Balancing agricultural development in the

Balancing Agricultural Development and Deforestation

in the Brazilian Amazon

Andrea Cattaneo

RESEARCHREPORT 129INTERNATIONAL FOOD POLICY RESEARCH INSTITUTEWASHINGTON, D.C.

Copyright 2002 International Food Policy Research Institute. All rights reserved.Sections of this book may be reproduced without the express permission of, but withacknowledgment to, the International Food Policy Research Institute.Printed in the United States of America

International Food Policy Research Institute2033 K Street, NWWashington, DC, 20006-1002 U.S.A.Telephone +1-202-862-5600www.ifpri.org

Library of Congress Cataloging-in-Publication Data

Cattaneo, Andrea.Balancing agricultural development and deforestation in the Brazilian

Amazon / by Andrea Cattaneo.p. cm. (Research report ; 129)

Includes bibliographical references.ISBN 0-89629-130-8 (alk. paper)1. Sustainable developmentBrazil. 2. Sustainable

developmentAmazon River Region. 3. DeforestationBrazilEconometricmodels. 4. DeforestationAmazon River RegionEconometric models. 5.Agriculture and stateBrazil. 6. Agriculture and stateAmazon RiverRegion. 7. Land reformBrazil. 8. Land reformAmazon River Region.9. Agricultural innovationsBrazilEconometric models. 10.Agricultural innovationsAmazon River RegionEconometric models. 11.BrazilEconomic policy. I. Title. II. Series : Research report(International Food Policy Research Institute) ; 129.

HC188.A5 C334 2002338.98107dc21 2002011967

Contents

Tables iv

Figures vii

Foreword ix

Acknowledgments x

Summary xi

1. Introduction 1

2. Deforestation in the Brazilian Amazon 5

3. Modeling Interactions between the Environment and the Economy 12

4. Model Database 40

5. The Effects of Macroeconomic, Interregional, and Intraregional Change 61

6. Policy Conclusions 106

Appendix A: The CGE Model 111

Appendix B: Structure of the Economy 118

Appendix C: Elasticities and Sensitivity Analysis 120

Appendix D: Results for Devaluation Scenarios 122

Appendix E: Results for Transportation Cost Reduction Scenarios 128

Appendix F: Results for Amazonian Technological Change Scenarios 131

Appendix G: Results for Non-Amazonian Technological Change Scenarios 134

References 137

iii

Tables

2.1 Regions and countries owning major tropical forest stocks and extent of deforestation, 19901995 (thousands of hectares) 5

2.2 Mean rate of gross deforestation by state, 19781998 (kilometers2/year) 7

2.3 Global carbon budget for the 1980s showing major emission sources and uptakes (million metric tons/year) 10

3.1 Mapping of economic activities, commodities produced, and factors used (as adopted in the model) 24

3.2 Production technology: Substitutability between agricultural commodities 26

3.3 Definitions of parameters and variables in the simplified model 28

3.4 Model description (simplified version with no intermediate goods) 30

4.1 A basic national social accounting matrix 40

4.2 Data sources for the macroeonomic social accounting matrix 43

4.3 Macroeconomic social accounting matrix for Brazil, 1995 (current R$ billion) 44

4.4 Value-added structure for Brazil, 1995 48

4.5 Structure of the Brazilian national commodity markets, 1995 49

4.6 Value added of regional agricultural output, differentiated by producer size (R$ billion) 50

4.7 Farm establishments by size and land ownership concentration 51

4.8 Evolution of land ownership concentration: The Gini coefficient, 19701995 52

4.9 Number of families that have benefited from land reform projects 52

4.10 Gross migration between regions inside Brazil, 19911996 (number of persons) 53

4.11 Average wages and population for each region 56

4.12 Propensity of unskilled labor to migrate and wage differential threshold (%) before movement occurs between two regions 57

iv

5.1 Factor mobility and utilization for short-run and long-run scenarios 63

5.2 Types of technological change 74

5.3 Shift from natural to planted pasture, by region, 19851995 84

5.4 Increase in annual yields in two regions 19851995 (%) 86

5.5 Replicating productivity improvements for the period 19851995: A retrospective scenario (% change) 87

5.6 Change in per capita agricultural income associated with non-Amazon technological change: Scenario replicating innovation for 19851995 90

5.7 Percent change in income to different types of producers based on the tax and subsidy scenario 104

6.1 A qualitative comparison of the impacts of technological change in the Amazon 108

A.1 Definition of parameters and variables in the CGE model 111

A.2 Model equations 114

B.1 Regional production by commodity, 1995 (R$ billion) 118

B.2 Factor intensities for Amazon agriculture in the base (in terms of output) 119

C.1 Model elasticities 121

D.1 Short-run changes in production with currency devaluation, balanced-adjustment scenario (%) 122

D.2 Short-run effects of currency devaluation on per capita income, balanced-adjustment scenario (%) 123

D.3 Short-run changes in macroeconomic aggregates with currency devaluation, balanced- adjustment scenario (%) 123

D.4 Long-run changes in production with currency devaluation, balanced-adjustment scenario (%) 123

D.5 Long-run effects of currency devaluation on per capita income, balanced-adjustment scenario (%) 124

D.6 Long-run changes in macroeconomic aggregates with currency devaluation, balanced-adjustment scenario (%) 124

D.7 Short-run changes in production with currency devaluation, capital-flight scenario (%) 125

D.8 Short-run effects of currency devaluation on per capita income, capital-flight scenario (%) 125

D.9 Short-run changes in macroeconomic aggregates with currency devaluation, capital-flight scenario (%) 126

D.10 Long-run changes in production with currency devaluation, capital-flight scenario (%) 126

TABLES v

D.11 Long-run effects of currency devaluation on per capita income, capital-flight scenario (%) 127

D.12 Long-run changes in macroeconomic aggregates with currency devaluation, capital-flight scenario (%) 127

E.1 Short-run changes in production with reduction in transportation costs (%) 128

E.2 Short-run effects of reduction in transportation costs on per capita income (%) 129

E.3 Short-run changes in macroeconomic aggregates with reduction in transportation costs (%) 129

E.4 Long-run changes in production with reduction in transportation costs (%) 129

E.5 Long-run effects of reduction in transportation costs on per capita income (%) 130

E.6 Long-run changes in macroeconomic aggregates with reduction in transportation costs (%) 130

F.1 Wage impact of technological change in the Amazon (% change) 131

F.2 Short-run impact of technological change on different producer types (%) 133

G.1 Change in per capita regional agricultural income for non-Amazonian technological change: Decomposing the impact of innovation on producers by type of activity 134

vi TABLES

Figures

1.1 Main agricultural development areas in the Amazon 2

2.1 Distribution of deforestation in the 19911996 period showing deforestation intensity 9

3.1 Factors affecting the appropriate structure of a CGE model 14

3.2 CGE structure showing the circular flow of income 20

3.3 Land transformation/conversion flows 24

3.4 Sectoral production technology 25

3.5 Flow of goods from regional producers to the national composite commodity 27

3.6 Markov chain representation of biophysical transformation processes 38

4.1 Migration as a function of wage differentials between regions 55

4.2 Net migration flows between regions used in the estimate of migration functions 56

4.3 Transition network for estimation of transformation processes 59

5.1 Logging in the Amazon: Balanced-contraction versus capital-flight scenarios in the short and the long run 64

5.2 Effects of balanced-contraction versus capital-flight scenarios on deforestation in the short and the long run 66

5.3 Change in deforestation rates if infrastructure in the Amazon is improved 69

5.4 Impact on the deforestation rate of regulating access to property rights 72

5.5 Change in deforestation rates from technological change in annuals production 75

5.6 Short-run change in value added in the Amazon region from technological change in annuals production, large and small farms 77

5.7 Short- and long-run changes in deforestation rates from technological change in perennial production 79

5.8 Short-run change in value added in the Amazon region from technological change in perennials production, small and large farms 80

vii

5.9 Short- and long-run changes in deforestation rates from technological change in animal production 81

5.10 Short-run change in value added in the Amazon region from technological change in animal production 83

5.11 Change in deforestation rates for non-Amazon technological change: Scenario replicating innovation, 19851995 87

5.12 Change in deforestation rates for non-Amazon technological change: What if innovation in livestock had not occurred? 88

5.13 Change in deforestation rates for non-Amazon technological change: Decomposing the impact of innovation among activities and regions where it occurs 93

5.14 Change in per capita regional agricultural income: Decomposing the impact of innovation by type of activity and region where it occurs 94

5.15 Impact of tax and subsidy scenarios on deforestation, logging, and extractive activities 102

5.16 Impact of tax and subsidy scenarios on extractive activities and government revenue (R$ billion) 103

viii FIGURES

Foreword

Balancing environmental sustainability and poverty alleviation is particularly chal-lenging in the rainforests of the Brazilian Amazon. Therefore, in line with its missionto identify and analyze policies that help meet the food needs of hungry people without further degrading the natural resource base, IFPRI undertook a three-year researchprogram in the Brazilian Amazon, as part of the CGIARs initiative on alternatives to slash-and-burn agriculture. This research is also highly relevant for an understanding of long-termclimate change linkages.

