in government revenues and concessionaire pro® ts at

May 2001

the time of initial imposition for each royalty. The high grading ratio is the change in harvesting of high-valued species relative to low-valued species. The larger is this ratio in absolute value, the more effective a royalty is at reducing high grading by shifting the distribution of harvesting. The degree of differentiation is the difference between the rate targeting harvesting of high-valued species and the rate targeting low-valued species. The changes in Table 4 re¯ ect a comparison of differentiated royalties with a uniform 10% royalty that is not differentiated, that is, where high- and low-valued harvests are subject to a 10% uniform payment or advalorem royalty.11 All units are in thousand Malaysian dollars or thousand cubic meters unless otherwise noted. Tables 3 and 4 collectively con® rm that policies to promote forest condition will cost substantially in terms of concessionaire pro® ts. The results also suggest that royalties should be higher given prevailing prices and harvesting volumes, at least for Sabah, if prevention of high grading is an objective. Government revenues increase and harvesting of high-valued tree species decreases substantially when royalty rates are increased for all types of policies. Generally, undifferentiated royalties are similar in terms of changes in concessionaire pro® t collections and changes in the high grading ratio. It is also clear it would be dif® cult to convince a government to use lump sum royalties to reduce high grading and improve forest condition, given that revenue collections could substantially decline. If the government is concerned about revenue collections, then clearly the advalorem undifferentiated royalty is preferable. The results in Table 3 address the assertion that differentiating policy rates across different species (or values) can be an important way to improve forest condition. Recall that the table describes the change in reve10 Of course, administration of differentiated royalties may be more costly to the government, or it might induce higher levels of timber theft and miss-reporting of harvest volumes where sales are not closely monitored. 11 The results are not sensitive to this baseline assumption. The critical issue is the degree to which royalties are differentiated.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

309

TABLE 3 One-Time Change in Government Revenues, High-Grading, and P rofits of Concessionaires for Different Degrees of Differentiation in Each Differentiated Royalty using OLS and Fixed Effect (FE) Estimates (thousand Malaysian dollars) Differentiated Advalorem Royalty

Change in Government Revenuesb

Change in HighGrading Ratioc

OLS

FE

OLS

FE

121,054.8 127,888.7 136,234.9

119,686.9 125,287.9 130,113.3

20.36 20.48 20.72

20.43 20.57 20.86

Change in Concessionaire Pro® t (marginal)d

Degree of Differentiationa 5% 10% 15% Differentiated Lump Sum Royalty

Change in Government Revenues

Change in HighGrading Ratio

OLS

FE

OLS

FE

12,690.6 12,816.3 1133.5

13,623.4 14,111.4 14,543.6

20.71 20.95 21.44

20.48 20.63 20.89

241,199.76 254,240.85 280,323.03 Change in Concessionaire Pro® t

Degree of Differentiation 5% 10% 15%

26,481.03 28,915.32 213.214.94

a Measures the difference between royalty rates targeting harvest of high-valued and harvest of low-valued tree species. This assumes the royalty rate on low-valued species is ® xed at 10%, and the elements of the ® rst column measure the increase in the high-valued royalty rate relative to the low-valued royalty rate. b From equations [8] ± [10] and results in Table 2. Measured in thousands Malaysian dollars c High-grading ratio 5 (change in volume harvested of high-valued species)/(total volume harvested), using state average for denominator. d Pro® ts are determined using royalty payments to recover constant average costs, and then computing revenues minus costs at prevailing prices and under different royalty rate assumptions. Measured in thousands Malaysian dollars.

TABLE 4 One-Time Change in Government Revenues, High-Grading, and P rofits of Concessionaires for Change in Undifferentiated Royalties, for OLS and Fixed Effects (FE) Estimates

Royalty Rate

Change in Government Revenuesb

Change in HighGrading Ratio

Change in Concessionaire Pro® t (marginal)

OLS

FE

OLS

FE

15,665.5 19,834.9

13,332.2 14,522.1

20.23 20.34

20.32 20.48

210,711.6 221,423.2

23,333.7 27,665.1

1204.8 1176.9

20.22 20.45

20.01 20.14

226,450.0 253,395.0

Undifferentiated Advalorem Royalty 10% 20% Undifferentiated Lump Sum Royalty 10% 20% a

Measured in thousands of Malaysian dollars.

