Does Credit Affect Deforestation? Evidence from a Rural Credit

Does Credit Aect Deforestation?

Evidence from a Rural Credit Policy in the

Brazilian Amazon

CPI Technical Report

Climate Policy Initiative

Juliano Assunção

Clarissa Gandour

Romero Rocha

Rudi Rocha

January 2013

About CPI

Climate Policy Initiative (CPI) is a policy eectiveness analysis and advisory organization whose

mission is to assess, diagnose, and support the eorts of key governments around the world to

achieve low-carbon growth.

CPI is headquartered in San Francisco and has oces around the world, which are aliated

with distinguished research institutions. Oces include: CPI Beijing aliated with the School

of Public Policy and Management at Tsinghua University; CPI Berlin; CPI Hyderabad, aliated

with the Indian School of Business; CPI Rio, aliated with Pontiﬁcal Catholic University of

Rio (PUC-Rio); and CPI Venice, aliated with Fondazione Eni Enrico Mattei (FEEM). CPI is an

independent, not-for-proﬁt organization that receives long-term funding from George Soros.

Descriptors

Sector Forestry and Land Use

Region Brazil

Keywords credit, land use, deforestation, conservation policies

Contact Juliano Assunção, CPI Rio

[email protected]

JEL codes Q23, Q24, Q28

purposes, such as policy discussions or educational activities, under a Creative

Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.

For commercial use, please contact admin@cpisf.org.

Author Aliations

J. Assunção, C. Gandour, Romero Rocha: Núcleo de Avaliação de Políticas Climáticas da

Pontifícia Universidade Católica do Rio de Janeiro (NAPC/PUC-Rio) & Climate Policy

Initiative Rio de Janeiro (CPI Rio)

Rudi Rocha: Universidade Federal do Rio de Janeiro (UFRJ)

Abstract

The concession of rural credit in the Brazilian Amazon became conditional upon stricter legal and

environmental requirements in 2008. This paper uses this policy change to investigate credit

constraints in the region, and how the variation in credit impacts deforestation. Dierences-in-

dierences estimation based on a panel of municipalities shows that the policy change led to a

reduction in rural credit, particularly for cattle ranching. The eect is concentrated on medium and

large loans. A two-stage estimation shows that the decrease in rural credit has curbed deforesta-

tion, especially in municipalities where cattle ranching is the main economic activity.

1. Introduction

How does the availability of rural credit aﬀect defo restati on ? Theory alone oﬀers

ambiguous answers to this question. On the one hand, should credit be used to increase

rural production by incorpora ting new lands for production, g reater availability of credit

will likely lead to rising deforesta tion, as forest areas are cleared and converted into

agricultural land. On the other hand, should it be used to fund the capital expenditures

required to improve agricultural technology and productivity, increases in availability of

rural credit may actually reduce land use and curb deforestation.

This paper shows that a reduction in rural credit has contributed to curb deforestatio n

in the Brazilian Amazon Biome. The evidence is provided by the evaluation of Resolu-

tion 3,545, introduced in 2008 by the Brazilian National Monetary Council (Conselho

Monetário Nacional, CMN). This resolution conditioned the concession of rural credit in

the Amazon Biome upon proof of compliance with legal and environmental regulations.

We investigate several aspects of b oth the implementation and impact of this novel credit

policy, and explore variations in rural credit concessions caused by this new regulation to

estimate the eﬀect of credit on deforestation.

We interpret the impact of credit on deforestation as evidence of credit constraints in

the region. Based on Banerjee and Duﬂo (2012), we argue that this potential rationing

in the availability of subsidized credit may have tightened credit constraints, leading to

changes in farmers’ production decisions and thereby aﬀecting deforestation. We develop

a very stylized econo mic model, which guides the interpretation of our results, showing

how a change in the availability of subsidized credit may either increase o r decrease forest

clearings.

Our results have two key policy implications. First, the evidence shows that condi-

tional rural credit can be an eﬀective policy instrument to combat deforestation. Second,

and perhaps more important, our analysis suggests that the ﬁnancial environment in the

Amazon is characterized by signiﬁcant credit constraints. Thus, policies that increase the

availability of ﬁnancia l resources may potentially lead to higher deforestation rates. This

issue lies at the core of the recent debate about REDD policies that ultimately invo lve

payments for environmenta l services (see Alston and Andersson (2011), Angelsen (201 0 ),

and Angelsen and Wertz-Kanounnikoﬀ (2008)).

Rural credit is an important source of funding for Brazilian agriculture, as well as

akeypolicyinstrumentinthecountry. Ruralcreditissubsidizedbythegovernment.

When Resolution 3,545 was implemented in 2008, the rural credit portfolio proposed by

the federal government and carried out by oﬃcial banks and credit cooperatives totaled

BRL 78 billion (USD 38.5 billion) for all of Brazil. In that same year, the Amazon Biome

portfolio provided BRL 2.5 billion (USD 1.24 billion) in rural credit. In this region,

rural credit is distributed mostly through government banks, such as Banco do Brasil,

Banco da Amazônia and Banco do Nordeste. According to MAPA (2003), the oﬃcial

rural credit portfolio covers about a third of the annual ﬁnancia l needs of the agricultural

sector in Brazil. As part of these needs are covered by producers’ own resources, this

portfolio represents an even larger share of external ﬁnance in the sector. Thus, any policy

measure that aﬀects rural credit also aﬀects one of Brazil’s main support mechanisms for

agricultural production.

Three key aspects in the implementation of Resolution 3,545 helped us design our

empirical analysis. First, its conditions applied only to rural credit used in landhold-

ings inside the Amazon Biome. Having access to b oth contract-level credit concession

microdata and deforestation satellite data for all of the Legal Amazon (a larger admin-

istrative region), we are able to use Legal Amazon municipalities that are not in the

Amazon Biome as a control group to evaluate the impact of the resolution in the biome.

This aspect allows us to use a diﬀerence-in-diﬀerences approach for the evaluation of the

resolution.

Second, credit takers who were beneﬁciaries of the National Program for the Strength-

ening of Family Agriculture (Programa Nacional de Fortalecimento da Agricultura Fa-

miliar, Pronaf) were subject to far less stringent conditions for credit concession. Pronaf

is designed to target sma ll-sca le producers, where eligibility for the program depends on

the producer’s farm size. As our data allows us to identify whether credit was contracted

through Pronaf or non-Pronaf channels, we explore how, for a given municipality, credit

concession diﬀered between small and mid-to-large-scale pro ducers after the adoption of

Resolution 3,545. We explore this second aspect in an alternative diﬀerence-in-diﬀerences

formulation.

Third, the resolution’s conditions were such that borrowers who proved that they

had the intention to comply with environmental regulation were allowed access to credit.

The key requirement was a formal commitment to adapt to the environmental rules over

time rather than at the current time. For practical purposes, this made intention to

comply equivalent to compliance, helping us validate our strategy for the estimation of

the relationship between credit and deforestation. A possible concern for our identiﬁca-

tion strategy co uld arise from ﬁnancially unconstrained farmers changing their choices

about deforestation to prevent future credit restrictions. The way Resolution 3 ,5 4 5 was

implemented mitigates the relevance of this issue.

Our analysis is based on a contra ct-level microdata set compiled by the Central Bank

from Common Registry of Rural Operations (Registro Comum de Operações Rurais,

Recor) data. This data set contains detailed information on all rural credit contracts

negotiated by oﬃcial banks (public and private) and credit cooperatives in the 2002

through 2011 period in the Legal Amazon states of Acre, Amazonas, Amapá, Ma ra nhã o ,

Mato Grosso, Pará, Rondônia, Roraima and Tocantins. We use this data to construct

apaneloftotalamountofruralcreditatthemunicipalitylevel. Wealsocollectmunic-

ipality level data on satellite-based deforestation from the National Institute for Space

Research’s (Instituto Nacional de Pesquisas Espaciais, INPE) Project for Monitoring

Deforestation in the Legal Amazon (Projeto de Monitoramento do Desﬂorestamento na

Amazônia Le gal, PRODES/INPE).

Our main diﬀerence-in-diﬀerences results indicate that Resolution 3,545 led to a reduc-

tion in the concessio n of rural credit in the Amazon Biome. In counterfactual simulations,

we estimate that approximately BRL 2.9 billion (USD 1.4 billion) were not loaned in the

2008 through 2011 period due to restrictions imposed by Resolution 3,545. A reduc-

tion in loans speciﬁc to cattle ranching activities accounted for 90% of this eﬀect. The

resolution also led to a decrease in the concession of non-Pronaf credit, as compared to

Pronaf credit, within a given municipality. This is to be expected in light of the legal

exemptions that were introduced for small-scale producers. Counterfactual simulations

conducted using the Pronaf vs non-Pronaf speciﬁcations yield results similar to those

obtained using the biome vs non-biome speciﬁcations, which is reassuring in terms of

support for our identiﬁcation strategy. Several robustness checks corroborate the use of

our main speciﬁcations.

We investigate p otential heterogeneous eﬀects of Resolution 3,545 by exploring speciﬁc

elements of municipality and contract-level heterogeneity. First, we test if the policy’s

impact diﬀered among municipalities with diﬀerent leading economic activities. Munici-

palities are categorized as being either cattle or crop-oriented. Our ﬁndings suggest that

the resolution had a signiﬁ cant negative impact on rural credit concession in both types

of municipalities, with a stronger eﬀect on cattle-oriented ones.

Second, we analyze how Resolution 3,545 aﬀected the size and the composition of

rural credit contracts in the Amazon Biome. Results show that the policy had a distribu-

tional eﬀect on ca ttle-speciﬁc contracts. The number of medium and large cattle-speciﬁc

contracts, as well as the number of medium crop-speciﬁc contracts decreased, while the

number of small cattle-speciﬁc contracts slightly increased. W e document no impact on

small crop-speciﬁc contra cts. This eﬀect is likely the consequence of credit agents striving

to reallocate resources away from credit takers subject to the resolutio n’s conditions and

towards those facing less stringent legal restrictions.

Having shown that Resolution 3,545 caused a reduction in rural credit concession in

the Amazon Biome, we move on to test whether this reduction aﬀected deforestation.

Results from instrumental variable regressions indicate that the decrease in rural credit

helped contain deforestation in the biome. In counterfactual simulations, we estimate that

over 2,700 square kilometers of forest would have been cleared from 2009 through 2011

in the absence of the resolution-induced credit constraint. Considering that deforestation

rates in the late 2000s and early 2010s were between 5,00 0 and 7, 00 0 square kilo meters

per year, the eﬀect attributed to the resolution is quite substantial. We also ﬁnd that

the relationship between rural credit and deforestation varies according to the regionally

leading economic activity, with municipalities where cattle ranching predominates over

crop farming showing a stronger impact of credit on deforestation. Overall, the results

suggest that there are relevant credit constraints aﬀecting the deforestation activity in the

Brazilian Amazon Biome, and that the expansion of agriculture at the extensive margin

in the biome - particularly in cattle-oriented municipalities - is ﬁnancially constrained.

Again, we run several robustness checks that corroborate our results.

The remainder of the paper is structured as follows. Next section presents a short

literature review. Section 2 presents an overview of related literature. Section 3 describes

the institutional co ntext of rural credit and Resolution 3,545 at the time of policy im-

plementation. Section 4 introduces the model that guides our empirical analysis and

deﬁnes the role of credit constraints within our framework. Section 5 describes the data

and presents stylized facts and trends on credit concession and deforestation. Section 6

details the empirical strategy used to calculate the impact of Resolution 3,545 on credit

and deforestation. Section 7 discusses results on policy eﬀectiveness in restricting credit.

Section 8 discusses results on the relationship between credit and deforestation. Section

9 clo ses with ﬁnal remarks.

2. Related literature

The literature analyzing the direct link between credit and deforestation is scarce.

Binswanger (1991) discusses whether there exists a positive relationship between credit

and deforestation. He argues that subsidized rural credit tends to increase the demand

for land, leading to a more rapid expansion of crop farmland and pasture, but does not

formally test his hypotheses. Yet, rural credit policies could also encourage producers

to increase productivity within a given area of land. Pfaﬀ (1999) empirically studies the

relationship between a proxy for credit supply and deforestation in the Brazilian Amazon.