Although the Brazilian government has recently eliminated policies that inadvertently of-fered incentives to clear the land, deforestation rates have not decreased. This suggests thatthere are additional causes of deforestation. From a broad perspective, this research looks atthe links between different types of agricultural producers, the logging industry, and the over-all institutional setting. It examines these interactions at different levels of geographic aggre-gation, ranging from survey-based research on small farms in the Western Brazilian Amazonto more aggregate regional and macroeconomic scales relying on secondary data. This reportillustrates the economic and environmental effects of macroeconomic and Amazon-wide poli-cies, considers them within a consistent framework, and shows how the Amazon fits into therest of the Brazilian economy.

To do this, a model was developed to simulate the effects on the Brazilian economy of pol-icy changes, currency devaluation, land tenure regime changes, infrastructure development,and the adoption of new agricultural technologies. The effects of these changes on deforesta-tion and the welfare of farmers and loggers are analyzed in depth. The interesting results,which are at times counterintuitive, shed new light on why slowing deforestation in the Ama-zon is so difficult, and on the trade-offs between environmental and economic goals.

While this report looks at the Amazon-wide mechanisms at work behind deforestation, acompanion IFPRI research report, Agricultural Intensification by Smallholders in the WesternBrazilian Amazon: From Deforestation to Sustainable Land Use, examines several small-holder settlements on the agricultural frontier. Together, these two reports offer a comprehen-sive look at an important set of issues around valuable, related resources.

Joachim von BraunDirector General, IFPRI

ix

xAcknowledgments

Iwish to thank Steve Vosti, Sherman Robinson, and John Boland for the support and train-ing received during the elaboration of this report. I am also grateful to Chantal Line Car-pentier, Julie Witcover, and Monica Scatasta for their considerable input and valuable in-sights, which helped shape this research. Participants in brown-bag seminars in the Trade andMacroeconomics and the Environment and Production Technology divisions at the Interna-tional Food Policy Research Institute (IFPRI) also provided useful suggestions. I am espe-cially grateful to Peter Hazell for his helpful comments as IFPRIs internal reviewer. To theexternal reviewers of the manuscript, Tom Tomich and an anonymous referee, I also owe aparticular debt of gratitude for a comprehensive evaluation. Their constructive criticism andextensive list of suggestions have vastly improved the quality of the work.

On the Brazilian front, I would like to thank Eustaquio Reis and all the staff at Institutode Pesquisa Econmica Aplicada in Rio de Janeiro for making this research possible and fortheir comments. I am also grateful to Marisa Barbosa, Yoshihiko Sugai, Antonio Teixeira-Filho, and the staff at the Brazilian Agricultural Research Corporation (Empresa Brasileira dePesquisa AgropecuriaEmbrapa) in Brasilia for the support they provided, for the crucialdata on agricultural technologies, and for the feedback. This research also relied on the sup-port of EmbrapaAcre and EmbrapaRondnia, and special thanks go to Judson Valentim andSamuel Oliveira. I would also like to thank the participants at the Center for InternationalForestry Research workshop on Technological Change in Agriculture and Deforestation,held in Costa Rica in March 1999, who provided useful comments.

For financial support, my gratitude goes to the U.S. Environmental ProtectionAgency (EPA) for the STAR graduate fellowship that supported me throughout the doctoralprogram. Substantial financial and logistical support was also given by IFPRI and the Dan-ish development agency (DANIDA) via the Alternatives to Slash-and-Burn Program led bythe International Center for Research in Agroforestry. The trips to the field in the BrazilianAmazon would not have been possible without the support of both EPA and IFPRI.

xi

Summary

The Brazilian Amazon, one of the worlds largest tropical forests, lost 128,000 squarekilometers to deforestation between 1980 and 1995. Agricultural development, log-ging, and ranching are often identified as the proximate causes. However, the under-lying causes of deforestation are rarely discussed in depth.

This report identifies the links between economic growth, poverty alleviation, and naturalresource degradation in Brazil. It examines the effects of the following national and regionalpolicies and events: (1) a major devaluation of the Brazilian real (R$); (2) improvements ofinfrastructure in the Amazon to improve links with the rest of Brazil and bordering countries;(3) modification of land tenure regimes in the Amazon agricultural frontier; (4) adoption oftechnological change in agriculture both inside and outside the Amazon; and (5) fiscal mech-anisms to reduce deforestation rates.

Studying the impact of such phenomena requires an economy-wide view, since the eco-nomic activities in other sectors and regions of the Brazilian economy are increasingly linkedto those in the Amazon. To this end, IFPRI developed a regionalized computable general equi-librium (CGE) model, which divides Brazil into four regions: the Amazon, Northeast, Cen-ter-West, and South/Southeast. In the model, relative product prices, factor availability, trans-portation costs, and available technology are all assumed to influence land use; biophysicalprocesses as well as decisions of economic agents are assumed to change land cover. Agri-cultural production activities are broken down by region, sector, and size of operations. A de-forestation sector produces arable land used by agricultural producers. Within this frame-work, land uses (including deforestation), incomes, wage rates, and other aspects of the econ-omy are estimated and differentiated by region.

Looking at the effects of devaluations ranging from 10 to 40 percent, the report finds thatunder a devaluation of 40 percent, nationally, GDP would decrease, urban poverty would in-crease, future growth would be undermined, and tradable agricultural goods would expand.In the Amazon itself, a devaluation of 40 percent has these results: Deforestation rates would vary depending on the governments crisis plan. If the govern-

ment balances reduction of private consumption, government demand, and investment,deforestation rates would decline by 10 percent in the short run and by 2 percent in thelong run. However, government inaction and capital flight would lead to a 6 percent in-crease in deforestation in the short run and 20 percent in the long run-about 4,000 addi-tional square kilometers per year.

Logging would increase by 1620 percent depending on government action. The Amazon would fill the domestic demand gap created as other regions move toward

tradables. Following the devaluation, agricultural expansion in the Amazon would

xii SUMMARY

center on production of a variety of annual crops and livestock, as other regions producemore coffee, soy, horticultural goods, and sugar.

The Brazilian governments strategy for Amazonian development, as part of its Avana Brasil(Forward Brazil) plan, includes an ambitious program of infrastructure investments of US$45billion in 19992006. This analysis finds that a resulting 20 percent reduction in transporta-tion costs for all agricultural products from the Amazon would increase deforestation by 15percent in the short run and by 40 percent in the long run (about 8,000 square kilometers ayear). As returns to arable land rise, the incentive to deforest would increase, leading to a 24percent increase in production by smallholders and a 9 percent increase by large farms. Na-tionally this would have little effect on welfare, because the increase in production in theAmazon, except for the share that is exported, would replace production from other regions.

Regulating tenure regimes is one of the best options for reducing deforestation in theAmazon. A substantial share of past deforestation occurred at the hands of deforesters whoacquired informal land tenure in the process. The Brazilian government is now uncoveringfraudulent land claims, reclaiming the land, and moving toward a unified land registry sys-tem. Removing the speculative incentive to deforest could reduce the deforestation rate by 23percent, saving up to 5,500 square kilometers per year.

Agricultural technologies play an important role in determining agricultural developmentand deforestation. Within the Amazon, the relative profitability and land intensities of differ-ent activities, combined with soil productivity and sustainability limits, are factors that affectagricultural producers incomes and determine, in part, the pressures on forests through thedemand for cleared land. The impact of improvements in Amazonian agricultural technolo-gies will depend on which activity is innovated. Livestock technology improvements appear to yield the greatest returns for all agricultural

producers in the Amazon and should improve food security in the region, but deforesta-tion increases dramatically in the long run.

Perennial crop technology improvements could theoretically reduce deforestation ratesconsiderably, but this is unlikely to happen. Small farmers stand to gain the most fromsuch improvements, but they are averse to the risks inherent in perennial crops. Largefarmers are unlikely to adopt the new technologies because their gains would be small.

Annual crop technology appears to have little potential. Income gains would be quitesmall. Before reaching the high land intensity required to reduce deforestation rates, therewould almost certainly be a period in which deforestation would increase substantially.

Outside of the Amazon, the agricultural technological change that took place during 198595affected deforestation in drastically different ways. Overall, deforestation rates were 1535percent lower than if improvements had not occurred outside the Amazon, largely as a resultof innovation in livestock technologies. In fact, improvements in annuals and some perenni-als alone would have led to a 20 to 27 percent increase in deforestation rates. Regionally, theNortheast and Center-West were the regions to gain in income from technological change.The income distribution gap apparently decreased in the Northeast and increased in the Cen-ter-West as a result of technological change outside the Amazon.

To take into account the nonmarket benefits and costs stemming from different land uses,the report considers both taxes and transfer payments. In spite of the link between logging anddeforestation, it finds that applying a logging tax in the Amazon would not lead to a decreasein the deforestation rate, but it would negatively affect the logging industry. A deforestationtax would be more effective: a tax of R$50 per hectare deforested would reduce deforestationrates around 9,000 square kilometers a year, with logging only minimally affected. The

downside of the deforestation tax is that it would have a substantial negative effect on smallfarmers in the Amazon.

An alternative scenario would be to subsidize forest conservation. For example, a 30 per-cent reduction in the deforestation rate could be obtained with a subsidy of R$240 per hectare.From a welfare standpoint, all regions stand to gain from a subsidy of this kind: the Amazonwould benefit directly, but the other regions would also gain by taking up the slack in the vol-ume of wood. Market benefits accrued nationwide would exceed the subsidy expenditures.The subsidy, equivalent to R$1.21 per carbon ton of reduced emissions, could be funded in-ternationally if Brazil were allowed compensation for reducing deforestation under carbontrading arrangements with other countries.