310

Land Economics

nues, pro® ts, and forest condition moving from a 10% undifferentiated policy rate to one that varies across high- and low-valued species harvests. Indeed, government revenue increased with differentiation in all cases. Clearly the best royalty to differentiate, in terms of reducing high-valued tree harvesting and improving forest condition, is the lump sum payment variety. Increased differentiation of this royalty also improves government revenues. Differentiation in the advalorem rates leads to much greater harvesting responses, mainly because the price differential of high- and low-valued species is already large (i.e., the elasticities are very different). This difference is much larger than that of existing lump sum royalty payments across species, and as a result increased differentiation in the advalorem rates (which are applied to the value of harvested material) leads to substantial reductions in harvesting. Consistent with this decrease is a very high reduction in concessionaire pro® ts. Implications for Forest and Land Use

The results are important to future land use, particularly given Malaysia has a current objective of improving forest value-added. If the royalty reforms we examine in Tables 3± 4 were implemented to improve forest condition, they certainly would mean decreased rents to concessionaires and production forestry. The highest and best use of the land may as a result shift to some non-production use, such as producing nontimber forest goods. Or, conversely, as rents decline, forests could be viewed as less important relative to other sectors. It is well known that differences in policies across different sectors in an economy contribute considerably to changes in land use. Of course, the royalties that decrease rents the most also improve forest condition in the long run and increase government revenues in the short run. Improved long-term forest condition might indeed make forest production even more attractive, and government revenues may indeed be higher in the long run as future revenue earning potential of the forest increases. The short-run estimates in rent and reve-

May 2001

nue changes might also be indicative of what revenues in other uses need to be before land use change results. For Malaysia, our results argue that these uses would have to return approximately $4± 8 per ha. to compensate for the royalty-induced change in pro® ts that might accompany differentiation. Nontimber market development has been a recent interest of NGOs and governments alike, but whether or not their values are this high remains an important future research area. If nontimber uses prove not to have this value, then royalty reform could encourage short run non-sustainable uses or shifting forest uses before forest condition improved enough to make forestry at the frontier pro® table enough to support competitive timber concessions. IV. CONCLUSION Royalties from timber concessionaires have been an important focus throughout the last decade, especially with regard to the role royalties play (or can play) in high grading and changes in forest condition. Many authors have also questioned whether royalty rates are too small, or whether royalties should have greater differentiation to impose a more substantial penalty on harvesting of the highest-valued species. We use species speci® c data to test these assertions for tropical forest concessions of Sabah, Malaysia. We examine both theoretical and empirical linkages between government revenues, forest condition, and harvesting behavior to determine the impact of different royalty systems on the distribution of harvesting between high- and low-quality species. The results establish that royalty rates could indeed be higher and more differentiated in Sabah, and that changes in high grading are similar for uniform and advalorem systems. However, government acceptance of higher royalties may be greatest with advalorem royalty increases, at least based on revenue collections. More important than the type of royalty is how differentiated it is. Differentiation is the best way to promote favorable changes in forest condition through reductions in high-valued (i.e., high-quality) species removal. It is more ef® cient for the

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

government to increase differentiation in uniform systems (rather than advalorem systems) for policies targeting high- and lowvalued species. For Sabah data, a 10% increase in differentiation yields an increase of revenue of $3,640 and $8,585 for every 1,000 cubic meter decrease in high-valued species harvests for the advalorem and uniform payment royalties respectively. The unit ``costs’ ’ to concessionaires in terms of reduced pro® ts are $7,079 and $59, respectively. The government should prefer to differentiate lump sum royalties from a revenue collection and high-grading perspective, but the ® rms would clearly prefer that the government increase forest condition by differentiating the advalorem royalty. For differentiated royalties the government clearly resides on the upward sloping part of the Laffer curve, that is, government revenue collections are increasing in degrees of differentiation. In only one case, for an increase in the undifferentiated uniform pro® t royalty, does government revenue decrease with a royalty rate increase. The dollars of government revenue raised per unit of concessionaire pro® t reduced are largest for advalorem differentiated and advalorem undifferentiated royalties. It is therefore an open question whether the political climate of the country would make it easy for a government to pick the best policy that jointly reduces high grading and increases government revenues by the largest proportions. Perhaps some revenue sharing between governments and ® rms could be used to ensure a movement toward the most ef® cient royalty system. Regardless of the politics of royalty choice, it is clear that failure to differentiate any royalty system will not provide enough of a decrease in harvesting for the highestvalued species to promote forest condition. Forest condition increases are largest, for each dollar of additional government revenue raised, when royalty reform includes increasing differentiation between rates targeting high- and low-valued species. This occurs mainly because harvesting of the best species is more price elastic. Royalty reform in Malaysia or any similar tropical country, could be used to promote sustainable forestry. Differentiating royalties