He ﬁnds that the number of bank branches is positively correlated with deforestation,

thought this result is not signiﬁcant in all speciﬁcations. Hargrave and Kis-Katos (2010)

also ﬁnd positive correlation between credit and deforestation in some regressions, and no

such impact in others (including their preferred speciﬁcation). Unlike previous studies,

we explore a policy-induced and exogeno us source of variation in rural credit. By using

the policy change as an instrumental variable for credit supply, we are able to identify a

causal relationship between credit and deforestation. We ﬁnd a positive and robust eﬀect

of rural credit on deforestation.

Our results also provide a better understanding of the determinants of the recent

Brazilian Amazon deforestatio n slowdown. After peaking at over 27,000 square kilome-

ters per year in 2004, the deforestation rate in the Brazilian Legal Amazon decreased

substantially in the second half of the 2000s to about 5,000 square kilometers in 2011

(INPE (2012)). The Amazon forest is the planet’s largest rainforest tract, as well as the

world’s most active agricultural frontier in terms of forest loss and CO

emissions (FAO

(2006), Morton et al. (2 0 0 6 ), Santilli et al. (2005)).

There is a substantial stream of literature documenting the impact of long-run so-

cioeconomic drivers of deforestation activity in the Amazon (for instance, previous works

investigate the eﬀect of population, road density, and agroclimatic characteristics on de-

forestation in the Amazon - see Chomitz and Thomas (2003), Reis and Guzmán (1994),

Reis and Margulis (1991)). However, there is scarce empirical evidence on the imme-

diate drivers of the recent and sharp deforestation slowdown in the region. This paper

complements the ﬁndings of Assunção et al. (2011). The authors show that, even when

controlling for commodity prices and relevant ﬁxed eﬀects, conservation policies intro-

duced starting in 2004 and 2008 helped avoid half of the forest clearings that would have

been observed from 2002 through 2009 should the policies not have been adopted. Our

paper isolates the credit channel and speciﬁcally tests whether credit constra ints have

been eﬀective in curbing deforestation vis-a-vis other recent conservation eﬀorts adopted

in the Brazilian Amazon.

Finally, our results contribute to a broader literature on rural credit. Previous studies

have found beneﬁcial eﬀects of the availability of credit in rural contexts. Credit sup-

ply ha s been positively associated with poverty reduction (Burgess and Pande (2005)),

agricultural investment and consumption smoothing (Rosenzweig and Wolpin (1993),

Conning and Udry (2007)). In this paper we unfold a potential negative externality by

documenting that the ava ila bility of rural credit may also lead to higher deforestation.

3. Institutional Context

Rural credit is one of Brazil’s most traditional ways to support agriculture (MAPA

(2003)). It encompasses signiﬁcant government subsidy. The Ministry of Agriculture

(Ministério da Agricultura, Pecuária e Abastecimento , MAPA) estimates that approxi-

mately 30% of the resources needed in a typical harvest year a re funded by rural credit

(MAPA (2003)). The remaining 70% come from producers’ own resources, as well as

from other agents of agribusiness (such as trading companies) and other market mech-

anisms (such as legal instruments that enable the pre-sale of production). In light of

this, any policy measure that aﬀects rural credit also aﬀects one of Brazil’s main support

mechanisms for agricultural production.

In Brazil, rural credit is loaned according to rules and conditions established in the

Central Bank’s Manual of Rural Credit (Manual de Crédito Rural, MCR). It is used to

ﬁnance short-term operating funds, investment, and commercialization o f rural produc-

tion.

All agents comprising the National System of Rural Credit (Sistema Nacional

de Crédito Rural, SNCR), which encompasses public banks, private banks, and credit

cooperatives, must abide by the MCR’s norms.

One of the key policy instruments for determining the concession of rural credit in

Brazil is the annual Crop and Livestock Plan (Plano Agrícola e Pecuário, PAP), com-

monly known as the Harvest Plan (Plano Safra). Published in the ﬁrst semester of each

year by the MAPA, the PAP summarizes the directives established for agricultural policy

in each harvest year.

The document is intended to serve as guidance for producers,

informing them about elements central to agricultural production in the country. At the

PAP’s core are the government’s plans regarding amounts of credit to be loaned to b o th

commercial and family production

,aswellasthesubsidizedinterestratesthatapplyto

alargefractionofplannedcreditloans.

From a total of BRL 14.7 billion (USD 7.3 billion) in the 2001/2002 harvest year, the

amount of credit planned for rural commercial production in the PAP increased to BRL

50.0 billion (USD 24.7 billion) in 2006/2007 and reached BRL 102.7 billion (USD 50.7

billion) in 2011/2012. A large share - typica lly over half - of these resources were loaned

under ﬁxed per year interest rates: 8.75% up to and throug hout the 2006/2007 harvest

year, and 6.75% thereafter (see MAPA (2001), MAPA (2006), and MAPA (2011)).

Considering the annua l SELIC

rate of over 18% in the beginning of the 2000s and over

8% in the beginning of the 2010s, the interest rates established in the PAP represented a

very signiﬁcant government subsidy for rural credit. Family production, whose planned

amount of credit increased from BRL 2.5 billion (USD 1.2 billion) in the 2001/2002 PAP

to BRL 10.0 and 16.0 billion (USD 4.9 and 7.9 billion) in the 2006/2007 and 201 1 /2012

PAPs respectively, received even greater subsidies, having access to annual interest rates

as low as 1% for speciﬁc categories of production.

3.1. Conservation Policy Eﬀorts and Central Bank Resolution 3,545

Brazilian conservation policies underwent profound revision in the 2000s. Most of

these changes occurred within the framework of the Action Plan for the Prevention and

According to MAP A’s classiﬁcation system, short-term operating funds credit is meant to cover the

usual expenses of production cycles; investment credit is to be applied to durable goods or services

whose beneﬁts last over an extended period of time; and commercialization credit is intended f or use

in ensuring supply and allowing for storage during periods of falling agricultural output prices. As, in

practice, some activities may fall into multiple categories, we restrict our attention to aggregate credit.

In Brazil, a harvest year is the period covering July of a current year through June of the following

year.

A family producer is described in Law 11,326/2006 as one who practices rural activities while simultane-

ously meeting the following criteria: (i) holds no more than four ﬁscal modules, where a ﬁscal module is

deﬁned as the minimum area needed in each municipality to ensure the economic viability of exploring

a rural establishment within that municipality; (ii) uses predominantly own family’s labor in establish-

ment’s economic activities; (iii) meets a minimum of family income originating from establishment’s

economic activities; and (iv) manages establishment alongside own family.

The SELIC rate is the Brazilian Central Bank’s overnight rate.

Control of Deforestation in the Legal Amazon (Plano de Ação para a Prevenção e o

Controle do Desmatamento na Amazônia Legal, PPCDAm). Launched in 2004, this plan

inaugurated a new form of dealing with deforestation. It integrated action across diﬀerent

government institutions, and introduced innovative procedures for monitoring, environ-

mental control, and territoria l management in the Legal Amazon (Ipea et al. (2011),

IPAM (2009)). Henceforth, policymakers sought to inhibit forest clearings and promote

forest conservation through three main policy eﬀorts: the strengthening of command and

control operations, the expansion of protected territory, and the adoption of more restric-

tive rural credit policies. The remainder of this section focuses on Resolution 3,54 5 , the

major rural credit policy change of the end of the 2000s.

Published by the CMN on February 29

,2008,Resolution3,545conditionedthe

concession of rural credit for use in agricultural activities in the Amazon Biome upon

presentation of proof of borrowers’ compliance with environmental legislation, as well as

of the legitimacy of their land claims and the regularity of their rural establishments. The

measure, aimed at restricting credit for those who infringed environmental regulations,

applied to all establishments in municipalities located entirely within the Amazon Biome.

Resolution 3,583, published on July 1

,2008,determinedthatinfrontiermunicipalities,

whose territory is only partially located within the biome, the conditions a pplied solely

to establishments located entirely or partially inside the biome. As the Amazon Biome

is contained within the Legal Amazon, all biome municipalities are necessarily located

in the Legal Amazon, but not all Legal Amazon municipalities are part of the Amazon

Biome (see Figure 1).

As all related institutions - public banks, private banks, and credit cooperatives - were

obligated to abide by the new rules, the resolution represented a potential restriction on

oﬃcial rural credit, and thereby on the fraction of it that is largely subsidized via lower

interest rates. However, other sources of ﬁna ncing for agricultural activity, such as traders

and supplier’s credit, suﬀered no such restriction.

As determined in Resolution 3,545, to prove eligibility for taking credit, the borrower

had to present: (i) the Certiﬁcate of Registry of the Rural Establishment (Certiﬁcado

de Cadastro de Imóvel Rural, CCIR); (ii) a declaration stating the absence of current

embargoes caused by economic use of illegally deforested areas; and (iii) a state-issued

document attesting the environmental reg ula rity of the establishment hosting the project

to be ﬁnanced, or, in the absence of such document, a state-issued certiﬁcate indicating

that the documentation necessary for regularization had been received. The resolution’s

requirements applied not only to landowners, but also to associates, sharecroppers and

tenants. Implementation of Resolution 3,545 terms by all credit agents was optiona l as

of May 1

,2008,andobligatoryasofJuly1

,2008.

Although seemingly restrictive at ﬁrst, the conditioning mea sures of Resolution 3,545

were subject to a series of qualiﬁcations that loosened the severity of the new credit

constraints. This was pa rticula rly relevant for small-scale producers. In its original text,

Resolution 3,545 already established exemptions for three groups of small credit takers.

The ﬁrst group, composed of Pronaf beneﬁciaries and rural producers operating in areas

smaller than or equal to four ﬁscal modules

, was still required to present the CCIR, but

could replace the rest of the documentation by an individual declaration attesting the

existence of the required legal reserve and area of permanent protection, and the absence

of current embargoes caused by the economic use of illegally deforested areas within the

credit taker’s establishment.

The second group, encompassing beneﬁciaries of the National Program of Land Re-

form (Programa Nacional de Reforma Agrária, PNRA) who ﬁt into Pronaf groups A

and A/C

,couldreplacealldocumentationbyadeclarationissuedbytheNationalIn-

stitute of Colonization and Land Reform (Instituto Nacional de Colonização e Reforma

Agrária,Incra)eitherattestingthelandreformsettlement’senvironmentalregularity,or

indicating that a term on the regularizatio n of the settlement had been agreed upon. An

annex with a list of all land reform settlement beneﬁciaries had to be included in the

documentation.

The third group, restricted to family producers ﬁtting into Pronaf group B

,was

not required to present any documentation. As the three groups referred to small-scale

producers, these exemptions implied that Resolution 3,545 established less restrictive

conditions for small producers.

Soon after the compulsory adoption of the resolution, new CMN measures further

loosened the requirements for the co ncessio n of rural credit to small producers. With the

passing of Resolution 3,599 on August 2 9

,2008,borrowersfromtheﬁrstgroupcould

The actual size of a ﬁscal module varies according to municipality. It depends mainly on the conditions of

pro duction in each municipality including, but not limited to, market dynamics, installed infrastructure,

technologic al availability, and natural features such as soil quality and water availability.

Pronaf Group A refers to family producers settled into PNRA land reform settlements, as well as

to beneﬁciaries of the National Program of Land Credit (Programa Nacional de Crédito Fundiário,

PNCF) who have not yet taken investment credit under the Program of Special Credit for Land Reform

(Programa de Crédito Especial para a Reforma Agrária,Procera)orthathavenotyetreachedthe

maximum credit limit for structural investment within Pronaf. The group also includes family producers

who were relocated due to the construction of dams for use in hydroelectric power generation or for the

provision of water to land reform settlements.

Pronaf Group A/C refers to family producers settled into PNRA land reform settlements, as well as to

beneﬁciaries of the PNCF who not only possess the Group A/C Declaration of Eligibility for Pronaf

(Declaração de Aptidão ao Programa, DAP), but who have also already contracted a ﬁrst op eration in

Group A, and have not obtained ﬁnancing for working capital in a Pronaf group other than A/C.