SUMMARY xiii

C H A P T E R 1

Introduction

The primary objective of this research is to identify the links between economic growth,poverty alleviation, and natural resource degradation in Brazil, with particular empha-sis on land use and deforestation in the Amazon. This report focuses on the impact ofpotential macroeconomic policy shifts in Brazil on deforestation and economic welfare, com-pared with the consequences of technological change in agriculture. The following set of pol-icy questions are applied to Brazil: What impact does a macroeconomic shock, such as currency devaluation, have on the

movement of the agricultural frontier in the Amazon? What will be the economic and environmental impact of forthcoming technological

changes in agricultural production inside and outside the Amazon region? What are the effects of lower transportation costs resulting from government investments

in physical infrastructure in the Amazon? What policy mechanisms are most effective in limiting deforestation without hindering

economic development? Policies considered are (1) fiscal incentives to account for the for-ests value in providing public goods, and (2) the modification of acquisition of propertyrights in the Amazon agricultural frontier to eliminate inefficient speculative behavior.

The Amazon rainforest covers an area of approximately 5.5 million square kilometers. Sixtypercent3.6 million square kilometersis located inside Brazil, encompassing nearly 40 per-cent of the countrys territory. In this report the Brazilian Amazon is defined as the North re-gion of Brazil plus northern Mato Grosso and western Maranho. This specification capturesthe ecological and agricultural characteristics typical of the tropical forest region. The Ama-zon so defined, however, still comprises a complex mosaic of forest (72 percent of land area),savanna (15 percent), inundated lowlands (8 percent), and ecological transition areas (5 per-cent). The savanna areas are important because large areas have been used for mechanizedsoybean cultivation and for pasture, despite generally poor soils.

The geographic expansion of the Brazilian agricultural frontier has certainly been the mostimportant activity directly involved in the Amazon deforestation process. Agropastoral landuses, particularly cropping and cattle ranching, have been the main cause of deforestation.Timber extraction, charcoal production, mining, and hydroelectric dams have been minor con-tributors, compared with agriculture, but to the extent that they stimulated agricultural settle-ments, they have played important causal roles.

Deforestation in the Brazilian Amazon occurs mainly along a band, varying in width be-tween 200 and 600 kilometers. This band stretches from the northeastern state of Maranho,through Par and Mato Grosso and includes colonization areas in Rondnia (Figure 1.1). Thefrontier expansion areas and the government-sponsored colonization areas came into being in

1

the 1960s and 1970s. However, there areareas in the floodplains of the Amazonbasin and upland regions in northeasternPar that were brought into agriculturalproduction in the 19th century. These latterforms of agriculture have adapted over timeto the environmental conditions. However,with the onset of roads, floodplain agricul-ture located along the riparian transporta-tion system lost its attractiveness.

In the state of Par, upland agriculture isa dynamic and diverse sector of the econ-omy but geographically constrained; there-fore, deforestation for agricultural purposesin this report implicitly refers to the frontierexpansion areas and the government-spon-sored colonization areas.

Since colonial times, the settlement ofnew frontiers has been undertaken to openaccess to land and other natural resources. Itis assumed that relative product prices, fac-tor availability, and transportation costs arethe main economic factors affecting themovement of a frontier. In this publication,the approach taken is the same as thatadopted by Findlay (1995), in which fron-tier movement is described as the process ofincorporating a periphery into an eco-nomic center through a network of trade,investment, and migration. In the Braziliancontext, high transportation costs betweenthe Amazon and the rest of the countryleading to high agricultural input costs andlimited interregional tradecharacterizethe frontier environment. This economic

2 CHAPTER 1

Figure 1.1 Main agriculture development areas in the Amazon

Floodplain (Varzea) Agriculture

Agriculture in Colonization Area

Agriculture in Northeast State of Para

Agriculture Frontier Expansion Area

Extraction Agriculture

Other Agriculture Occupation Areas

Amazonas

RoraimaBoa Vista

AmapMacap

Amazon River

Par

AltamiraSantarm

Mato GrossoTocantinsRondnia

Acre

Porto Velho

Rio Bravco

Cuiab

Belm

Maranho

So LuisManaus

PA

167

(Rio Amazonas)

Trans AmazonHig

hway

Source: Adapted from Nascimento and Homma 1984. In Serro and Homma 1993. Reprinted with thepermission of Dr.A.K.O. Homma.

intuition is confirmed by the work of Pfaff(1997), who finds that greater distance frommarkets south of the Amazon leads to lessdeforestation.

In Brazil, macroeconomic policies,credit and fiscal subsidies to agriculture,and technological change in Brazilian agri-culture have all acted as push factors in themigration process. Regional developmentpolicies have pulled economic resourcesthrough fiscal incentives for agropastoralprojects to attract investment, the expansionof the road network to stimulate trade, andcolonization programs to facilitate migra-tion (Binswanger 1991). While some ofthese policies are directed toward reducingpoverty, the most harmful ones from an en-vironmental perspective are not driven byequity concerns. The most important exam-ple is the fact that agricultural income hasbeen taxed at lower rates than nonagricul-tural income (a 1.56.0 percent tax rate onagricultural income versus a 3545 percentcorporate tax rate in manufacturing andservices), thereby converting agricultureinto a tax shelter. Small farmers and poorindividuals have been negatively affectedbecause the market price of land includes acomponent capitalizing these tax prefer-ences. This implies that the poor must cutconsumption below the imputed value offamily labor to pay for the land. Such a pol-icy leads to an increase in deforestation be-cause it creates an incentive for urban in-vestors and corporations to compete forland at the frontiers of settlement as well asin areas of well-established settlement, andbecause it encourages poor individuals tomove to the frontier in search of unclaimedland.

At an intraregional level, several inter-esting distorting provisions have been re-ported in the literature, including (1) rulesof public land allocation that provide incen-tives for deforestation because the securityof a claim is determined by land clearing(Binswanger 1991); (2) a progressive landtax that encourages the conversion of forestto crop land or pasture (Almeida and Uhl

1995); and (3) a tax credit scheme aimedtoward corporate livestock ranches thatsubsidized inefficient ranches establishedon cleared forest land (Browder 1988). Thefiscal incentives for agricultural productionin the Amazon, however, were withdrawnin the late 1980s in response to domesticfiscal concerns plus international criticismof Amazon policy (Lele et al. 2000). Withdiminished federal support, it was expectedthat some ranchers, where productivity waslow, would abandon their lots, as livestockproducers have done in other regions ofBrazil. However, profits for pasture systemspersisted even with less government sup-port (Faminow and Vosti 1998; Hecht 1993;Mattos and Uhl 1994; Valentim and Vostiforthcoming). While some have argued thatintensifying pasture systems could removepressure to deforest (Mattos and Uhl 1994;Arima and Uhl 1997), they did not alwaystake explicit account of all farm resources(Faminow, Pinho de Sa, and de MagalhesOliveira 1996) or long-run effects.

Other work at the regional level has em-phasized the combined role of expandingroad networks and rising agricultural de-mand in prompting population growth anddeforestation (see, for example, Pfaff1997), while documenting some role forgovernment policy. Using county-leveldata, Pfaff confirms the importance for de-forestation of some of the trends comingout of the policy push to develop the Brazil-ian Amazon. For instance, developmentprojects were linked to deforestation in the1970s but not the 1980s (but no robust rela-tionship regarding credit emerged). Closerproximity to markets to the south of theAmazon as well as higher road densitieswere associated with more deforestation,and early arrivals to a regionnot simplyhigher population densitieshad greaterimpact on the environment. Andersen(1996) similarly found that the importanceof federal policy for deforestation faded inthe 1980s in the face of local marketforceseconomic growth, populationgrowth, and locally funded roads.

INTRODUCTION 3

Schneider (1994) argued that increasedroad density in already settled areas andfewer roads reaching into new forest areaswould help provide sustainable livelihoodsfor forest inhabitants while protecting fur-ther encroachment on the forest. Somestudies point to the importance of propertyrights in Brazilian Amazonian deforesta-tion, including a role for land speculation(Alston, Libecap, and Mueller 1999;Kaimowitz and Angelsen 1998). Still othershave found that climatic conditions, princi-pally high precipitation levels, in effect pre-vent conversion of forest to agriculture (orpromote abandonment of that land) evencontrolling for some market linkages, andthat agriculture offers low private returnswhere practiced (Chomitz and Thomas2000).

At the local level, an issue open to de-bate is whether deforestation is primarilycarried out by smallholders or by large farmenterprises, and whether the smallholdersgoal is to plant crops or install pasture. Ac-cording to Homma et al. (1998), each of the600,000 smallholders present in the Brazil-ian Amazon clears, on average, 23hectares of forest and cultivates it for two tothree years. This implies that smallholdersclear approximately 600,000 hectares annu-ally. An alternative view holds that com-mercial ranching has been the largest con-tributor to the deforestation process. How-

ever, as Mahar (1989) points out, some ofthe deforestation attributed to livestock op-erations may have been caused by thespread of small-scale agriculture, since landdevoted to annual crops is often convertedto pasture after a few years when yields decline.