311

can lead to substantial revenues, which could be channeled into reforestation programs or intensive management of second growth forests. Revenues could be used to restore the over 3 million hectares subjected to slashand-burn agriculture (a major goal of Malaysia’ s new Land Use Master Plan). Or revenues might be transferred to develop and enhance markets for nonwood forest products such as rattan, palm, resins, etc. Ensuring that second growth forests remain productive will be the key to protecting native forests and providing Malaysia with ¯ exibility to meet future policy commitments and agreements. It is reasonable to question whether our Malaysian results apply more broadly. Recall the results hinge on the size of the rents to concessionaires for harvesting. One reason for the result that increased differentiation of advalorem royalties and increased uniform royalties provide the largest decrease in highvalued species harvesting, per dollar of government revenue raised, is the magnitude of the export price. A small percentage change in price results in relatively elastic changes in harvesting for high-valued species. Clearly, the data supports that rents to harvesting are very high in Malaysia. Perhaps this is because the market in Sabah is close to sea ports, with easy access to the Paci® c Rim countries. In other cases, it has been argued that the rents for harvesting frontier forests can sometimes be small (Hyde, Amacher, and Magrath 1996).12 If this is the case, then we would likely discover lower changes in government revenues for any policy-induced decrease in high grading caused by differentiating a royalty or increasing its rate. It would also not make sense for the government to spend time or effort enforcing royalty collections or monitoring timber sales. But these 12 Many note that any rents present must therefore be the result of either policy failures (e.g., Hyde, Amacher, and Magrath 1996) or of uncertainty in the political climate of the country. More work is needed on royalty design in these cases. This represents the classic second best situation where royalties are designed with preexisting distortions, and the results may differ from our analysis. The question would then become whether a royalty system in general is ef® cient.

312

Land Economics

are the areas where the economic margin is so distant that the government would have dif® culty making the monitoring investment anyway. Most important, if rents are indeed low, than we should ask whether raising royalties, or differentiating them, is practical in the ® rst place. Perhaps the best reform in royalty systems for these cases would be to increase differentiation among royalty rates subject to a requirement that an equivalent amount of government revenue is collected. If this can be done than substantial gains in forest condition might occur if harvesting remains relatively elastic with respect to highvalued species. There are some important assumptions that could be relaxed in future work. We have assumed that administrative costs are absent or, more appropriately, are constant across royalty types. We have also not addressed timber theft. Including these complications, even if there was data available, would only change the magnitude of results. If administrative costs increase for differentiated subsidies, government revenues would be reduced accordingly but would not affect results for changes in high grading and concessionaire pro® ts. Timber theft could be introduced as an additional cost incurred by the government (which again reduces government revenues) for each royalty system, and high grading would be correspondingly affected. As royalty rates increase, or as royalties increase on harvests of high-valued species, we would expect illegal logging and inaccurate reporting of harvesting to increase. This is also similar to an increase in administrative costs.13 Finally, our results all assume that the average costs of harvesting high- and lowvalued species do not change as royalties become more differentiated. It is an open question whether costs of harvesting low-valued species are higher than costs of harvesting high-valued species. Any difference in costs wouldreducepro® ts of concessionaires.If harvesting costs decreased for high-valued species (perhaps because they are easier to log), then changes in concessionaire pro® ts would be smaller in magnitude under royalty differentiation. However, if costs increasedforhighvaluedspecies,thenpro® ts forconcessionaires