Pronaf Group B refers to family producers that meet all of following criteria: (i) explore a fraction of

land as owner, land reform settler, tenant or associate; (ii) live in or near the property; (iii) have access

to an area no greater than four ﬁscal modules; (iv) derive at least 30% of family income from both the

agricultural and non-agricultural exploration of the property; (v) base exploration of the prop erty on

family labor; and (vi) have total family income of up to BRL 6,000 (USD 2,962.20), excluding social

beneﬁts. This group is known as Pronaf’s rural microcredit group.

present a declaration attesting not the existence, but the recomposition or regeneration of

the required legal reserve and area of permanent protection within their establishments.

Similarly, instead of attesting the land refo rm settlement’s environmental regularity, the

Incra declaration presented by borrowers from the second group could attest that the land

reform settlement had an environmental license or that the process for obtaining such

license had already been started. Resolution 3,599 and Resolution 3,61 8 from September

, 2008, also included the fo llowing in the list of Pronaf beneﬁciaries exempted from

presenting any documentation: indigenous people, quilombolas,

,small-scaleﬁshermen,

inhabitants or users of conservation units of sustainable use, and reg ula r inhabitants of

river margins.

Small producers also beneﬁtted from several exceptions introduced for speciﬁc harvest

years. The CMN determined that the CCIR could be substituted either by the DA P (a

document held strictly by eligible Pronaf beneﬁciaries) for program beneﬁciaries, or by a

CCIR request for all other rural producers operating in areas no greater than four ﬁscal

modules. Resolution 3,618 introduced this new rule for the 2008/2009 harvest year, and

Resolution 3,735, published on June 17

,2009,extendedittothe2009/2010harvest

year.

From an institutional perspective, Resolution 3,545 established the same restrictions

and exceptions in all municipalities to which it applied and, thus, represented a homoge-

neous policy change across regions. Even so, its impact on rural credit concession and,

consequently, on deforestation may have diﬀered across economic sectors due to struc-

tural heterogeneity. A key structural diﬀerence we take into account is the composition

of sources used to meet ﬁnancial requirements for crop and cattle production. According

to FAO (2007), the relative participation of oﬃcial rural credit contracts has decreased,

as agricultural ﬁnancing, particularly for soybean production, has increasingly been ob-

tained through contracts with traders, input and processing industries, and retailers and

market operators. Government credit meets an estimated 30% of the ﬁnancial require-

ments of the soybean production sector in Brazil, but the remaining funds are provided

by traders and the processing industry (40%), the input industry (15%), and farmers’

own resources (10%), with the remaining 5% being attributed to other sources, such as

manufacturers of agricultural machinery (FAO (2007)). A crop farming sector that is not

heavily dependent on oﬃcial rural credit, as appears to be the case in Brazil, could com-

pensate the decrease in oﬃcial rural credit imposed by Resolution 3,545 with alternative

sources of ﬁnancing. Producers operating in this sector would thus be able to sustain

investment and deforesta tio n at the same levels as before the credit policy intervention.

Crop production in Brazil also experienced relevant technological advances starting in

the early 1990s, particularly with the widespread adoption of direct seeding (FAO (2007)).

Quilombolas are inhabitants of quilombos settlements traditionally founded by escaped slaves.

No such pattern was observed for livestock farming, which remains a low-produ ctivity

practice in the country. Cro p fa rmers likely invest a larger share of rural credit loans

in the intensiﬁcation of production, instead of expanding production by operating in the

extensive margin as cattle ranchers do. In this case, a decrease in rural credit for crop

farmers might not lead to a decrease in forest clearings, since resources were not originally

being used to push agricultural production into forest areas. We examine these potential

sectoral diﬀerences.

Resolution 3,545 was originally proposed as a means to restrict credit to those who did

not abide by environmental regulations in the Amazon Biome. However, the resolution

was ﬂexible for small producers already a t its outset, and became increasingly so in a

relatively short period of time. Additionally, altho ugh the resolution constituted an insti-

tutional change at the federal level, regional heterogeneities may have inﬂuenced the way

in which the resolution impacted local access to credit and, thus, aﬀected deforestation.

It is therefore likely that Resolution 3,545’s eﬀects diﬀered among diﬀerent segments of

producers and regions. We explore this idea in the empirical analysis described in the

remainder o f this paper.

4. Model

This section presents the theoretical model that guides our empirical analysis. The

model, inspired by Banerjee and Duﬂo (2012), focuses on how credit constraints can inﬂu-

ence deforestation when diﬀerent production technologies are available to the producer.

Suppose a farmer operates in a forest a rea and chooses one among two agricultural

production technologies - traditional or modern. With the traditional technology, the

farmer produces agricultural output using labor and land inputs. This traditional tech-

nology is described by:

f(L, T ) (1)

where L is labor employed and T is area used for production. With the modern technol-

ogy, in addition to labor and land, the farmer also uses other inputs, K,suchastractors

and fertilizers. This modern technology is described by:

F (K, L, T )=A(K)f(L, T ) (2)

Assume that labor can be paid at the end of the harvest period, but that expenditures

with non-labor inputs must be paid in advance. Taking M as total working capital

av ailable to the farmer, working capital constraints are given by p

T ≤ M and p

K +

T ≤ M for the traditional and modern technologies, respectively. These constraints

allow for the possibility of existing binding credit ﬁnancing as in Feder (1985) and Udry

(2010). A farmer using the traditional technology therefore faces the following decision

problem:

traditional

(M)=max

L,T

f(L, T ) − p

L − p

T (3)

subject to p

T ≤ M

Similarly, the decision problem for a farmer using the modern technolo gy can be described

as:

modern

(M)=max

K,L,T

A(K)f (L, T ) − p

K − p

L − p

T (4)

subject to p

K + p

T ≤ M

Thus, a farmer with available working capital M chooses the modern technology if, and

only if, π

modern

(M) ≥ π

traditional

(M).DeﬁneM

such that π

modern

)=π

traditional

We assume that p

and A(K) are such that all farmers with M ≥ M

choose the modern

technology. In summary:

π (M)=



traditional

(M)ifM<M

modern

(M)ifM ≥ M

(5)

Within this framework, with the farmer operating in a forest area, the choice of area

to be used for productio n is equivalent to deforestation. We are therefore pa rticul arly

interested in how optimal farmland size is aﬀected by the availability of capital when the

farmer is allowed a choice of production technology.

To simplify the analysis, we consider speciﬁc functional fo rms for the production

functions, assuming that f (L, T )=L

and A(K)=K

, where α > 0, β > 0, γ > 0

and α + β + γ < 1. The assumption of decreasing returns to scale helps determine a ﬁnite

optimal farmland size. We focus on the characterization of the optimal land input. For

the traditio na l technology, the optimal choice of farmland is given by:

traditional

(M)=







, ifM<M

∗

traditional

ifM ≥ M

(6)

(7)

where T

∗

traditional

≡





1−β

1−β−γ





1−β−γ

and M = p

∗

traditional

.Forthemoderntech-

nology, the optimal choice of farmland is given by:

modern

(M)=







α + γ

ifM<M

∗

modern

ifM ≥ M

(8)

where T

∗

modern

≡





1−α−β−γ





1−α−β−γ





1−α−β

1−α−β−γ

and M = p

∗

modern

∗

modern

The relative values of M

M and M deﬁne diﬀerent possible cases. For example, a

conﬁguration such that M

< M<M implies in the optimal farmland size graph shown

in Figure 2.

Deﬁne M

∗

as the farm’s total investment if the farmer can borrow as much as he

wants at the interest rate r. Thus,

∗

(r)=argmax

Π(M) − (1 + r)M (9)

represents the ﬁrst-best investment level.

We assume that a typical farmer can be ﬁnanced by two diﬀerent sources and ignore,

for the sake of simplicity, the possibility of self-ﬁnancing. A subsidized rural credit line

is available at cost r

, which is b elow the market interest rate r

, r

.Denotingthe

amounts of subsidized rural credit and market credit as M

and M

,respectively,total

investment is given by M = M

+ M

. Following Banerjee and Duﬂo (2012), we say

that a farmer is credit rationed at the subsidized interest rate if M

∗

),andthat

afarmeriscredit constrained if M<M

∗

As argued in Section 3, Resolution 3 ,5 45 may have reduced the availability of sub-

sidized rural credit for some farmers in the Amazon Biome. Yet, the supply of credit

supplied at the market rate by agents other than oﬃcial banks (private and public) and

credit cooperatives was not directly aﬀected by the resolution. Our theoretical model

oﬀers intuition on how farmers are expected to react to this change in the supply of

credit, and thereby potentially aﬀect deforestation, under diﬀerent assumptions about

the availability of ﬁnancial reso u rces.

To restrict the ana lysis to a simple, yet interesting, situation, consider the case de-

picted in Figure 2, where M

< M<M .Otherconﬁgurationscanbeconsideredanal-

ogously. Start with the region where total investment lies below M .Increasesinthe

availability of resources within each technology region - (0,M

) or (M

, M) -aﬀect land

size positively. There being no change in the choice of production technology, a reduction

in credit leads to a decrease in optimal farmland size a nd thereby reduces deforestation.

However, changes in the availa bility of resources that cause farmers to switch between

technology regions - from (0,M

) to (M

, M) or vice-versa - have an ambiguous eﬀect on

land size. A reduction in credit may lead the farmer to substitute the modern technology

for the traditional one, potentially leading to an increase in optimal farmland size and

deforestation. In the region where total investment lies above M ,farmersarenotcredit

constrained, so changes within this region do not aﬀect o ptimal farmland size. Thus, a

reduction in M

that keeps the farmer in the unconstrained region does not aﬀect defor-

estation, but a reduction in the availability of resources that pushes that farmer into the

, M) interva l will reduce optimal farmland size and deforestation. An even stronger

reduction in the availability of resources that further pushes the farmer into the (0,M

)

interval has an ambiguous impact on deforestation. Propositions 1-3 summarize these

results in the context of the credit reduction implied by Resolution 3,545.

Proposition 1: If the reduction in the availability of subsidized rural credit causes a

reduction in deforestation, we can conclude that: (i) farmers are credit constrained; and

(ii) credit and deforestation have a positive relationship in the region.

Proposition 2: If the reduction in the availability of subsidized rural credit does

not aﬀect the amount of cleared land, we can conclude that: (i) either farmers are not

credit constrained (they could simply be substituting subsidized rural credit by market

credit); or (ii) farmers are credit constrained, but are changing from the modern to the

traditional technology.

Proposition 3: If the reduction in the availability of subsidized rural credit implies

an increase in deforestation, we can conclude that: (i) farmers are credit constrained; and

(ii) they are changing from the modern to the traditional technology.

In summary, a subsidized credit policy restriction can: (i) serve as evidence of credit

constraints if we observe an impact on deforestation; and (ii) reveal whether the rele-

vant margin is change in optimal farmland size within a given technology (decreasing

deforestation) or change across production technologies (increasing deforestation).

5. Data, Descriptive Statistics and Stylized Facts

This section introduces the data used in our empirical evaluation of Resolution 3,545,

presents descriptive statistics and discusses stylized facts to characterize aggregate trends

for our variables of interest.

5.1. Data on Rural Credit

Our analysis is based on two panels of municipality-level data covering the 2002

through 2011 period. The ﬁrst panel is constructed from a contract-level microdata

set of rural credit loan contracts compiled by the Central Bank from Recor data. This

is an administrative microdata set enco mpa ssin g all rural contract records negotiated by

oﬃcial banks - both public and private - and credit cooperatives in the Legal Amazon

states of Acre, Amazonas, Amapá, Maranhão, Mato Grosso, Pará, Rondônia, Roraima

and Tocantins.

It contains detailed information about each contract, such as the exact

day on which it was signed, its value in BRL, the contracted interest rate and maturation

date, its intended use by agricultural activity, and the category under which credit was

Only a fraction of the state of Maranhão is considered part of the Le gal Amazon. Our sample only

includes mu n ic ipalities from this fraction.

loaned (short-term operating funds, investment, or commercialization). The data set also

contains information about the oﬃcial source of funding for each contract, allowing us to

identify whether lo a ns ﬁt into Pronaf or non-Pronaf credit lines. All contracts are linked

to a code identifying the municipality in which the borrower’s landholding is located. We

add up the value of the contract loans across all days in each month and each municipality

to co nvert the microdata panel into a municipality-by-month panel.