The Brazilian Amazon, with a popula-tion of 16 million (61 percent urban), de-pends to a large extent on local productionmarketed by both small-scale farmers andlarge-scale enterprises. While the move-ment of the agricultural frontier is the majorcontributor to the deforestation process, therole of agricultural producers in ensuringfood security for the region requires a care-ful analysis of how to reduce deforestationrates without negatively affecting farmerslivelihoods and the regional food supply.Agricultural technologies play an importantrole in determining agricultural develop-ment and deforestation. The relative prof-itability and land intensities of different ac-tivities, combined with soil productivityand sustainability limits, are all factors thataffect agricultural producers incomes anddetermine, in part, the pressures on forests.The results to be presented in this reportcompare the magnitude of these effects relative to those of economic processes andpolicy changes occurring outside the Amazon.

4 CHAPTER 1

C H A P T E R 2

Deforestation in the Brazilian Amazon

Brazil in Context: Comparing Tropical Deforestation RatesAround the World

A comprehensive assessment of the state of the worlds forests, released by the Food andAgriculture Organization of the United Nations (FAO), indicates that total forestedarea continued to decline significantly in the 1990s (FAO 1999). According to FAOsanalysis, deforestation is concentrated in the developing world, which lost approximately 62million hectares between 1990 and 1995 (Table 2.1).1 The result is an average annual loss indeveloping countries of 12.5 million hectares. This constitutes a slight decline relative to198090, when annual forest loss in developing countries was estimated at 15.5 millionhectares.

Various combinations of agricultural development, logging, and ranching claimed much ofthe 239,000 square kilometers of forest lost in this South America between 1980 and 1995the largest loss of forest in the world during those years. Brazil alone lost 128,000 square

5

Table 2.1 Regions and countries owning major tropical forest stocks and extent of de-forestation, 19901995 (thousands of hectares)

Total forest Total Annual change Annual change

Region/country 1990 1995 199095 change rate (%)

Africa 538,978 520,237 18,741 3,748 0.7Congo, Democratic Republic of 112,946 109,245 3,701 740 0.7

Asia 517,505 503,001 14,504 2,901 0.6Indonesia 115,213 109,791 5,422 1,084 1.0

Central America 84,628 79,443 5,185 1,037 1.3Mexico 57,927 55,387 2,540 508 0.9

South America 894,466 870,594 23,872 4,774 0.5Brazil 563,911 551,139 12,772 2,554 0.5

Total developing countries 2,035,577 1,973,275 62,302 12,460 0.6

Source: FAO 1999.

1In North America, Europe, and Oceania, reforestation efforts, new forest plantations, and the gradual regrowthand expansion of forest account for an increase of about 6 million hectares of forest cover.

kilometersmore than one-fifth of all trop-ical forest lost worldwide during that time.Nevertheless, South America maintainsvast areas of intact tropical and temperateforest. The northern Amazon Basin and theGuyana Shield house the largest tropicalfrontier forests anywhere. In fact, the an-nual deforestation rate for South America,in percentage terms, is lower than the aver-age in the developing world.

Addressing deforestation at a systemiclevel requires the removal of both marketfailures and policy failures. While some is-sues may be addressed at the internationallevel others are best solved at the nationallevel. The data in Table 2.1 concerningstanding forests and deforestation highlightthe future role that a few countries likeBrazil, Indonesia, Mexico, and the Democ-ratic Republic of Congo can play in reduc-ing deforestation rates. These four countriesinclude within their borders approximately42 percent of standing forest in developingcountries and account for 39 percent of alldeforestation. Since policy solutions needto be tailored to specific national politicaland economic environments, it makes senseto focus on those countries that own thelargest shares of standing forest and, in par-ticular, on Brazil.2

Trends and Geographic Distribution of Deforestationin the Brazilian AmazonOfficial estimates of Brazilian deforestationrates are released on an annual basis with a

delay of one-to-two years. Forest conver-sion to agriculture is readily monitoredfrom space using imagery from the LandsatThematic Mapper (TM) satellites, permit-ting the development of deforestation mapsof large regions at a reasonable cost andspeed. As can be seen in Table 2.2, there isconsiderable variation from year to yearand across states. After a substantial declineduring 198991, the trend in deforestationappears to have spiked sharply in 1995.There is some debate, however, aboutwhether deforestation rates did indeedspike in 1995. A possible explanation isthat forest losses that took place over theprevious two years or so did not register onaerial images due to cloud cover or othercomplexities of interpretation. If so, what isperceived to be a rapid rise in deforestationin 1995 would instead be a cumulative ef-fect.3 Another possibility might be that theincrease in deforestation during 199395was mainly the result of accidental forestfires (Lele et al. 2000).

In the second half of the 1990s defor-estation rates varied between 13,000 squarekilometers per year in 1996/97 to 18,200square kilometers per year in 1999/2000.Although the average deforestation rate inthe second half of the decade (approxi-mately 17,000 square kilometers per year)is much lower than the 1994/95 peak (Table2.2), it is apparently increasing and may re-turn to the 197795 historical average ofabout 19,400 square kilometers per year.

The state-by-state information in Table2.2 indicates that Par, Mato Grosso, and

6 CHAPTER 2

2There are countries, for example in Central America, where the need to curb deforestation may be greater be-cause their forest base is smaller and they are deforesting at a faster rate. This is particularly relevant for biodi-versity maintenance and local benefits such as hydrologic functions. It is less relevant for greenhouse gas emis-sions for which tropical deforestation is quite similar no matter where it occurs. 3Alves (1999) reports that approximately one-sixth of the area of study was covered by clouds in the first threesurveys. In the 1995/96 survey, 30 percent of the area of study was not observed because of clouds. Cloud-covered areas appeared predominantly in Amap, Roraima, and some areas near the Atlantic Ocean in Maranhoand Par. The state of Amap was excluded from the present analysis because of frequent cloud cover over 60percent or more of its territory.

DEFORESTATION IN THE BRAZILIAN AMAZON 7

Tabl

e 2.

2 M

ean

rate

of g

ross

def

ores

tatio

n by

sta

te, 1

978

1998

(kilo

met

ers2

/yea

r)

Stat

e19

77/8

8a19

88/8

9a19

89/9

0a19

90/9

1a19

91/9

2a19

92/9

4 b

1994

/95

1995

/96

1996

/97

1997

/98

1998

/99

1999

/200

0

Acr

e62

054

055

038

040

048

21,

208

433

358

536

441

547

Am

ap

6013

025

041

036

9

18

30

A

maz

onas

1,51

01,

180

520

980

799

370

2,11

41,

023

589

670

720

612

Mar

anh

o2,

450

1,42

01,

100

670

1,13

537

21,

745

1,06

140

91,

012

1,23

01,

065

Mat

o G

ross

o5,

140

5,96

04,

020

2,84

04,

674

6,22

010

,391

6,54

35,

271

6,46

66,

963

6,36

9Pa

r6,

990

5,75

04,

890

3,78

03,

787

4,28

47,

845

6,13

54,

139

5,82

95,

111

6,67

1R

ond

nia

2,34

01,

430

1,67

01,

110

2,26

52,

595

4,73

02,

432

1,98

62,

041

2,35

82,

465

Ror

aim

a29

063

015

042

028

124

022

021

418

422

322

025

3To

cant

ins

1,65

073

058

044

040

933

379

732

027

357

621

624

4A

maz

on21

,050

17,7

7013

,730

11,0

3013

,786

14,8

9629

,059

18,1

6113

,227

17,3

8317

,259

18,2

26To

tal

42,1

0035

,540

27,4

6022

,060

27,5

7229

,792

58,1

1836

,322

26,4

5434

,766

34,5

1836

,452

Sour

ce:

INPE

200

2.N

ote:

The

lead

ers

indi

cate

a n

il or

neg

ligib

le a

mou

nt.

a Mea

n ov

er th

e de

cade

b Bie

nnia

l mea

n

Rondnia have consistently been the stateswith the largest areas being deforestedthroughout the 1990s. Where deforestationis occurring is important to this research be-cause, given the aggregate nature of theanalysis, it would be ideal to include in therepresentation of the Brazilian Amazononly those areas that are along the currentarc of deforestation or likely to face defor-estation pressures in the future.

Alves (2001) presents the geographicdistribution of deforestation over the period199196 using a 1/4 grid cell decomposi-tion of the Amazon. These cells are dividedby the author into three major deforesta-tion intensity categories (high, medium,low), based on the extent of deforestationthat occurred during 199195 (Figure 2.1).The high intensity cells, for example, aredefined as the group of 1/4 grid cells rep-resenting 33 percent of total deforestation,that is, the set formed by the cells that rep-resented the most deforested area andamassed 33 percent of total deforestation.The other categories are similarly defined.In Figure 2.1 one can see that a small sub-set of cells accounted for a large share ofdeforestation. Alves (2002) reports that approximately 25 percent of the total ob-served deforestation can be accounted forby just the 3.8 percent of grid cells with themost deforestation while 9.7 percent of thecells accounted for 50 percent of total de-forestation. Furthermore, 75 percent of thetotal observed deforestation is accounted by19.4 percent of these cells. This shows thatthe deforestation process tended to be con-centrated over an arc extending from Acreand Rondnia in the west, through northern

Mato Grosso, into Par and Maranho inthe east.