May 2001

would be further reduced by decreases in high grading, and governmentrevenues might possibly suffer in the long run if total harvesting declined. Future work could also consider the costs of harvesting in sustainable manners across all tree species. If concessionaires are required to meet some minimal damage constraintfromaharvest,this will entail additional costs and affect their harvesting response to a change in royalties. References Barbier, E., and J. Burgess. 1995. ``The Linkages Between the Timber Trade and Tropical Deforestation: Indonesia.’ ’ The World Economy 18 (3): 411± 42. Bushbacher, R. 1990. ``National Forest Management in the Humid Tropics: Ecological, Social, and Economic Considerations.’ ’ Ambio 19 (5): 253± 58. Douglas, J. 1993. ``Forestry Development Approaches.’ ’ In The Earthscan Reader in Tropical Forestry, ed. Rietbergen. London: Earthscan Publications Ltd. FAO. 1990. Forest Resources Assessment: Tropical Countries. Rome. Ð Ð Ð . 1993. Forestry Policies of Selected Countries in Asia and the Paci® c. Rome. Ð Ð Ð . 1997. Asia-Paci® c Forestry Sector Outlook Study. Working Paper No. APFSOS/WP/ 07. Rome. Ð Ð Ð . 1999. Malaysia. APFSOS Working Paper No. 17. Rome. Greene, W. 1997. Econometric Analysis. Upper Saddle River, N.J.: Prentice Hall Press. Grut, M., J. Gray, and N. Egli. 1991. ``Forest Pricing andConcessionPolicies: Managing the High Forests of West and Central Africa.’ ’ World Bank Discussion Paper, Washington D.C. Hyde, W. F., and R. A. Sedjo. 1992. ``Managing Tropical Forests: Re¯ ections on the Rent Distribution Discussion.’ ’ Land Economics 68 (Aug.): 343± 50. Hyde, W., and J. Kulluvainen. 1996. ``Timber Pricing Policy in Lao PDR.’ ’ World Bank Discussion Paper. Washington, D.C. 13 Moreover, results presentedconcerning the advantage of differentiating royalties versus having uniform rates would probably not be affected by timber theft, since similar levels of theft can be envisioned for each policy. Timber theft is probably more a function of the change in concessionaire pro® ts that would accompany any royalty reform.

77(2)

Amacher, Brazee, and Witvliet: Royalty Systems and Government Revenues

Hyde, W., G. Amacher, and W. Magrath. 1996. ``Deforestation, Scarce Forest Resources, and Forest Land Use: Theory, Empirical Evidence, and Policy Implications.’ ’ World Bank Research Observer 11 (2): 223± 48. Poore, E. 1993. ``The Sustainable Management of Tropical Forests: The Issues.’ ’ In The Earthscan Reader in Tropical Forestry, ed. Rietbergen, London: Earthscan Publications Ltd.

313

Repetto, R., and M. Gillis, eds. 1988. Public Policy and the Misuse of Forest Resources. Cambridge: Cambridge University Press. Vincent, J. R. 1990. ``Rent Capture and the Feasibility of Tropical Forest Management.’ ’ Land Economics 66 (May): 212± 23. Witvliet, M. 1996. ``Royalty Systems and Tropical Countries with Forest Resources.’ ’ Master’ s thesis no. D050-708, Wageningen Agricultural University, The Netherlands.

Edited by Daniel W. Bromley print ISSN: 0023-7639, e-ISSN: 1543-8325 Published four times per year Land Economics is dedicated to the study of land use, natural resources, public utilities, housing, and urban land issues. Established in 1925 by the renowned economist and founder of the American Economic Association, Richard T. Ely at the University of Wisconsin, Land Economics has consistently published innovative, conceptual, and empirical research of direct relevance to economists. Each issue brings the latest results in international applied research on such topics as transportation, energy, urban and rural land use, housing, environmental quality, public utilities, and natural resources.

Please visit us online at http://le.uwpress.org › Search across full text, abstracts, titles, tables of contents, fi ­ gures, and journals › View tables of content and abstracts › E-mail article information to a friend Land Economics 2010 Subscription Rates Individuals o Print & Electronic, $90 o Electronic Only, $84 Institutions o Print & Electronic, $268 o Electronic Only, $242

› Sign-up for e-mail alerts › View most read papers list › View most-cited papers list › View FREE sample issue

___________________________ Name

___________________________ Address

___________________________ Address

___________________________ City

State

Payment Information:

Please remit funds in U.S. currency. Canadian subscribers add 5% goods and services tax.

o VISA o MC $ ___________

Total charge

__ __ __ __-__ __ __ __-__ __ __ __-__ __ __ __ Credit Card no.

Non U.S. Postage (no postage

(_____) ______ - __________ ___________________________ ____________ Card Exp. Date home phone no.

o airmail: add $35/year

___________________________ ____________________________

charges for electronic only subscriptions)

Zip

E-Mail address

Country

Signature

University of Wisconsin Press, Journals Division 1930 Monroe Street, 3rd Fl., Madison, WI 53711-2059 Phone: (608) 263-0668 • Fax: (608) 263-1173 • US only Toll-free Fax: (800) 258-3632 [email protected] • uwpress.wisc.edu/journals ebook