Since Resolution 3,545 conditions applied to all esta blishments in municipalities lo-

cated entirely within the Amazon Biome, we merge our panel of municipality-by-month

credit loans data with a geocoded map containing information on the biome’s geographic

limits and municipalities’ location. We then identify, for each sample municipality,

whether it is entirely located inside the Amazon Biome. We also construct variables

indicating the shortest lin ear distance, in kilometers, between each municipality’s border

and the biome frontier. Using this variable, we can create subsamples of municipalities,

both inside and outside the Amazon Biome, located within sp eciﬁc distances to the biome

frontier.

To smoothen the large cross-sectional variation in values of credit contracts generated

by diﬀerent municipality sizes, we use a normalized measure of rural credit. This nor-

malization ensures that our analysis captures relative variations in credit lending within

municipalities. The variable is constructed according to the following expression:

Credit

− C

sd (C

)

(10)

where Credit

is the normalized amo unt of rural credit loaned in municipality i and

month-year t;thetermC

is the amount of rural credit loaned in municipality i and

month-year t in BRL; and the terms C

and sd (C

) are, respectively, the mean and the

standard deviation of the amount of rural credit loaned in municipality i over the 2002

through 2011 period. The variable C

replaces Credit

in robustness checks.

The ﬁnal data set containing information on rural credit, time, and geographic vari-

ables at the municipality-by-month level is used to evaluate the impact of Resolution

3,545 on rural credit loans. Our sample does not include municipalities crossed by the

biome frontier, since only those farmers whose landholdings are entirely located within

the biome in these municipalities are subject to the resolution’s conditions. The full sam-

ple is composed by 713 Legal Amazon municipalities, of which 458 are located inside the

Amazon Biome and 255 outside it. We use our constructed distance variable to create

a restricted subsample that o nly includes municipalities located within 100 kilometers of

the biome frontier. This subsample has 387 municipalities, of which 132 are inside the

Amazon Biome and 255 outside it. We construct a second restricted subsample composed

by municipalities where the pre-2 0 0 8 average value of annual credit loans for cattle ranch-

ing was higher than that for crop production. This "cattle-oriented" subsample co ntains

301 municipalities located inside the Amazon Biome and 154 outside it. Finally, we con-

struct an analogous restricted subsample of municipalities where the pre-2008 average

value of annual credit loans for crop production was higher than that for cattle ranching.

This "crop-oriented" subsample contains 133 located inside the Amazon Biome and 101

outside it.

5.2. Data on Deforestation

The second panel of data relates rura l credit loans to deforestation at the municipality

level. Data on deforestation is built from satellite-based images that are processed at the

municipality level and publicly released by PRODES/INPE. Because PRODES data is

reported annually, we must ﬁrst convert our municipality-by-month credit panel into a

municipality-by-year credit panel. We deﬁne deforestation as the annual deforestation

increment, that is, the area in square kilometers of forest cleared over the twelve months

leading up to August of a given year.

We recode credit loans accordingly, summing up

monthly data into an annual basis, where year t data sums information over the twelve

months leading up to August of t.

For any given municipality, cloud cover during the perio d of remote sensing may

compromise the accuracy of satellite images, requiring images to be produced at a diﬀerent

time. As a result, image records for diﬀerent years may span from less to more than twelve

months. To control for measurement error, variables indicating unobservable areas are

included in all regressions. This data is also publicly available at the municipality-by-year

level from PRODES/INPE.

To smoo then the cross-sectional variation in deforestation that arises from municipal-

ity size heterogeneity, we use a normalized measure of the annual deforestation increment.

The normalization ensures that our analysis considers relative variations in deforestation

increments within municipalities. The variable is constructed according to the following

expression:

Deforest

ADI

− ADI

sd (ADI

)

(11)

where Def or e st

is the normalized annual deforestation increment for municipality i and

year t; ADI

is the annual deforestation increment measured in municipality i between

the 1

of August of t − 1 and the 31

of July of t;andADI

and sd (ADI

) are,

respectively, the mean and the standard deviation of the annual deforestation increment

calculated for each i over the 2002 through 2011 period. The variable ADI

replaces

Deforest

in robustness checks. Our sample does not include municipalities that showed

More precisely, the annual deforestation increment of year t measures the area in square kilometers

deforested between the 1

of August of t − 1 and the 31

of July of t.

no variation in deforestation throughout the sample yea rs, as this variation is needed to

calculate the normalized variable.

The ﬁnal data set containing information on deforestation, rural credit, time, and

geographic variables at the municipality-by-year level is used to estimate the eﬀects of

the credit restriction on deforestation. Aga in, we do not include municipalities crossed

by the biome frontier. The full sample comprises 575 Legal Amazon municipalities, of

which 439 are located inside the Amazon Biome and 136 outside it. We also construct the

three restricted subsamples introduced in Section 5.1. Their composition is as follows:

251 municipalities within 10 0 kilometers of the Amazon Biome frontier (115 inside the

biome and 136 outside it); 406 cattle-oriented municipalities (301 inside the biome and

105 outside it); and 141 crop-oriented municipalities (110 inside the biome and 31 outside

it).

5.3. Agricultural Output Prices and the Demand for Credit

Agricultural prices a re endogenous to local agricultural production. Thus, to control

for ﬂuctuations in the demand for rural credit at the municipality level, we must construct

output price series that capture exogenous variations in the demand for agricultural com-

modities produced locally. As argued in Assunção et al. (2011), agricultural commodity

prices recorded in the southern Brazilian state of Paraná are highly correlated with av-

erage local crop prices calculated for the Legal Amazon sample municipalities. Hence,

we use the Paraná agricultura l commodity price series as exogenous indicators of lo ca l

market conditions within our empirical context. Prices for beef cattle, soybean, cassava ,

rice, corn, and sugarcane were collected at the Agriculture and Supply Secretariat of the

State of Paranáą (Secretaria de Agricultura e do Abastecimento do Estado do Paraná,

SEAB-PR). Soybean, cassava, rice, and corn are predominant crops in the Legal Amazon

in terms of harvested area. Although not a predominant crop in the region, sugarcane is

also included to take into consideration the recent expansion of Brazilian ethanol biofuel

production. Together, the ﬁve crops account for approximately 70% of total harvested

area averaged across sample years.

The Paraná price series are used to build two variables of interest. The ﬁrst of these

variables, an annual index of crop prices, is constructed in three steps. First, we calculate

nominal monthly price series for each calendar year-month and culture. Annual prices

are deﬂa ted to year 2011 BRL and are expressed as an index with base year 2011.

Second, we calculate a weighted real price for each of the crops according to the

following expression:

PPA

itc

= PP

∗ A

ic,2000−2001

(12)

where PPA

itc

is the weighted real price o f crop c in municipality i and year t; PP

is the

Paraná-based real price of crop c in year t expressed as an index with base year 2000; and

ic,2000−2001

is the share of municipal area used as farmland for production of crop c in

municipality i averaged over the 2000 through 2001 period.

This latter term captures

the relative importance of crop c within municipality i’s agricultural production in the

years immediately preceding the sample periods. It thus serves as a municipality-speciﬁc

weight that introduces cross-sectional variation in the commodity price series.

Third, we use principal component analysis on the weighted real crop prices to derive

the annual index of crop prices. This technique allows the price variations tha t are

common to the ﬁve selected crops to be represented in one single measure. The resulting

index of crop prices captures the ﬁrst principal component of the ﬁve weighted real

prices. As the index maximizes the price variance, it represents a more comprehensive

measure of the agricultural output price scenario for this analysis than the individual

prices themselves. Moreover, by using the index of crop prices, which absorbs b o th cross-

sectional and time-speciﬁc trends at the municipality level plausibly correlated with credit

demand, we overcome an important empirical limitation.

The second variable of interest is a n annual index of cattle prices, which is derived anal-

ogously to PPA

itc

.However,aslandpastureisnotobservable,inthiscaseA

ci,2000−2001

is the ratio of heads of cattle to municipal area in municipality i averaged over the 2000

through 2001 period.

5.4. Stylized Facts and Trends

This section presents stylized facts and trends for rural credit and deforestation over

the past ten years. In this set of descriptive results, the evolution of agg reg ate credit

is inﬂuenced by both economic conditions and changes in the regulatory and institu-

tional landscape. Tables and ﬁgures shown were built from the contract-level credit loan

microdata set and the deforestation data set.

Table 1 presents descriptive statistics for average annual amounts of rural credit and

deforestation at the municipality level. Figures for Amazon Biome municipalities before

and after the implementation of Resolution 3,545 are compared with those for Legal

Amazon municipalities outside the biome. The statistics point towards more accentuated

aggregate rural credit growth in the post-resolution period in Legal Amazon municipalities

located outside the Amazon Biome, where borrowers were not subject to the resolution’s

restrictions. Average Pronaf loans inside the biome also seem to have grown more than

non-Pronaf loans, as is to be exp ected in light of the exemptions introduced for small-scale

producers operating in the Amazon Biome. Average deforestation, which is signiﬁcantly

Variables on annual municipality crop production (harvested area, quantum,orvalueincurrentprices)

are based on data originally from the Municipal Crop Survey of the Brazilian Institute for Geog-

raphy and Statistics (Pesquisa Agrícola Municipal do Instituto Brasileiro de Geograﬁa e Estatística,

PAM/IBGE).

higher in Ama zo n Biome municipalities, dropped sharply both inside and outside the

biome.

Stylized facts also hint at potential eﬀects of Resolution 3,545. In Figure 3, average

rural credit loan trends for Amazon Biome municipalities do not seem to follow those for

its unrestricted counterpart outside the biome in the post-2008 period. The diﬀerence

in growth rates appears in both cattle and crop-oriented samples, suggesting that the

resolution imposed a binding credit constraint for rural producers within the biome.

Yet, as the pattern of credit loan trends diﬀer for cattle and crop-oriented samples, our

empirical analysis explores whether the resolution’s impact depends on economic activity

heterogeneity.

Acloserlookattheevolutionofruralcreditloansthroughout2008revealsthatthe

constraint seems to have been anticipated by credit takers. Figure 4 illustrates this

phenomenon. While credit concession in 2006 and 2007 concentrated in the second half

of each year, it followed a very diﬀerent pattern in 2008, peaking in April and again

in June. The total amount of credit negotiated in 2008, however, is similar to that of

previous years. Considering that the adoption of the measures established in Resolution

3,545 was optional as of May 1

and compulsory as of July 1

,theunseasonablepeaks

in 20 0 8 likely reﬂect borrowers’ eﬀorts to have early access to resources that would soon

be restricted. Consequently, an accentuated decrease in the concession of credit was not

observed in 2008 despite the restrictions imposed by Resolution 3,545. This behavio r is

more clea rly seen in the trends for cattle-speciﬁc contracts than for crop-speciﬁc ones,

perhaps due to the intrinsically seasonal component of crop production. Alternatively,

this could be because crop farmers were less vulnerable to the reduction in subsidized

rural credit. As they were able to access credit from other sources even after the resolution

had been adopted, they had less need to anticipate credit prior to the policy.

In addition to assessing the resolution’s impact on rural credit concession, we are

interested in understanding how it aﬀected deforestation. Figure 5 portrays the evolution

of average municipality-level deforestation. Deforestation dropped signiﬁcantly starting

in the mid-2000s in Lega l Amazon municipalities both inside and outside the Amazon

Biome. In the post-2008 period, however, trends for biome municipalities appear to

behave diﬀerently from those for non-biome municipalities - deforestation exhibits more

accentuated drops inside the biome immediately after the adoption of Resolution 3,545

both in the cattle and the crop-oriented subsamples. Although not yet conclusive, this

could be an indication both that deforestation is a credit-dependent activity and that

there are relevant credit constraints for forest clearing in the Amazon Biome.

The stylized facts discussed in this section provide a descriptive overview of how

Resolution 3,545 might have aﬀected the evolution of rural credit and deforestation in

the Amazon Biome. However, these facts capture the eﬀects of other inﬂuencing factors,

particularly those of eco no mic circumstances in the Amazon Biome during the period of

interest. To evaluate the impact of Resolution 3,545, we must isolate the eﬀect of the

resolution from that of other determinants of rural credit.