Since this report is focused on the de-forestation frontier and how it interacts withthe rest of the economy, the regional distri-bution of deforestation presented by Alveswas influential in determining the regionaldisaggregation adopted in the model pre-sented here. In particular, the areas of MatoGrosso, Maranho, and Tocantins to be in-cluded as part of the Amazon were deter-mined by including all cells with high in-tensity deforestation occurring and the ma-jority of the medium intensity cells (com-patibly with the definitions of micro-re-gions adopted by IBGE).4 The reason thiscriterion was adopted is that for modelingpurposes, given the aggregate level ofanalysis, the Amazon should include onlyeconomic activities on land that is stillforested or along the arc of deforestation.The deforestation frontier shown in Figure2.1 is an approximate representation ofwhat is considered here to be the border ofthe Amazon in terms of areas facing defor-estation pressures.

Although the deforestation rates re-ported in Table 2.2 are referred to through-out this report, the exact rate at which theAmazon forest is presently being destroyedis not known. Besides the margin of errorassociated with ambiguous scenes andcloud cover, the classification by Institutode Pesquisas Espaciais (INPE) reflects a di-chotomy between forest and nonforest thatis indeed useful for capturing the mainhuman effects on tropical forests (such asdeforestation by ranchers and farmers). Butit neglects those forest alterations that re-duce tree cover but do not eliminate it, such

8 CHAPTER 2

4The Legal Amazon is defined under Brazilian law as the area comprised by the states of Acre, Amap, Ama-zonas, Maranho (west of the 44 meridian), Mato Grosso, Par, Rondnia, Roraima and Tocantins. The regionaldisaggregation adopted here using the intensity of deforestation as the criterion excludes southern Mato Grosso,Eastern Maranho, and most of Tocantins. The micro-regions in Mato Grosso that were included in our defini-tion of the Amazon are: Alta Floresta, Alto Guapor, Alto Tele Pires, Arinos, Aripuan, Colder, Jauru, NorteAraguaia, Parecis, Sinop, Tangar da Serra. For Maranho we included: Alto Mearim e Graja, Gurupi, Impera-triz, Pindar, Porto Franco. For Tocantins the included micro-regions are: Araguana, Bico do Papagaio.

as logging and surface fires in standingforests. The forest openings created by log-ging and accidental surface fires are visiblein Landsat TM images, but they are coveredover by regrowing vegetation in one to fiveyears and are easily misclassified withoutaccompanying field data. Although loggingand forest surface fires usually do not killall trees, they severely damage forests. Log-

ging companies in the Amazon kill or dam-age 1040 percent of the living biomass offorests by harvesting trees (Verissimo et al.1992). Based on field surveys of wood millsand forest burning across the BrazilianAmazon, Nepstad et al. (1999) find thatlogging crews severely damage10,00015,000 square kilometers per yearof forest that is not included in deforestation

DEFORESTATION IN THE BRAZILIAN AMAZON 9

High intensity

Medium intensity

Low intensity

Deforestation frontier

Figure 2.1 Distribution of deforestation in the 19911996 period showing deforestationintensity

Source:Notes: The deforestation frontierwas defined for this report relying on data in the agricultural census bymicro-

region and on data inAlves 2001 and 2002.Reprintedwith permission ofDr.D. S.Alves.

Adapted fromAlves 2002.

5Nepstad et al. (1999), by stating that the area impacted by logging is additional to official estimates of defor-estation, implicitly assume that areas deforested for agricultural purposes and those impacted by logging are sep-arate and independent from one another.

mapping programs.5 While this additionalforest area is not explicitly included in theresults presented here on the effects ofhuman use, it is taken into considerationwhen the complementary relationship be-tween logging and deforestation is mod-eled. This is an important aspect of themodeling effort, given that robust domestictimber demand combined with the exhaus-tion of forest in Southeast Asia mean log-ging in the Amazon is likely to grow in thenear future. Therefore, even though logginghas been determined to be a historically lessimportant factor in Brazilian Amazoniandeforestation than agriculture, its role in theregion is becoming more significant (Leleet al. 2000; Reis and Margulis 1991).

Greenhouse Gas Emissionsfrom Deforestation in Brazil

Evidence is building in the scientific com-munity that the continued release of green-house gases (GHG) threatens to raise thetemperature of the earth and disrupt the cli-mates we depend on. Most of the increasein atmospheric carbon dioxide (CO2) con-centrations has come from the use of fossilfuels (coal, oil, and natural gas) for energy,but 2025 percent of the increase over thelast 150 years can be attributed to changesin land use: for example, the clearing offorests and the cultivation of soils for foodproduction. This contribution is confirmedby Table 2.3, which represents the emis-sions and uptakes (absorption) of carbonduring the 1980s, compiled by the Intergov-ernmental Panel on Climate Change(IPCC). The net release of carbon fromchanges in land use averaged 1.60.7 mil-

lion metric tons6 per year, representing 23percent of the total emissions.

Estimates of the emissions from defor-estation in the Brazilian Legal Amazon varyaccording to the accounting frameworkadopted: (1) net committed emissions refersto the long-term total of emissions and up-takes set in motion by the act of deforesta-tion, and it is calculated only for the areacleared in a given year (for example, the13.8 x 103 kilometers2 cleared in 1990); (2)annual balance refers to the emissions anduptakes in a single year (such as 1990) overthe entire landscape (the 415.2 x 103 kilo-meters2 cleared by 1990). The current bestestimate for 1990, according to Fearnside(1999), is 267 x 106 tons of carbon for netcommitted emissions, or alternatively, 353x 106 carbon tons for the annual balancefrom deforestation (plus an additional 62 x10 6 carbon tons from logging).7 The mag-nitude of these emissions can be appreci-ated by comparison with global emissionsfrom automobiles: the worlds 400 millionautomobiles emit 550 x 106 carbon tons

10 CHAPTER 2

Table 2.3 Global carbon budget for the 1980s showing majoremission sources and uptakes (million metric tons/year)

Source Emissions

Fossil fuels 5.5 0.5Tropical deforestation 1.6 1.0

Sink Uptake

Atmospheric buildup 3.3 0.2Ocean uptake 2.0 0.8Forest regrowth (Northern Hemisphere) 0.5 0.5Land sink (by difference) 1.3 1.5

Sources: IPCC 1996.

6For the purposes of this report, all tons are metric tons.7Considerable uncertainty still surrounds the overall extent of Brazils contribution to greenhouse gas emissions.This uncertainty will be reduced when the national inventory, now being compiled by Brazils Ministry of Sci-ence and Technology, is completed. It is following the standardized methodology developed by the IPCC.

annually (Flavin 1989). If one comparesBrazils emissions from land use and coverchange with those from fossil fuels (ap-proximately 75 x 106 carbon tons), one real-izes the importance deforestation has in de-termining Brazils greenhouse gas emis-sions. Therefore, in the future Brazil maystand to gain financial benefits from reduc-ing deforestation if the international com-munity decides it is a viable tool for limit-ing global warming.

It is now widely accepted that one of themain problems, if not the main problem, forattempts at maintaining forest cover is thatit is only rarely a viable financial proposi-tionwhile forest exploitation, like one-offlogging and deforestation, continue to behighly profitable activities. The global ex-ternality associated with greenhouse gasemissions associated with deforestation canbe viewed as a case of missing markets for

environmental services such as carbon se-questration and biodiversity conservation.International payments, transferring finan-cial resources from consumer nations inrecognition of the global public good valuesof forests, appear to have real potential. Ex-amples of such mechanisms are the GlobalEnvironment Fund (GEF), set up in 1991 asa financing mechanism for the InternationalConventions on Climate Change and Bio-logical Diversity, and the Clean Develop-ment Mechanism (CDM) defined under theKyoto Protocol.8

This report considers two categories ofcorrective actions: one exploits fiscalmechanisms that create disincentives to de-forest; the other provides payments (eithernational or international funds) to compen-sate producers for the forgone profits asso-ciated with reduced emissions. These issuesare discussed in greater detail in Chapter 5.

DEFORESTATION IN THE BRAZILIAN AMAZON 11

8In December 1997 more than 160 nations met in Kyoto, Japan, to negotiate binding limitations on greenhousegases for the developed nations, pursuant to the objectives of the Framework Convention on Climate Change ofl992. The outcome of the meeting was the Kyoto Protocol, in which the developed nations agreed to limit theirgreenhouse gas emissions, relative to the levels emitted in 1990.

C H A P T E R 3

Modeling Interactions between the Environment and the Economy

This chapter presents a regionalized computable general equilibrium (CGE) model inwhich Brazil is subdivided into regions compatible with the major administrative sub-divisions adopted by the Brazilian government: Amazon, Northeast, Center-West, andSouth/Southeast. For the Legal Amazon, the following processes are considered: (1) conver-sion of forests to cleared land (which depends on agents economic decisions), and (2) trans-formation of land from cleared land to grassland, and (3) subsequent transformation fromgrassland to an unproductive state.

The starting point for the regionalized CGE model is a nationwide model developed in199596 as an ongoing collaborative effort between the International Food Policy ResearchInstitute (IFPRI) and the Brazilian National Development Bank (BNDES).

The regional model has two components: a CGE model, which represents the behavior ofeconomic agents, and a land transformation model, which is a simplified representation of bio-physical processes affecting land productivity. This chapter begins with a brief survey of theapproaches that have been adopted to address deforestation issues using CGE models, fol-lowed by a description of the characteristics of the CGE model used for this research and a de-scription of how the biophysical processes are represented.