6. Empirical Strategy

This section describes the empirical strategy used to identify the causal eﬀect of rural

credit on deforestation. Since only equilibrium prices and quantities are observed in the

credit market, many of the identiﬁcation concerns mentioned in the literature have been

related to reverse causality and omitted variables. Reverse causality might take place if

the expansion of agricultural activities beyond the farmer’s landholding is done at the ex-

pense of public areas of forest. In this case, deforestation and appropriation of public land

increase real estate and collateralized asset-based lending, thereby allowing the farmer to

overcome borrowing constraints. Omitted variables that are simultaneously associated

with credit and land use, such as regional growth and agricultural output prices, may

also trivially jeopardize identiﬁcation. In this paper we limit identiﬁcation problems by

exploring the diﬀerent margins of a credit policy break that generated exogenous varia-

tions in credit supply, within and across municipalities of the Brazilian Amazon, in the

late 2 0 0 0 s.

Our strategy foll ows a two -sta g e procedure. First, we evaluate Resolution 3,545’s

impact on rural credit market quantities. Then, we use diﬀerent margins of this policy

change to derive instrumental variables for rural credit, which help us identify a causal

eﬀect of rural credit on deforestation in a two-stage least squares (2SLS) estimation.

6.1. First Stage: Impact on Rural Credit

We explore diﬀerent dimensions of Resolution 3,545 to provide a general characteri-

zation of its impacts on rural credit. First, the resolution speciﬁcally established that the

conditions to credit concession applied to municipalities in the Amazon Biome only. This

generates an explicit geographic cleavage between two groups of municipalities within

the Legal Amazon. We refer to municipalities located entirely within the Amazon Biome

as the intervention group. Legal Amazon municipalities located outside the biome form

the control group. Although intervention and contro l groups may systematically diﬀer in

terms of long-persistent geographical characteristics, both have been exposed to similar

economic ﬂuctuations and political cycles. This is particularly consistent for municipal-

ities near the biome frontier. This speciﬁc geographic break in Resolution 3,545, nested

within our monthly data on municipalities, allows us to perform a diﬀerence-in-diﬀerences

estimation deﬁned by the following equation:

Credit

= α

+ φ

+ β

(Biome

∗ Post2008

)+β

P rices

+ β

P riority

+ �

(13)

where Cred it

is the normalized amount of rural credit loaned in municipality i and time

t, which indexes a speciﬁc month throughout the period.

We estimate models for total municipality rural credit loans, as well as for cattle and

crop-speciﬁc loans separately. Our variable of interest is the interaction of a dummy indi-

cating whether the municipality is located within the Amazon Biome, Biome

, with a vari-

able that indicates the period after the implementation of Resolution 3,545, Post2008

which includes all years from 2008 o nwards. We consider all of 2008 as part of the post-

resolution p eriod to avoid ﬁnding a signiﬁcant coeﬃcient when, in fact, we are simply

capturing the anticipation eﬀect (recall that credit concession in the biome was very low

in the ﬁrst half of 2008 and very high in the second half).

The term α

includes municipality ﬁxed eﬀects, which absorb initial conditions and

persistent municipality characteristics, such as geography and transport infrastructure.

The term φ

includes month ﬁxed eﬀects, which also absorb year ﬁxed eﬀects by con-

struction. These variables control fo r common time trends, such as seasonal ﬂuctuations

in agricultural activity, macro eco nomic conditions, common rural policies, and the po-

litical cycle. The term P rices

proxies for municipality-speciﬁc demand for credit, as it

includes annual cattle and crop price indices (current and lagged) varying over time at

the municipality level. Finally, the term P riority

indicates municipalities included in a

priority list of top deforesters, which have recently been the focus of conservation policy

eﬀorts. The parameter of interest β

captures the causal eﬀect of Resolution 3,545 on

rural credit if the residuals contain no omitted variables driving the correlation between

the policy and either the demand for and/or supply of credit loans.

We rely on another diﬀerence-in-diﬀerences strategy to explore a second dimension

of Resolution 3,545’s design. As explained in Section 3.1, Resolution 3,545 established

exemptions regarding requirements for small producers. Considering that Pronaf bene-

ﬁciaries are, by deﬁnition, small producers, and taking into account that the restrictive

conditions were eased speciﬁcally for credit loans contracted via Pronaf credit lines, the

comparison between Pronaf and non-Pronaf loans within municipalities, before and af-

ter the resolution, oﬀers another source of exogenous variation in credit supply. The

speciﬁcities of the policy design therefore enable us to construct intervention and control

groups within municipalities located in the Amazon Biome. We take advantage of this

by estimating a triple-diﬀerences model that compares rural credit loans before and after

the implementation of Resolution 3,545, in municipalities located inside and outside the

biome, a nd between Pronaf and non-Pronaf groups within municipalities. This strategy,

based on an extension o f model (13), is deﬁned by the following equation:

Credit

itk

= α

+ φ

+ β

Nonpronaf

+ β

(Nonpronaf

∗ Biome

+β

(Nonpronaf

∗ Post2008

)+β

(Biome

∗ Post2008

)+ (14)

+β

(Nonpronaf

∗ Biome

∗ Post2008

)+β

P rices

+ β

P riority

+ �

itk

where Credit

itk

is now the amount of rural credit loaned in municipality i,attimet,and

of type k ∈ {Pronaf, non-Pronaf}. The term Nonpronaf

is a dummy assuming value 1

if (k = non-Pronaf) and 0 otherwise, which we interact with the dummies Post2008

and

Biome

. The parameter of interest β

is expected to capture the marginal variation in

non-Pronaf credit loans within the Amazon Biome after the implementation of Resolution

3,545.

We discuss some caveats associated with models (13) and (14) in robustness checks.

In all speciﬁcations, standard errors are robust to arbitrary forms of heteroscedasticity

and clustered at the municipality level to allow for correlation at a given time, and across

time within municipalities.

6.2. Second Stage: Impact on Deforestation

Using the municipality-by-year panel detailed in Section 5.2, we can therefore use

models (13) and (14) as ﬁrst stage regressions in a 2SLS approach, which helps us identify

acausaleﬀect of rural credit on deforestation.

We begin by deﬁning a second stage speciﬁcation for which model (13) is the ﬁrst

stage:

Deforest

= α

+ φ

+ δ



Credit

+ δ

P rices

+ δ

P riority

+ X

+ �

(15)

where Def or e st

is deforested area in municipality i and year t,and



Credit

is the total

value of rural credit loans in municipality i and year t,instrumentedbytheinteraction

variable (Biome

∗ Post2008

). The terms α

, φ

, P rices

,andP riority

are deﬁned

as in model (13), although the subscript t now indexes year. All regressions include the

term X

, which adds controls fo r the size of unobservable areas (or measurement error)

during the period of remote sensing.

The second stage speciﬁcation alternatively uses model (14) as ﬁrst stage, and is

deﬁned analogously to model (15). In this case, however, the unit of observation is

the cell indexed by itk, where k ∈ {Pronaf, non-Pronaf}. The dependent variable is

now Def orest

,invarianttok. The variable of interest is



Credit

itk

,thetotalvalueof

rural credit loans in municipality i and year t,instrumentedbytheinteractionvariables

Biome

∗ Post2008

, N onpronaf

∗ Post2008

,andN onpronaf

∗ Biome

∗ Post2008

The identiﬁcation hypo thesis in both 2SLS speciﬁcations is th at, conditioned upon

the control variables, instrumented credit is orthogonal to any latent determinant of

deforestation. Alternatively, the instrumental variables should be strongly correlated

with rural credit loans, but orthogonal to the error term in the second stage reg ressio n.

As arg ued above, this strategy is valid because Resolution 3,545 provides sources of

plausibly exo g eno us varia tion in rural credit loans in the Legal Amazon.

Resolution 3,545 was formally designed to restrict the concession of credit in the

Amazon Biome, with exceptions rendered to small producers. As argued in the previ-

ous section, these two dimensions of the resolution’s design provide potential sources of

exogenous variation in rural credit loans within the Legal Amazon.

Athirddimensionofitsdesignstrengthensthevalidityofusingtheresolution-induced

change in rural credit as an exogenous source of v ariation. Resolution 3,545 conditions

were such that borrowers who proved that they had the intention to comply with en-

vironmental regulation were allowed access to credit. This meant that producers who

feared the resolution might aﬀect their future access to credit could signal an intent to

change their deforestation behavior in the future and be considered compliant with envi-

ronmental regulation at the present. This could invalidate the exclusion restriction of our

instrumental variable, si nce farmers who are not meeting environmental regulation at the

present alter their deforestation behavior for rea sons other than a concurrent reduction

in credit caused by Resolution 3,545. However, this introduces a bias that runs in the

opposite direction of our expected positive coeﬃcient. After all, these producers will

suﬀer no credit eﬀect (as their intention to comply makes them compliers), but still re-

duce deforestation, driving downwards the coeﬃcient of interest. In terms of meeting the

resolution’s requirements, intention to comply is equivalent to compliance. Thus, there

were no other channels linking Resolution 3,545 and deforestation during the period of

interest. This eliminates a potential source of concern regarding the validity of using

the policy change as an instrument for credit and enables us to empirically examine the

relationship between credit and deforestation.

7. Policy Implementation and Eﬀectiveness

This section evaluates the impacts of Resolution 3,545 on rural credit loans. We

start by describing the main results obtained using the empirical strategies detailed in

Section 6.1. We then explore heterogeneity and characterize the policy impact. Finally,

we discuss caveats and present robustness checks.

7.1. Credit Constraint

Table 2 presents the results for regressions based on model (13) (Panel A) and model

(14) (Panel B), using normalized rural credit at the municipality level as the dependent

variable. Coeﬃcients in Panel A show that, conditioned upon controls, Resolution 3,545

was associated with a reduction in rural credit concession in the Amazon Biome, as com-

pared with the rest of the Legal Amazon. This eﬀect is signiﬁcant for total rural credit,

as well as for cattle and crop-speciﬁc loans. The impact on cattle-speciﬁc loans is larger

than that on crop-speciﬁc lo a ns, and the diﬀerence b etween their respective coeﬃcients is

statistically signiﬁcant. A plausible explanation for this diﬀerence is that cattle ranching

is relatively more land-intensive than crop farming, and therefore less compliant with

environmental regulation. In fact, previous studies show that three quarters of the land

cleared in the Legal Amazon have been used for cattle ranching-related activities, al-

though the relative participation of crop farming-related activities in cleared forest areas

has recently increased (see Chomitz and Thomas (2003) and Morton et al. (2006)).

Panel B corroborates the ﬁndings of Panel A, reporting even larger impacts when

considering the resolution’s impact on d iﬀerent kinds of credit takers. For a given mu-

nicipality in the Amazon Biome, credit for the non-Pronaf category decreased relatively

more than for the Pronaf category. These results are to be expected, given that Resolu-

tion 3,545 and associated legislation exempted small-scale producers - particularly Pro na f

beneﬁciaries - from the more restrictive conditions.

In Table 3, we present results from a counterfactual analysis to quantify the magnitude

of the impact. Our estimates indicate that, in the absence of the policy, total rural credit

and cattle-speciﬁc credit in the Amazon Biome would have been much greater than was

actually observed from 2008 onwards. Based on coeﬃcients from Panel A, in which the

control group is composed of municipalities outsid e the Amazon Biome, we estimate that

approximately BRL 2.9 billion (USD 1.4 billion) were not loaned in the 2008 through

2011 period due to restrictions imposed by Resolution 3,545. This is equivalent to about

BRL 725 million (USD 350 million) less credit p er year. The reduction in cattle-speciﬁc

credit l oans accounts for 90% of this diﬀerence. Estimates obtained from comparing

the resolution’s impact on Pro na f and non-Pronaf credit are very similar, although the

relative role of cattle-speciﬁc loans is slightly smaller in this simulation.

7.2. Heterogeneity

Having estimated a signiﬁcant impact of Resolution 3 ,5 4 5 on the concession o f rural

credit in the Amazon Biome, we are now interested in exploring potential heterogeneous

eﬀects arising from sectoral heterogeneity. Using the cattle-oriented and crop-oriented

subsamples described in Section 5.1, we test whether the resolution’s impact diﬀered

among municipalities with diﬀerent leading econo mic activity. These restricted samples

further allow us to test whether our estimated coeﬃcients have been jeopardized by

comparing municipalities that are not, in fact, economically comparable.