General Equilibrium Models: From Theory to Practice

In general equilibrium theory, the goal is to formulate a model of simultaneous equilibrium incompetitive markets for all commodities that is a precise logical representation of the interac-tion of consumers and producers. The simplest form of general equilibrium model is the input-output model pioneered by Leontief (1941). In the static input-output model, there is no jointproduction, only one technique exists for producing each output, and all technologies haveconstant returns to scale. Input requirements for each unit of output are given by fixed coeffi-cients, and final demand is exogenous. The appeal of this approach is its conceptual simplic-ity and the tractability afforded by computing equilibrium prices by matrix inversion. Thescheme of using matrices to keep track of flows between sectors persists to this day withinmore complex models of general equilibrium. Isard and Kaniss (1973) give a good account ofthe uses and shortcomings of the input-output model.

Activity analysis generalizes the production structure by representing it in terms of alter-native activities, that is, combinations of inputs and outputs where the ratios between inputsand outputs are fixed in each instance but vary between activities. Joint production is permit-ted in activity analysis, and there may be more than one activity producing the same output

12

MODELING INTERACTIONS BETWEEN THE ENVIRONMENT AND THE ECONOMY 13

(Koopmans 1951; Dorfman, Samuelson,and Solow 1958). Within the linear pro-gramming environment, prices are assumedexogenous, multiple consumers are not per-mitted, and the model contains no price dis-tortions. Under these conditions it could beproved that shadow prices coincided withmarket prices

CGE modeling originated with the workof Johansen (1960). He was the first to in-troduce a feedback from production levelsand endogenous prices to final demand. Jo-hansen solved the general equilibriummodel for growth rates by linearizing themodel in logarithms and applying matrixinversion techniques. He introduced nonlin-ear neoclassical substitution possibilities inproduction and consumption and endoge-nous determination of market-clearingproduct and factor prices. The Johansen ap-proach was further developed by Dixon etal. (1982) in their multisectoral ORANImodel for the Australian economy. Darwinet al. (1995) and Hertel (1990, 1997) arealso in the same tradition.

A technique that is becoming widelyadopted is to recast equilibrium problems asmixed complemetarity problems (MCP).The MCP is a fundamental problem in opti-mization that encompasses many of thecontinuous optimization problems, such asquadratic programming and nonlinear pro-gramming, as special cases. It is useful forexpressing systems of nonlinear inequali-ties and equations. A common representa-tion of an MCP has two components: thefirst represents a set of underlying condi-tions defined by a system of nonlinearequations, and the second constitutes thecomplementarity conditions that are onlyapplied to some of the variables and func-tions. The problem can be specified as fol-lows: given a nonlinear function

let I and J be a partition over {1, 2,, n}such that

Where the perpendicular notation sig-nifies that, in addition to the stated inequal-ities, the equation xTJ FJ (x) = 0 also holds.For existence and uniqueness of the solu-tion to this problem, see Ferris and Kanzow(1998).

The connection between traditional op-timization techniques in economics and thiswider problem class was first made by Cot-tle and Dantzig (1970). A natural connec-tion was also the use of mathematical pro-gramming methods in partial equilibriummodels pioneered by Samuelson (1949).For a review of papers on the formulationand solution of computable equilibriumproblems such as MCP, see Manne 1985;Cottle, Pang, and Stone 1992; Ferris andPang 1997.

An area that has received wide attentionin the field of complementarity problemshas been the development of robust and ef-ficient algorithms for solving large-scaleapplications efficiently. Along with the re-search in the design of algorithms came thelinkage of these algorithms with program-ming model languages such as the GeneralAlgebraic Modeling System (GAMS). Theresearch results to be presented here havebeen obtained using the PATH solver (avail-able with GAMS), which uses a searchmethod that is a generalization of a linesearch technique (Dirkse and Ferris 1995).

Modeling Approaches

CGE models have been categorized fromanalytical through stylized to applied(Robinson 1989). Analytical and stylized: , find an n n nF x R R R

( ) 0, free, and

( ) 0 0J J

F x x

F x x

=

14 CHAPTER 3

numerical models explore the magnitude ofthe effects of particular causal mechanismsand usually do not provide sufficient detailto analyze and support specific policy rec-ommendations. Applied models consist of amore detailed specification of the institu-tional side of the country-specific economyunder study. Although applied modelsallow for detailed analysis, there is a dangerof concealing the basic causal mechanismsof the model without enhancing its empiri-cal significance, a fact that should be kept inmind when choosing detailed features foran applied model specification (Devarajan,Lewis, and Robinson 1994).

In the domain of the applied models, thedetailed nature of CGE models is driven byconcerns about policy objectives, externalshocks being imposed, and the policy tools

being considered to meet the objectives andface the exogenous shocks (Figure 3.1).The combination of these three factors de-termines the adequate geographic and sec-toral aggregations and indicates the appro-priate way of representing time. More im-portantly, the underlying theoretical para-digm will also be affected by these factors.

Although the core of CGE models isneoclassical microeconomic theory, com-bined with the multisectoral intermediateinput links adapted from input-output mod-els, modelers have had to abandon some ofthe strict neoclassical assumptions in orderto meet the imperfections of the actualeconomies under observation. Instead ofperfect competition with perfectly flexibleprices and free product and factor mobility,applied CGE models often incorporate

Figure 3.1 Factors affecting the appropriate structure of a CGE model

Policy objectives Welfare improvement

Avoiding environmental damageRegional developmentIntergenerational allocation

Model characteristics

Geographic aggregationTime dimensionFinancial . real economyMarket rigiditiesFactor, sector, and householdspecification

vs

External shocks

Exchange rate shockNatural disasters, such as droughtTrade sanctions

Policy levers

Structural adjustmentAgricultural development policiesTrade liberalizationRegional development policies

MODELING INTERACTIONS BETWEEN THE ENVIRONMENT AND THE ECONOMY 15

structural rigidities, which seek to capturenonneoclassical behavior, macro imbal-ances, and institutional rigidities typical ofdeveloping economies.9 The relevant theoretical features that describe macro ad-justment, political economy, uncertainty, in-complete markets, and temporary equilib-rium are not directly incorporated into themodels, but imposed through ad hoc con-straints, which are not directly related to theagents endogenous rational behavior.

Geographic Aggregation

CGE models can be divided into sub-national, single-country, and multicountrymodels. All are open-economy models andincorporate the rest of the world as an in-tegral component that permits the consider-ation of worldwide capital and commodityflows and consequently their influence onthe economy under observation. The analyt-ical focus of the study to be carried out de-termines the geographic aggregation to beapplied. Single-country models are used foranalyses with a single, national focus. Multi-country models are used to address ques-tions such as global trade liberalization, re-gional trade agreements, interregional mi-gration, and climate change issues.10

Although less common, the focus issometimes at a subnational level. In suchcases one can choose among a spectrum ofoptions for capturing the regionality insidethe country. If there are several economi-cally distinguishable regions to be fully rep-resented, a separate CGE model can be con-structed for each region connected by flowsof factors and commodities as in the multi-

country models (Robinson, Hoffman, andSubramanian 1994) .11 Lofgren and Robin-son (1999) present a spatially disaggregatednational CGE model that incorporates inter-regional and national-regional feedbacks toanalyze the spatial impacts of economicpolicies. On the other hand, if regionality isrelevant only to a subset of the economicprocess, such as the presence of a regionallyspecified activity or factor or both it may besufficient to maintain a national specifica-tion for the model as a whole, while distin-guishing the few relevant regional charac-teristics (Coxhead and Warr 1991; Coxheadand Jayasuriya 1994).

Another reason to model at the subna-tional level is that the interest is in a naturalresource base that is geographically de-fined. In such cases modeling the single re-gions, for example, a watershed, may be theappropriate solution (Mukherjee 1996).Isard et al. (1998) present a detailedoverview on applied general interregionalequilibrium models.

Specification of Time

If the focus of the analysis is comparativestatics, the appropriate approach is a single-period model in which all time flow is col-lapsed into the time before and after an ex-ogenous, unexpected shock. In this casetime plays a limited role, agents expecta-tions are assumed static, only impacts onflows are considered and not impacts onstocks, and the timeframe for adjustment isgenerally captured by the mobility of factormarkets expressed in the model closure.12While this approach may appear to be an

9These deviations from the Walrasian paradigm and their corresponding methodological problems are criticizedin Srinivasan 1982; Bell and Srinivasan 1984; and Shoven and Whalley 1984.10For surveys on this matter refer to Shoven and Whalley 1992; Brown 1992; Goldin, Knudsen, and van derMensbrugghe 1993; OECD 1990.11In this case care has to be taken because the different regions share a common exchange rate.12Usually in the short-term factors reflect limited intersectoral mobility in the labor markets and none in the cap-ital markets (1 year); in the medium-term labor has full mobility but capital is still fixed (24 years); finally, inthe long-term both factors are mobile (510 years).

16 CHAPTER 3

oversimplification, it is a useful indicator ofthe order of magnitude of the impact of ashock or policy measure over an approxi-mate timeframe. At the opposite extreme ofthe single-period model, there are perfectforesight, intertemporally specified CGEmodels. This type of model is appropriatewhen the main focus is the transition pathassociated with a shock. This interest mayarise from a concern with the distribution ofincome over generations, associated for ex-ample with an aging population, or from in-efficiencies that could arise from fluctua-tions in the tax burdens over time. In caseslike these, dynamic CGE and other modelsare best suited to compare the long-termgains of a policy and its short-term costs.