Table 4 presents results obtained by reproducing the speciﬁcations of Table 2 us-

ing the restricted cattle and crop-oriented subsamples. The signiﬁcant negative coef-

ﬁcient estimated for our variable of interest in the cattle-oriented subsample conﬁrms

that Resolution 3,545 reduced cattle-speciﬁc loans inside the Amazon Biome, even when

the comparison is made using only cattle-o riented municipalities both inside a nd outside

the biome. This supports our main results, providing further evidence that the credit

constraint was caused by the policy change. Results are similar for the crop-oriented

subsample. In this case, however, restricting the sample to municipalities that are eco-

nomically more alike has highlighted the policy impact, as estimated coeﬃcients are even

higher tha n in our original speciﬁca tio ns.

We also consider the distribution of loan size to analyze the impact of heterogeneity

along a diﬀerent dimension. The dependent variable is now the number of credit con-

tracts in each municipality categorized according to contract size. This exercise uses the

speciﬁcation that takes all municipalities in the Amazon Biome as the treatment group,

and all Legal Amazon municipalities outside the biome as the control group. We start

by dividing credit contracts into three categories based on their size: small (up to the

median), medium (between the median and the 75

percentile), and large (above the

percentile). We then aggregate these contracts by municipality to use as dependent

variable. Given that small producers beneﬁtted from less stringent conditions for credit

concession, while medium and large producers faced more restrictive conditions, we ex-

pect Resolution 3,545 to have a negative impact on the number of medium and large

contracts, and a neutral impact on small contracts. In fact, a positive impact on small

contracts could also be expected, should credit agents choose to reallocate resources away

from the more restricted group of credit takers and towards small producers.

Results shown in Table 5 indicate that, as expected, Resolution 3,545 had a signiﬁcant

negative impact on the number of medium and large cattle-speciﬁc credit contracts. The

eﬀect on crop-speciﬁc contracts, which had been small when estimated in speciﬁcations

using normalized rural credit as the dependent variable, is now signiﬁcant and negative

only for medium credit contracts. The absence of a signiﬁcant negative impact on small

producers is to be expected, but the coeﬃcient estimated for the large producers shows

that, contrary to what was originally thought, large-scale crop producers were also not

aﬀected by the conditional rural credit policy. This might be because they could more

easily meet the conditions required in Resolution 3,545, be it due to their better orga-

nizational capacity o r to their greater access to resources needed to follow through with

the regularization procedure. After all, in regions where crop pro duction is predominant

in the Amazon Biome, rural economic activities are mostly based on larg e-sca le soybean

farmers, who might have other sources of credit ﬁnancing and a large organized chain

of production. Table 5 shows, moreover, that the resolution had a signiﬁcant positive

eﬀect on small cattle-speciﬁc credit contracts. This corroborates the hypoth esis stating

that Resolution 3,545 had a distributio nal e ﬀect, leading credit agents to change their

allocation of resources, moving it away from the group of credit takers subject to the

resolution’s conditions and towards the group that received legal exemptions.

Overall, the results obtained so far show that Resolution 3,545 produced diﬀerentiated

eﬀects within the Amazon Biome. While the policy change led to a reduction in the

number of medium and large cattle-speciﬁc contracts and in the number of medium crop-

speciﬁc contracts, it had no impact on both large and small crop-speciﬁc loans, and a

positive impact on small cattle-speciﬁc loans. The resolution also appears to have led

to greater credit reduction in municipalities where cattle ranching is the main economic

activity, though the impact on rural credit in municipalities where crop production is the

main eco no mic activity is not negligible.

7.3. Caveats and Robustness Checks for Credit Results

Although our results are generally consistent with Resolution 3,545’s institutional

context, we run a series of tests to check their robustness. We focus on three sources of

potential concern. First, it could be that our regressions capture a spurious eﬀect due to

the deﬁnition of the intervention and comparison groups that we erroneously attribute

to the policy change. Second, our identiﬁcation strategy relies on the hypothesis that,

after controlling for observable characteristics and municipality and time ﬁxed eﬀects,

both intervention and control groups are comparable. Yet, municipalities in diﬀerent

regions could diﬀer in terms of regional economic dynamics or non-observable regional

characteristics. In this sense, comparing municipalities that are near to the Amazon

Biome frontier with those that are far from it could jeopardize our results. Third, the

use of a normalized depen dent variable could be driving our results.

We tackle these issues in three sets of robustness checks. First, we explore the res-

olution’s conditions - namely, its restriction to the Amazon Biome and the exemptions

created for small producers - to perform falsiﬁcation tests. We start by considering the

speciﬁcations presented in Table 2, Panel A, in which municipalities inside the Amazon

Biome are taken as the treatment group and compared to Legal Amazon municipalities

outside the biome. In the ﬁrst falsiﬁcation test, we rep eat the exercise using the amount

of credit loaned strictly to Pronaf beneﬁciaries as the dependent variable. As this group

of borrowers beneﬁtted from the aforementioned institutional exemptions, we expect this

falsiﬁcation test to capture no signiﬁcant impact of Resolution 3,545 on Prona f credit

inside the biome. If anything, the test could return a positive signiﬁcant impact of the

resolution on Pronaf credit, should its distributional impact actually have caused a shift

of credit towards unrestricted small producers inside the biome, as discussed in Section

7.2. We then move on to the speciﬁcations presented in Table 2, Panel B, in which we

compare Pronaf and no n-Pro na f credit inside and outside the Amazon Biome. In the

second falsiﬁcation test, we again repeat the speciﬁcations, but restrict our sample to

Legal Amazon municipalities located outside the biome. As Resolution 3,5 45 conditions

did not apply to the Legal Amazon as a whole, we expect this falsiﬁcation test to capture

no signiﬁcant negative impact of the resolution on Pronaf and non-Pronaf credit outside

the biome.

Panels A.1 and B.1 of Table 6 present the results for the two falsiﬁcation tests. In Panel

A.1, we note that the coeﬃcient for the Biome ∗ Post2008 variable is now insigniﬁcant,

as expected. When using crop-speciﬁc contracts, the coeﬃcient is positive, although

relatively small, indicating that a small shift in allocation of ﬁnancing resources towards

Pronaf beneﬁciaries did occur for crop-speciﬁc loans. In Panel B.1, the coeﬃcient of

interest becomes non-negative, a s expected. In fact, when focusing on cattle-speciﬁc

contracts, results indicate that non-Pronaf credit was increasing more than for Pronaf

beneﬁciaries. Interestingly, should this pattern also apply to non-Pronaf credit inside the

Amazon Bio me, the real impact of Resolution 3,545 could be even higher than we have

estimated.

Our second set of robustness checks tests if results have been driven by regional or

economic diﬀerences between municipalities. To do this, we create a subsample restricted

to municipalities that are within 100 kilometers of the Amazon Biome border. Treatment

and control groups still refer to Legal Amazon municipalities inside and outside the

biome, respectively, but now both treatment and control municipalities must meet the

maximum distance to biome border criteria. This subsample contains municipalities that

are geographically closer to one another, and are thus thought to b e more alike across

both observable and non-observable ﬁxed characteristics and trends.

Panel A.2 and B.2 in Table 6 show the results for this second robustness check.

Speciﬁcations are analogous to those of Table 2. Robustness results shown in Panel A.2

are consistent with those of our main speciﬁcations and estimated coeﬃcients are, in fact,

higher. This suggests that municipa lities closest to the biome frontier, which are probably

in a more economically dynamic region, are more heavily dependent on credit and were

thus more aﬀected by the conditions established in Resolution 3,545. Moreover, it seems

plausible that, in a region with more dynamic agricultural production, producers are

less compliant with environmental regulations. Estimated coeﬃcients of interest remain

negative and signiﬁcant in Panel B.2, except in the speciﬁcation for crop-speciﬁc loans,

which is now less signiﬁcant, although of magnitude similar to that of Table 2.

The third and ﬁnal robustness check consists of replacing the normalized credit vari-

able with an absolute measure of total rural credit to test if our results have been driven

by inaccuracies introduced via the normalization of the dependent v ariable. Results are

presented in Panels A.3 and B.3 of Table 6, again using the speciﬁcations of Table 2.

Estimated coeﬃcients are essentially the same as those obtained in our main regressions.

Overall, the robustness of our results support the speciﬁcations chosen for our main

regressions, as well as the interpretation of our results.

8. Credit Constraint and Deforestation

This section presents the results of the second stage of our 2SLS strategy, that is, the

impact of the credit constraint on deforestation in the Amazon Biome. As mentioned

in Section 1, the directio n of the eﬀect of a reduction in credit on deforestation depends

on how rural credit is used. If used to improve production techniques and intensify

productivity per unit of land used for agricultural production, a credit restriction could

limit productivity gains and thereby generate an incentive for producers who seek to

expand their production to move into new areas. If, however, credit is used to expand

production at the extensive margin - to clear forest areas and accommodate a growing

herd, for example - the expected result of a reduction in credit should be a decrease in

deforestation. Which of these mechanisms prevails in the Ama zo n Biome is an empirical

question.

We can also think about our regressions as an econometric test ab out the existence of

binding credit constra int in activities related to deforestation. The usual way to assess if

aﬁrmiscreditconstrainedintermsofinvestment,forinstance,istomeasureiftheﬁrm

increases its amount of investment following a policy change that facilitates its access

to credit (Banerjee and Duﬂo (2012)). Sh ou ld the ﬁrm do so, it had not yet reached

its optimal amount of investment before the policy - the ﬁrm faced credit constraints.

The arg ument we make here is analogous to that of the credit constrained ﬁrm, but runs

in the opposite direction. If a deforester decreases the amount of forest cleared due to

arestrictioninsubsidizedcredit-meaninghewouldincreasedeforestationifhehad

more subsidized credit - then the deforester is credit constrained. In other words, the

deforester requires access to credit to implement his activities, be it the actual clearing

of forest or activities practiced after the forest area has been converted. If he is changing

his investment (deforestation) behavior as a response to a reduction in the availability of

subsidized credit, then he is not substituting the lost subsidized credit for market credit,

which serves as evidence that he is credit constrained.

At the same time, if farmers are decreasing the amount of deforestation in response to

a reduction in subsidized credit availability, it means that, within the theoretical frame-

work discussed in Section 4, they are not switching from the modern to the traditional

technology. Rather, they a re adjusting their optimal farmland decision within regions of

ﬁxed technology. Should they be changing between technologies, we would observe an

increase in deforestation as a respo nse to a reduction in availability of subsidized credit.

8.1. Main Results

Table 7 presents results for the evaluation of how Resolution 3,5 4 5 aﬀected deforesta-

tion in the Amazon Biome. Column 1 displays coeﬃcients estimated in a ﬁxed eﬀects

speciﬁcation without the use of instrumental variables (IV). The coeﬃcient capturing the

impact of rural credit on deforestation is positive, but very small. As discussed in Section

6, however, these speciﬁcations suﬀer from an endogeneity problem, due both to omitted

variables and simultaneity. The IV approach is used to tackle this problem. In column

2, we present coeﬃcients estimated in IV regressions that instrument rural credit by the

policy variable Biome ∗ Post2009.Resultssuggestthatcredithasapositiveandstrongly

signiﬁcant relationship with defo restation, and that the reduction in the concession of

rural credit in the Amazon Biome ca used by Resolution 3,545 contributed to the curbing

of deforestation in the biome. In particular, municipalities that were most aﬀected by the

resolution-induced credit reduction were also the ones that presented the sha rper drops

in deforestation. Similar results, shown in column 3 are obtained when using the triple

interaction term Nonpronaf ∗ Biome ∗ Post2009 as an instrumental variable, in addition

to the varia bles Biome ∗ Post2009 and Nonpronaf ∗ Post2009.

To better grasp the economic signiﬁcance of our results, we conduct counterfactual

simulations analogous to those presented in Section 7.1 but now using the speciﬁcation

of Table 7, column 2. Results shown in Table 8 indicate that, in the absence of Resolu-

tion 3,545, deforestation in the Amazon Biome would have been 2 ,7 83 square kilometers

greater than was actually observed in the 2009 through 2011 period. This is equiva lent

to an increase of about 18% over observed deforestation. In 201 0, for example, deforesta-

tion would have totaled 7,398 square kilometers had the resolution not been introduced,

compared to the 5,657 square kilometers that were actually observed. This result is

particularly impressive if we take into consideration the comparatively low deforestation

rates reco rded during the same period.