Between the two extremes of static andrational expectations models there is abroad spectrum of options. A deeper treat-ment of time in CGE models reflectsmainly on the stock-flow relationships andthe assumptions about agent behavior overtime. First, if a model is to be inter-temporal, an equation of motion has to bespecified to update the factor stocks forlabor through population growth, for capitalthrough investment, and for the natural re-source base through degradation/regenera-tion. Second, one must represent the agentsexpectations concerning prices and pro-jected incomes. The latter point can be dealtwith in a variety of ways ranging frombackward-looking expectations (which canbe solved recursively) to perfect foresightmodels (Dixon and Parmenter 1996). Therecursive approach is often considered the

appropriate choice for capturing the transi-tion path and, in fact, it is often used forforecasting purposes. There are two ap-proaches to macro forecasting in a CGEframework: the first option is to rely onCGE-generated macro implications, and thesecond is to rely on exogenously suppliedmacro forecasts, using the CGE model tocarry out structural forecasts (Dixon andParmenter 1996).13

The perfect foresight approach is ap-pealing for its model-consistent expecta-tions. Forward-looking models will gener-ally have four distinguishing characteris-tics. First, consumption is represented aspart of life-cycle behavior of consumers.Household behavior is determined by themaximization of an additively separable,time-invariant, intertemporal utility func-tion subject to a lifetime intertemporalbudget constraint. Households can be repre-sented as being constituted by overlappinggenerations or as infinitely lived agents.14Second, firms are assumed, first, to maxi-mize their market value, which is equal tothe present value of their dividend streams,and second, to face imperfect capital mobil-ity due to adjustment costs (q-theory).15Third, the government faces an intertempo-ral budget constraint, and if the governmentis allowed to run deficits, the debt path isendogenously determined (Pereira 1988;Pereira and Shoven 1988). Finally, the bal-ance of trade and international capital flowshave to be specified; not much has beendone in this area, and most models assumebalanced trade and no capital flows.16

13Forecasting with CGE-generated macro scenarios has not been very successful. When using an external macroforecast, compatibility with the CGE model is ensured by endogenizing variables like total factor productivityand the propensity to save (see Dixon and Parmenter 1996 for more on this matter).14Early work on the overlapping generations dynamic models was done by Auerbach and Kotlikoff (1983); Bal-lard (1983); Ballard and Goulder (1985); for the infinitely lived agent approach see Bovenberg 1985 and Ander-sson 1987.15Examples of such firm behavior specifications can be found in Bovenberg 1984; Summers 1985; Goulder andSummers 1987; and Devarajan and Go 1998.16Exceptions are Andersson 1987 and Erlich, Ginsburgh, and Van der Heyden 1987.

Infinite-horizon formulations face se-vere computational problems when used inapplied models. Another drawback of thistype of approach is that the baseline towhich the simulations will be compared is abalanced growth path (which may or maynot occur in reality). Finally, the discountfactor, which is generally specified exoge-nously, will often generate an unrealistic se-quence of savings rates (Ginsburg 1994). Agood compromise is to build a two-periodintertemporal model for a policy measure orshock that takes place during the first period(Erlich, Ginsburgh, and Van der Heyden1987; Persson 1995).

For an early survey on dynamic CGEmodels (concentrating on tax policy evalu-ation), see Pereira and Shoven (1988). Inthe final part of a book edited by Mercenierand Srinivasan (1994), four contributionsby different authors are concerned withmodeling intertemporal trade-offs. Azis(1997) compares the impacts of economicreform on rural-urban welfare in a static anda dynamic framework and thereby focusesnot only on the economic objectives of thestudy, but also on the differences of its re-sults with respect to the different method-ological approaches. In this vein, Abbink,Braber, and Cohen (1995) demonstrateunder what assumptions a simple staticCGE model can be extended to a dynamicCGE specification, and they apply both ver-sions simultaneously. Very few applicationsshow explicit interest in and specification ofintertemporal aspects of the developmentprocess, such as the multisectoral CGE withoverlapping generations and intertemporal

optimization presented by Keuschnigg andKohler (1995).17 Another example is Go(1995), who highlights the intertemporaltrade-offs of tariff reforms when examiningthe sensitivity of investment and growth toexternal shocks and adjustment policy. Dy-namic CGE models are very useful in orderto simulate the overall economic develop-ment path of an economy. Diao, Yeldan,and Roe (1998) construct a dynamic ap-plied general equilibrium model of a smallopen economy in order to investigate thetransition path and convergence speed ofout-of-steady state growth paths in responseto trade policy shocks.

Environmental Externalitiesand Natural Resource Use

Since the 1970s there have been numerousapplications of CGE modeling to energyand natural resource issues. Models relatingto energy range from those with highly dis-aggregated specifications of the energy sec-tor, allowing for substitution between en-ergy sources and specifying different de-mand types, to those focusing more on therest of the economy, containing a simplifiedrepresentation of the energy sector.18 Thelatter generally focus on the differential im-pact of a natural-resource boom or crisis onthe tradable and nontradable sides of theeconomy (Benjamin1996; Martin and vanWijnbergen 1986). As an example of theformer, Hudson and Jorgenson (1974) con-structed an econometric general equilib-rium model that captured the interrelation-ships between energy policies and

MODELING INTERACTIONS BETWEEN THE ENVIRONMENT AND THE ECONOMY 17

17Keuschnigg and Kohler (1995) analyze the dynamic effects of trade liberalization in Austria.18Surveys for the disaggregated approach are Devarajan 1988; Bergman 1988; and Bhattacharyya 1996.

economic growth. The authors examinedthe role of energy taxes in promoting con-servation and how to employ the price sys-tem to adapt to changes in the availability ofenergy resources.

The role of taxation to compensate forenvironmental externalities and its generalequilibrium effects are fertile topics forCGE analysis both because the societalcosts of such a tax can be estimated throughits effect on prices and income (positiveanalysis), and because optimal taxes may becomputed (normative analysis). Jorgensonand Wilcoxen (1990) examine the costs tothe economy of emissions regulation andthe implications of a carbon tax.19 For a pe-riod there was debate over the so-calleddouble-dividend hypothesis, postulatingthat if the revenue from emission charges isused to reduce the tax on wage income thenpositive employment effects can result insecond-best situations with preexistingdistortions (Terkla 1984). While this debatehas not been resolved, the hypothesis seemsto hold only in the short run and under re-strictive assumptions (Carraro, Galeotti,and Gallo 1996; Scholz 1998). An interest-ing development, as the theory of marketincentives evolved, was to include marketsfor tradable emission permits where theequilibrium prices of permits reflect themarginal costs of emission control(Bergman 1991). In reality, the problemwith this approach is that a tradable permitprogram, compared with taxation, has norevenue-raising mechanism to cover thehigh monitoring costs.20

Because of the local and global exter-nalities associated with tropical deforesta-tion, the results presented in the previousparagraphs are important in the context of

the research described in this report; how-ever, deforestation occurs mostly on pri-vately owned land. This implies that theeconomic agent owning the land will viewit as an input to production, either agricul-tural or for timber where externalities arenot taken into consideration, or maybe forconservation if externalities are fully inter-nalized. It is therefore important to under-stand how land as a factor of production isrepresented in CGE models.

Land is a heterogeneous factor in agri-cultural production and this poses interest-ing challenges and possibilities from amodeling standpoint. The productive possi-bilities of a given hectare of land depend onsoil type, drainage, declivity, and climate.These characteristics affect the yield for anyspecific crop given labor and capital inputs,and therefore determine (along with consid-erations of the other factors) the most suit-able economic activity on a parcel of land.A CGE model focusing on agriculture mustmanage to capture the constraints on supplyresponse arising from land heterogeneity.Perhaps the simplest method available is tosegment the land market along land typesthat can be put to similar uses. For example,rice and corn can be substituted in produc-tion if the land is good, but a producer can-not switch from mediocre pasture to pro-ducing rice or corn on that land. This ap-proach implies that activities are either per-fectly substitutable or not substitutable atall. A more flexible approach is that adoptedby Robidoux et al. (1989) who also differ-entiate between land types and land uses,but the land types substitute imperfectly inthe production of a given crop.21 In both approaches the land-specific rental ratemust be equal across uses. An alternative

18 CHAPTER 3

19See Bhattacharyya (1996) for a survey on the use of CGE for environmental policy analysis.20Revenues can be generated by auctioning off permits, but this one-time inflow will not cover monitoring costs.21The authors of this study on Canada specify constant elasticity of substitution (CES) aggregator functions thatcombine land types, each of which is used to some degree in each crop.

approach is that adopted by Hertel and Tsi-gas (1988); they specify a transformationfunction that takes aggregate farmland as aninput and employs it in various uses basedon the elasticity of transformation and rela-tive rental rates.

Unlike labor and capital, land is geo-graphically immobile. Regional or climaticdifferences can be expressed in a number ofways. If farmland is represented as an ag-gregate input as in Hertel and Tsigas (1988),regionality is difficult to incorporate unlessit is embedded in the crop specification. Toportray regionality appropriately, land typeshave to be differentiated along geographicor climatic lines as in Darwin et al. (1995).Land classes are then employed differen-tially across sectors according to currentpatterns of production.