Our results serve as evidence of the existence of binding credit constraints in the

Amazon Biome. Farmers appea r to have responded to a reduction in the availability of

subsidized credit by changing their optimal allocation of resources and thereby reducing

deforestation. Were there no binding credit constraints, farmers’ actions would not have

resulted in a change in deforestation in the post-policy period. Moreover, our results

suggest that the prevailing mechanism relating rural credit and forest clearings in the

Amazon Biome is that in which credit is used to expand production by operating in the

extensive margin of land use, and not by increasing productivity. The predominance of

cattle ranching in the region and the correlation between this activity and extensive land

use in the Amazon could explain these results. Intensiﬁcation of land use for crop farming

in Brazil is also much more develop ed than for ca ttle ranching.

8.2. Heterogeneity

To test whether the relationship between credit and deforestation depends on regional

heterogeneity, particularly that of leading economic activity, we repeat the empirical exer-

cise of Section 8.1 using cattle a nd crop-oriented municipalities. This allows to investigate

if diﬀerent types of economic activity use credit diﬀerently, increasing production either

at the extensive margin or at the intensive margin.

Table 9 presents heterogeneity results. Coeﬃcients estimated for cattle-oriented mu-

nicipalities, shown in Panels A.1 and B.1, are very si milar to those of Table 7, conﬁrming

the positive relationship between credit and deforestation. The reduction in credit conces-

sion caused by Resolution 3,545 therefore implied a reduction in deforestation in munici-

palities where cattle ranching predominates. Yet, coeﬃcients estimated for crop-oriented

municipalities, shown in Panels A.2 and B.2, indicate that rural credit has no impact

on deforestation. As has been documented before (see FAO (2007), crop production in

Brazil underwent several technological improvements, allowing pro duction to increase in

the intensive margin. If farmers are able to increase production via intensiﬁcation, they

do not use credit to clear new land for production. In this case, chang es in rural credit

do no t aﬀect deforestation.

The results can also be regarded as evidence that crop farmers are not credit con-

strained. They may have compensated the reduction in subsidized rural credit with

an increase in other sources of ﬁnancing (market credit), thus maintaining their pre-

resolution levels of investment, farmland and deforestation. Cattle ranchers, on the other

hand, seem to be credit constrained (or at least became constrained after the policy

change), since they reduce the amount of deforestation in response to the reduction in

the availability of credit.

8.3. Robustness Checks

The concerns mentioned in Section 7.3 also apply to our second-stage regressio ns. As

the Amazon is an extremely large area, municipali ties in the treatment groups system-

atically diﬀer f rom those in the control groups. To tackle this issue, we run robustness

checks using the subsample of municipalities within 100 kilometers of the biome border.

Results for this test, shown in Panels A.1 and B.1 of Table 10, indicate that no t only are

estimated coeﬃcients signiﬁcant, but also greater in magnitude. As the frontier covers

an area with very high deforestation rates - it contains a large fraction of the so-ca lled

"Arc of Deforestation," deforestation hotspots in the Brazilian Amazon - this result is to

be expected. As argued in Section 7.3, the economic dynamics of this region likely makes

this area more dependent on rural credit. An exogenous reductio n in rural credit would

therefore have a higher impact on agricultural activities in the region, and thereby more

strongly aﬀect deforestation.

To test whether our results are driven by the use of the normalized variables, we

also repeat 2SLS estimations using absolute values for the deforestation and rural credit

variables. Results presented in Panels A.2 and B.2 of Table 10 show that coeﬃcients

estimated using IV regressions remain positive and signiﬁcant, although their magnitude

varies according to the speciﬁcation.

Finally, to test whether our results are driven by a natural convergence of high to low

deforestation rates, we control for an interaction between a non-linear trend and the 2002

deforestation rates in each municipality. If the results are driven only by the convergence

process, our main coeﬃcient should become zero. Results presented in Panels A.3 and B.3

of Table 10 show that coeﬃcients remain positive and signiﬁcant even when we control

for these trends. First stage results are also signiﬁcant when controlling for these trends,

indicating that our instruments are not weak. These estimates corroborate the idea that

our results are not driven by a natural process of convergence.

The robustness tests conducted support our main speciﬁcations and interpretation of

results. The signiﬁca nt reduction in rural credit concession caused by Resolutio n 3,545

seems to have played an important role in curbing forest clearings in the Amazon Biome

in the late 2000s and early 2010s.

9. Final Comments

In this work, we investigate the reduction in the availability of credit implied by

Resolution 3,545. We focus on key aspects of the implementation and consequences

of this novel credit policy, exploring the associated exogenous variation in rural credit

concessions to empirically evaluate the resolution’s impact on both rural loans and forest

clearings.

We document that credit takers anticipated the credit constraint imposed by Resolu-

tion 3,545, but that the policy change did, in fact, lead to a reduction in the concession of

rural credit in the Amazon Biome. Counterfactual simulations suggest that, in the 2008

through 2011 period, approximately BRL 2.9 billion (USD 1.4 billion) were not loaned

due to restrictions imposed by the resolution. This result was largely driven by a reduc-

tion in cattle-speciﬁc loa ns. Resolution 3,545 also ca used a decrease in total amo unt of

non-Pronaf credit, as compared to Pronaf credit, within the biome. This is to be expected

in light of the legal exemptions that were introduced for small-scale producers regarding

legal requirements to access rural credit.

Estimations from instrumental variable regressions further show that the resolution-

induced restriction in credit helped contain deforestation in the Amazon Biome. This

result sug g ests that the expansion o f agriculture at the extensive margin in the biome

is ﬁnancially constrained. Counterfactual simulations indicate that over 2,700 squa re

kilometers of forest would have been cleared from 2009 through 2011, had Resolution

3,545 not been implemented. Considering that deforestation rates in the late 2000s and

early 2010s were around 7,000 square kilometers per year, the eﬀect attributed to the

resolution is quite substantial.

The results have two key policy implications. First, the evidence shows that condi-

tional rural credit can be an eﬀective p olicy instrument to combat deforestation. Along

these lines, the diﬀerential eﬀects across sectors and regions suggest that it might comple-

ment rather than substitute other conservatio n eﬀorts. The pre-existent socioeco no mic

circumstances matter - credit reduction came mostly from the reduction of cattle loans

rather than crop loans. The implementation details also matter. The lag between the

announcement and enforcement of the resolution induced farmers to anticipate credit in

2008, mitigating part of the eﬀect. Also, less stringent requirements and exemptions have

determined that large producers were more aﬀected than small producers.

Second, our analysis suggests that the ﬁnancial environment in the Amazon is charac-

terized by signiﬁcant credit constraints. Especially in municipalities where cattle ranching

is the predominant activity, fewer resources correspond with less deforestation. This is a

key ﬁnding with implications for po licy design. In particular, policies that increase the

av ailability of ﬁnancial resources (for example, payments for environmental services) may

lead to higher defo resta tio n rates, depending on the economic environment and exist-

ing resources in the area. Our results do not suggest that these policies will necessarily

increase defo resta tio n, but that these policies should take into account the na ture of ﬁnan-

cial constraints that are prevailing in the Amazon, avoiding potentially adverse rebound

eﬀects.

Acknowledgments

Arthur Bragança, Luiz Felipe Brandão, Pedro Pessoa and Ricardo Dahis provided

excellent research assistance. We thank the Ministry of the Environment, particularly

Roque Tumolo Neto, for their continuous support. We are also grateful to Arthur Bran-

gança, Dimitri Szerman, Jo a na Chiavari, and Pedro James Hemsley for helpful comments.

References

Alston, L. J. and Andersson, K. (2011). Reducing Greenhouse Gas Emissions by Forest Protection: The

Transaction Costs of REDD. Working Paper 16756, The National Bureau of Economic Research.

Angelsen, A. (2010). Policies for Reduced Deforestation and Their Impact on Agricultural Production.

Proceedings of the National Academy of Sciences,107(46):19639.

Angelsen, A. and Wertz-Kanounnikoﬀ,S.(2008). Moving Ahead with REDD: Issues, Options and Im-

plications. Center for International Forestry Research (CIFOR).

Assunção, J., Rocha, R., and Gandour, C. C. e. (2011). Deforestation Slowdown in the Legal Amazon:

Prices or Policies? Working Paper 1.

Banerjee, A. V., M. K. and Duﬂo, E. (2012). Do ﬁrm want to borrow more? testing credit constraints

using a directed lending program. Mimeo, MIT.

Binswanger, P. H. (1991). Brazilian policies that encourage deforestation in the amazon. World Devel-

opment,19(7):821–829.

Brasil, Ministério da Agricultura, Pecuária e Abastecimento (2001). Plano Agrícola e Pecuário 2001–

2002.SecretariadePolíticaAgrícola,Brasília.

Brasil, Ministério da Agricultura, Pecuária e Abastecimento (2003). Plano Agrícola e Pecuário 2003–

2004.SecretariadePolíticaAgrícola,Brasília.

Brasil, Ministério da Agricultura, Pecuária e Abastecimento (2006). Plano Agrícola e Pecuário 2006–

2007.SecretariadePolíticaAgrícola,Brasília.

Brasil, Ministério da Agricultura, Pecuária e Abastecimento (2011). Plano Agrícola e Pecuário 2011–

2012.SecretariadePolíticaAgrícola,Brasília.

Burgess, R. and Pande, R. (2005). Do rural banks matter? evidence from the indian social banking

experiment. American Economic Review,95:780–795.

Chomitz, K. and Thomas, T. (2003). Determinants of Land Use in Amazônia: A Fine-Scale Spatial

Analysis. American Journal of Agricultural Economics,85(4):1016–1028.

Conning, J. and Udry, C. (2007). Rural Financial Markets in Developing Countries. In The Handbook

of Agricultural Economics, Vol. 3.

Conselho Monetário Nacional (2008). Resolução 3.545 de 29 de Fevereiro de 2008.

FAO (200 6) . G lob al For est Reso urc es As ses sme nt 200 5: Pr ogr es s Towards Su st a in abl e Fore st M an age -

ment. Rome: Food and Agriculture Organization (FAO).

FAO (200 7) . Fu t ur e e x pa nsi on of s oy b ea n 2 00 5 -2 014 : imp li c at ion s f or fo o d s ecu ri ty, su st ain abl e r ura l

development and agricultural policies in the countries of mercosur and bolivia. Technical report, FAO

Regional Oﬃce for Latin America and the Caribbean.

Feder, G. (1985). The Relation Between Farm Size And Farm Productivity: The Role Of Family Labor,

Supervision And Credit Constraints. Journal of Development Economics,18(2-3):297–313.

Hargrave, J. and Kis-Katos, K. (2010). Economic Causes of Deforestation in the Brazilian Amazon: A

Panel Data Analysis for 2000s. In Proceedings of the German Development Economics Conference,

Hannover. Verein für Socialpolitik, Research Committee Development Economics.

INPE (2012). Projeto prodes - monitoramento da ﬂoresta amazônica brasileira por satélite. Technical

report, INPE. http://www.obt.inpe.br/prodes/index.php.

IPAM (2009). Evolução na Política para o Controle do Desmatamento na Amazônia Brasileira: O

PPCDAm. Clima e Floresta,15.

Ipea, Cepal, and GIZ (2011). Avaliação do Plano de Ação para a Prevenção e Controle do Desmatamento

da Amazônia Legal. Technical report.

Morton, D. C., DeFries, R. S., Shimabukuros, Y. E., Anderson, L. O., Arai, E., Espirito-Santo, F.,

Freitas, R., and Morisette, J. (2006). Cropland Expansion Changes Deforestation Dynamics in the

Southern Brazilian Amazon. Proceedings of the National Academy of Sciences,103(39):14637Ű–14641.

Pfaﬀ,A.S.(1999). WhatDrivesDeforestationintheBrazilianAmazon? EvidencefromSatelliteand

Socioeconomic Datal. Journal of Environmental Economics and Management,37(1):26–43.