This section concludes with anoverview of the use of CGE models to ana-lyze issues relating to forestry and defor-estation. Following Xie, Vincent, andPanayoutou (1996), CGE models dealingwith forest resources can be broadly classi-fied into three groups. The first group con-sists of applications of standard CGE mod-els that include a forestry sector alongsidethe other production sectors of the economy(Cruz and Repetto 1992; Coxhead andJayasuriya 1994; Coxhead and Shively1995). The second group considers the dy-namic nature of forests reaction to eco-nomic processes and resolves the intertem-poral forest harvesting problem by model-ing a steady state (Dee 1991; Thiele andWiebelt 1992; Wiebelt 1994; Thiele 1994).The steady-state specification assumes thatforesters choose an economically optimalharvest pattern. The limitation of this ap-proach for deforestation in tropical areassuch as Brazil is, first, that logging is closerto an extractive process, as opposed to asustainable, managed forest operation. Sec-ond, deforestation is driven mostly by landclearing for agricultural purposes. The thirdgroup of models differentiates land uses andtypes and introduces property rights consid-erations (Persson and Munasinghe 1995;

Persson 1995). They include logging andsquatter sectors and therefore markets forlogs and cleared land. The model adopted inthis paper extends the approach of Perssonand Munasinghe (1995) to include landdegradation as a feedback mechanism intothe deforestation process. A more in-depthreview of CGE model applications to defor-estation can be found in Kaimowitz and An-gelsen (1998).

In their comprehensive review of eco-nomic models of deforestation spanningtheoretical constructs and scales,Kaimowitz and Angelsen (1998) note somecommonality in findings-that ease of accessto forest and to long-distance trade paths aswell higher agricultural and timber prices orlower rural wages increase deforestationrates. However, problems at each scale ofanalysis contribute to what Kaimowitz andAngelsen highlight in their review as incon-clusive or ambiguous findings about the ef-fects on deforestation of macroeconomicforces, population and migration, changesin productivity and input markets (includ-ing land markets and tenure security), andhousehold wealthor poverty. Since thatreview, Barbier (2001) has collected papersanalyzing deforestation that emphasize eco-nomic modeling techniques or that incorpo-rate spatial features and institutional factors(including placement of parks and reserves).

CGE Model Structure: APrimer

In the standard approach to CGE models,one first distinguishes between differentagents, such as producers, consumers, andgovernment, and then between goods andfactors and the associated markets throughwhich agents interact. The behavioral as-sumptions of agents are rooted in conven-tional microeconomic theory: producersmaximize profits subject to certain technological constraints (nonincreasing-returns-to-scale production functions) whileconsumers maximize utility subject to

MODELING INTERACTIONS BETWEEN THE ENVIRONMENT AND THE ECONOMY 19

budget constraints, all within the frame-work of competitive markets. Equilibriumin this type of model is characterized by aset of prices and levels of production suchthat the market demand equals supply forall commodities. Factors are either fullyutilized with flexible market-clearingwages or rent, or alternatively, the wage ofa factor has a lower bound below whichthere is excess supply of that factor. The in-tersectoral allocation of factors is endoge-nously determined. The model is specifiedas a system of nonlinear simultaneous equa-tions. The basic elements of the model canbe represented by the circular flow diagramof the economy presented in Figure 3.2. Thestarting point for the development of thismodel is a standard CGE model as de-scribed in Dervis, de Melo, and Robinson(1982), and the structure of the modeldraws most directly on Robinson, Kilkenny,and Hanson (1990) and Robinson (1990).

Factor incomes generated by productionactivities are divided among households infactor-specific shares representing factorownership. Total household income is usedto pay taxes, save, and consume. Govern-ment revenue comes from the collection ofad valorem direct taxes and indirect taxes.Government transfers income to house-holds, and expenditure is a fixed share oftotal absorption. The rest of the world sup-plies imports and demands export goods.Brazil is treated like a small country inthe sense that the export demands and im-port supplies that it faces are infinitely elas-tic at prevailing prices (with the exceptionof coffee and sugar).

The macro system constraints (or macroclosures) determine the manner in whichthe accounts for the government, the rest ofthe world, and savings and investment arebrought into balance. On the spending sideof the savings-investment balance, nominal

20 CHAPTER 3

Figure 3.2 CGE structure showing the circular flow of income

Notes: RoW is rest of world, C is consumption, G is government, and I is investment.

Factor income Domestic savings

Suppliers Household Govt.

Factormarket

InvX

t

C G I

Intermediate

Goodsmarket

Domestic

Imports

RoW

Exports

ForeignSavings

aggregate investment is either a fixed shareof total absorption, or it adjusts according tothe households savings rate. On the savingsside, if investment is fixed, the averagehousehold saving rate adjusts to achieve thelevel of savings that matches the exoge-nously specified level of investment. In thegovernment account, total nominal govern-ment expenditure is a fixed share of totalabsorption, and government saving is en-dogenously determined by the model. For-eign savings is exogenous and the exchangerate adjusts the current account balance.

Model Characteristics

In the modeling approach adopted here, aregionalized CGE model is developed inwhich Brazil is subdivided into regionscompatible with the major administrativesubdivisions adopted by the Brazilian gov-ernment: Amazon, Northeast, Center-West,and South/Southeast.22 For the Amazon thefollowing processes are considered: (1)conversion of forest to cleared land (de-pends on agents economic decisions), (2)transformation of cleared land to grassland,and (3) subsequent transformation fromgrassland to unproductive states.23

The overall model has two components:the CGE model, representing the behaviorof economic agents, and the land transfor-mation model, which is a simplified repre-sentation of biophysical processes affectingland productivity.

The model allows for two-way trade(cross-hauling) assuming that imports anddomestic demand as well as exports and do-mestic supply are imperfect substitutes

(Armington assumption). Producers maxi-mize profits with respect to their nestedconstant elasticity of substitution (CES)production functions, and households max-imize utility with respect to Cobb-Douglashousehold consumption.24

The model is nonfinancial because itdoes not explicitly include money and assetmarkets. This choice is based on the as-sumption that the types of shock considered(changes in the nominal exchange rate,transportation costs, and agricultural tech-nologies) affect most directly the real sideof the economy, such as quantities of pro-duction and commodities consumed, ratherthan monetary effects, inflation, and interestrates. While the above hypothesis is some-what unrealistic in certain situations, thelack of data on the functioning of financialmarkets necessary to integrate supply anddemand variables for money and assets is alimiting factor in modeling financial inter-mediation of the savings and investmentprocess.25

The model is static and solves for a newequilibrium within a single period, given aspecified external shock, internal shock, orpolicy change. The previous section on dy-namic CGE models provides some insightinto the pros and cons of examining changeover time via CGE models used for differ-ent analytical purposes. The underlying mo-tivation for choosing a comparative staticsapproach is that the issues of interest heredo not depend on intertemporal optimiza-tion by agents, whether it be firms invest-ment behavior or households life-cyclesaving patterns. The scenarios to be ana-lyzed involve one-time shocks or policy

MODELING INTERACTIONS BETWEEN THE ENVIRONMENT AND THE ECONOMY 21

22For the definition of the Amazon region adopted in this report, refer to footnote #4.23The methods adopted could be used to study subsequent regeneration processes through secondary forestgrowth or planting improved pasture.24The reason for specifying consumption as being Cobb-Douglas is that the income shifts for most of the simu-lations are sufficiently small that a unitary income elasticity for the components of final demand will not affectthe outcome of the simulations for the variables in this report.25See Bourguignon, Branson, and de Melo (1992) for an example of the integration of asset portfolio behaviorof macroeconomic models in Tobins tradition into a CGE model.

measures to which the structure of the econ-omy must adjust in order to return to equi-librium. In terms of expectation models, theshock is a surprise, requiring adjustmentsto reestablish the macro balance of theeconomy.

A complete CGE model also includes anumber of closure rules. Closure rules placeaggregate constraints on the economic ac-tivity simulated in the CGE model. Theypertain to how the major macroeconomicaccounts (government, trade, labor and cap-ital accounts) adjust to regain equilibrium inresponse to changes in economic activity.When specifying the model, the system willbe overdetermined and one of the con-straints of the model must be relaxed to finda solution. Choosing a particular closurerule means precisely deciding which con-straint should be dropped. There is no clear-cut theoretical justification for the choice ofa particular closure rule except the mod-elers general view of an underlying macro-economic behavior that is assumed exoge-nous to the CGE model. The closure ruleshave been shown to have a considerable im-pact on model structure and the policy con-clusions reached (Lysy 1982; Dewatripontand Michel 1987; Robinson 1991). Themacroeconomic closure rules of the modeland the specification of its factor markets(presented in detail in a later section) willdetermine the short-, medium-, or long-term character of the model.

The present approach incorporates anumber of distinctive model features inorder to capture the mechanisms underlyingdeforestation and agricultural developmentin a complex setting like Brazil. First, theresearch is centered on the role of land as afactor of production; therefore, differentland classes, with distinct productive possi-

bilities, are specified based on geographiclocation and vegetative cover. Land in eachregion is differentiated according to its landtype on the basis of cover: (1) forested land,(2) arable land, (3) grassland/pasture, and(4