Presidência da República (2006). Lei No. 11.326 de 24 de julho de 2006.

Reis, E. and Guzmán, R. (1994). An Econometric Model of Amazon Deforestation. In K. Brown and

D. Pearce, eds. The Causes of Tropical Deforestation. London: UCL Press,pages172–191.

Reis, E. and Margulis, S. (1991). Options for Slowing Amazon Jungle Clearing. In R. Dornbusch and

J.M. Poterba, eds. Global Warming: Economic Policy Responses. Cambridge, MA: MIT Press,pages

335–375.

Rosenzweig, M. and Wolpin, K. I. (1993). Credit Market Constraints, Consumption Smoothing, and

the Accumulation of Durable Production Assets in Low-Income Countries: Investments in Bullocks

in India. Journal of Political Economy,101:223–244.

Santilli, M., Moutinho, P., Schwartzman, S., Nepstad, D., Curran, L., and Nobre, C. (2005). Tropical

Deforestation and the Kyoto Protocol. Climatic Change,71:267–276.

Udry, C. (2010). The economics of agriculture in africa: Notes toward a research program. Mimeo, Yale.

Figure 1: The Brazilian Amazon Biome and Legal Amazon

Figure 2: Theoretical Model - Optimal Farmland Size (M

< M<M)

!"#$%&'

&%()'*+,-'

./%)+#!(%&'

.-01(!&!23'

40!(*./%+(-)5'

$!)-/('

.-01(!&!23'

40!(*./%+(-)5'

$!)-/('

.-01(!&!23'

46(0!(*./%+(-)5'

Figure 3: Annual Concession of Rural Credit in the Legal Amazon (municipality-level

average value of rural credit loans, 2002-2011)

(a) Cattle-oriented sample

!""

#""

$""

%""

&""

!"#$%&'()*+'&,-./01$2

'""

(""

(""( (""! (""# (""$ (""% (""& ("") (""* ("'" ("''

+,-./0123456+728/+41

/7./0123456+728/+41

(b) Crop-oriented sample

!"###

!"$##

%"###

%"$##

!"#$%&'()*+'&,-./01$2

$##

%##% %##& %##' %##$ %##( %##) %##* %##+ %#!# %#!!

,-./01234567,8390,52

08/01234567,8390,52

Notes: Authors’ elaboration based on data from Recor/Central Bank.

Figure 4: Monthly Concession of Rural Credit in the Amazon Biome (total value of rural

credit loans, 2006-2008)

(a) Cattle-speciﬁc contracts

!"#

$##

$"#

%##

%"#

!"#$%&'()*+',%--%./0

!##

$##&

$##'

$##(

(b) Crop-speciﬁc contracts

$""

$%"

$&"

!"#$%&'()*+',%--%./0

%""!

%""'

%""#

Notes: Authors’ elaboration based on data from Recor/Central Bank.

Figure 5: Deforestation in the Legal Amazon (municipality-level average deforestation,

2002-2011)

(a) Cattle-oriented sample

)(

!"#$%"&'"!()%")(*+,

(

)

(

)(

*((* *(() *((! *((" *((# *(($ *((% *((& *('( *(''

+,-+./012345,06+52/078/9:03;+-<

5=:-+./012345,06+52/07>+?@:03;+-<

(b) Crop-oriented sample

)(

!"#$%"&'"!()%")(*+,

(

)

(

)(() )((* )((! )((" )((# )(($ )((% )((& )('( )(''

+,-+./012345,06+52/078/9:03;+-<

5=:-+./012345,06+52/07>+?@:03;+-<

Notes: Authors’ elaboration based on data from PRODES/INPE.

Table 1: Municipality-Level Average Annual Rural Credit and Deforestation Before and After Resolutio n 3,545 (credit in BRL thousand

by type of contract; deforestation in km

)

Aggregate Credit Pronaf Credit Non-Pronaf Credit

Deforestation

Total Cattle Crop Total Cattle Crop Total Cattle Crop

Panel A: All municipalities

Inside biome Pre-resolution 5,509 3,531 1,978 1,254 841 413 4,255 2,690 1,565 40.36

Post-resolutio n 6,3 6 0 4,429 1,9 3 0 1,976 1,458 518 4,383 2,97 1 1,413 14.90

Outside biome Pre-resolution 12,092 2,865 9,227 784 475 309 11,308 2,390 8,918 3.03

Post-resolutio n 15,594 5,371 10,223 1,110 776 334 14,484 4,595 9,8 8 9 1.72

Panel B: Cattle-oriented municipalities

Inside biome Pre-resolution 5,799 4,655 1,144 1,581 1,152 430 4,218 3,504 714 49.57

Post-resolutio n 6,9 5 4 5,843 1,1 1 0 2,396 1,941 455 4,558 3,90 2 656 18.07

Outside biome Pre-resolution 4,865 2,987 1,878 757 490 267 4,108 2,497 1,611 2.64

Post-resolutio n 7,8 9 4 5,715 2,1 8 0 1,024 774 250 6,870 4,940 1,930 1.95

Panel C: Crop-oriented municipalities

Inside biome Pre-resolution 5,841 1,620 4,221 738 289 449 5,103 1,3 3 1 3,772 21.60

Post-resolutio n 6,1 2 7 2,010 4,1 1 8 1,352 613 739 4,775 1,397 3,379 8.27

Outside biome Pre-resolution 23,112 2,679 20,433 826 453 373 22,286 2,226 20,060 4.41

Post-resolutio n 27,334 4,848 22,486 1,242 780 462 26,093 4,068 22,025 0.90

Notes: Descriptive statistics for average values of annual rural credit are calculated using municipality-level data constructed from the Recor

contract-level microdata set. Statistics are shown for aggregate rural credit, and for Pronaf and Non-Pronaf credit separately. Descriptive

statistics for deforestation are calculated using municipality-level satellite-based deforestation data from PRODES/INPE. Figures in Panel A are

for the full, unrestricted sample; in Panel B for the cattle-oriented subsample; and in Panel C for the crop-oriented subsample.

Table 2: The Eﬀect of Resolution 3,545 on Rural Credit in the Amazon Biome

Panel A: Biome as treatment group

Total Cattle Crop

Biome * Po st2008 -0.216 -0.289 -0.078

(0.029)*** (0.033)*** (0.024)***

Observations 85,440 85,440 85,440

Number of Municipalities 712 712 712

Year and Municipality FE Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon

Panel B: Biome and non-Pronaf as treatment groups

Total Cattle Crop

Biome * Non-Pronaf * Post2008 -0.369 -0.407 -0.189

(0.039)*** (0.041)*** (0.033)***

Non-Pronaf * Post2008 0.064 0.116 -0.005

(0.030)** (0.031)*** (0.025)

Biome * Non-Pronaf 0.148 0.163 0.076

(0.016)*** (0.016)*** (0.013)***

Biome * Post2008 0.026 0.007 0.064

(0.030) (0.033) (0.025)***

Non-Pronaf -0.025 -0.046 0.002

(0.012)** (0.012)*** (0.010)

Observations 170,880 170,880 170,880

Number of Municipalities 712 712 712

Year and Municipality FE Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon

Notes: Coeﬃcients shown are estimated using ﬁxed eﬀects speciﬁcations in a municipality-by-

month panel data set covering the 2002 through 2011 period. The sample includes all Legal

Amazon municipalities that are not crossed by the Amazon Biome frontier. The dependent

variable is normalized rural credit at the municipality level calculated using all contracts (column

1), cattle-speciﬁc contracts only (column 2), and crop-speciﬁc contracts only (column 3). All

regressions include controls for municipality ﬁxed eﬀects, month-year ﬁxed eﬀects, cattle prices

(current and lagged), crop prices (current and lagged), and priority municipality status. Robust

standard errors are clustered at the municipality level. Signiﬁcance: *** p<0.01, ** p<0.05, *

p<0.10.

Table 3: Counterfactual Simulations - The Eﬀect of Resolution 3,545 on Rural Credit in the Amazon Biome (credit in BRL million)

Panel A: Biome as treatment group

Year

Total Cattle Crop

Observed Estimated Diﬀerence Observed Estimated Diﬀerence Observed Estimated Diﬀerence

2002 1,595 1,092 503

2003 2,306 1,312 994

2004 3,002 1,679 1,324

2005 2,982 1,945 1,037

2006 2,623 1,856 767

2007 2,630 1,818 812

2008 2,506 3,174 668 1,740 2,253 512 765 944 179

2009 2,772 3,594 821 1,845 2,564 719 927 1,079 152

2010 3,203 3,852 649 2,271 2,873 601 932 1,008 76

2011 3,170 3,928 758 2,258 3,037 779 912 945 33

2008-2011 Total 11,651 14,547 2,896 8,114 10,727 2,611 3,536 3,976 440

Panel B: Biome and non-Pronaf as treatment groups

Year

Total Cattle Crop

Observed Estimated Diﬀerence Observed Estimated Diﬀerence Observed Estimated Diﬀerence

2002 1,595 1,092 503

2003 2,306 1,312 994

2004 3,002 1,679 1,324

2005 2,982 1,945 1,037

2006 2,623 1,856 767

2007 2,630 1,818 812

2008 2,506 3,036 530 1,740 2,052 312 765 943 178

2009 2,772 3,607 835 1,845 2,449 604 927 1,085 158

2010 3,203 3,795 592 2,271 2,694 423 932 1,002 70

2011 3,170 3,905 736 2,258 2,886 628 912 933 21

2008-2011 Total 11,651 14,342 2,693 8,114 10,081 1,967 3,536 3,963 427

Notes: Counterfactual simulations are conducted using the sample, speciﬁcations and estimated coeﬃcients from Table 2. Results shown are

for total rural credit (columns 1-3), cattle-speciﬁc contracts (columns 4-6), and crop-speciﬁc contracts (columns 7-9). Observed rural credit is

obtained from Recor/Central Bank and estimated rural credit is calculated by replacing the coe ﬃcient of the policy variable by zero.

Table 4: Sectoral Heterogeneity - Cattle-Oriented and Crop-Oriented Sub-

samples

Panel A: Biome as treatment group

Cattle-oriented Crop-oriented

Biome * Po st2008 -0.300 -0.128

(0.042)*** (0.040)***

Observations 54,600 27,960

Number of Municipalities 455 233

Year and Municipality FE Yes Yes

Sample Legal Amazon Legal Amazon

Panel B: Biome and non-Pronaf as treatment groups

Cattle-oriented Crop-oriented

Biome * Non-Pronaf * Post2008 -0.490 -0.246

(0.053)*** (0.060)***

Non-Pronaf * Post2008 0.193 -0.030

(0.040)*** (0.041)

Biome * Non-Pronaf 0.196 0.098

(0.021)*** (0.024 )* * *

Biome * Post2008 0.069 0.048

(0.041)* (0.042)

Non-Pronaf -0.077 0.012

(0.016)*** (0.016)

Observations 109,200 55,920

Number of Municipalities 455 233

Year and Municipality FE Yes Yes

Sample Legal Amazon Legal Amazon

Notes: Coeﬃ cients shown are estimated using ﬁxed eﬀects speciﬁcations in a

municipality-by-month panel data set covering the 2002 through 2011 period.

The cattle-oriented subsample includes Legal Amazon municipalities that are

not crossed by the Amazon Biome frontier where the pre-2008 average value of

annual credit loans for cattle ranching was higher than that for crop production;

the crop-oriented subsample is deﬁned analogously. The dependent variable is

normalized rural credit at the municipality level calculated using cattle-speciﬁc

contracts only (column 1), and crop-speciﬁc contracts only (column 2). All re-

gressions include controls for municipality ﬁxed eﬀects, month-year ﬁxed eﬀects,

cattle prices (current and lagged), crop prices (current and lagged), and priority

municipality status. Robust standard errors are clustered at the municipality

level. Signiﬁcance: *** p<0.01, ** p<0.05, * p<0.10.

Table 5: The Eﬀect of Resolution 3,545 on the Distribution of Rural Credit Contracts by Contract Size

Number of cattle-speciﬁc contracts, by contra ct size Number of crop-speciﬁc contracts, by contract size

Small Medium Large Small Medium Larg e

Biome * Post2008 3.405 -0.798 -1.345 1.642 -0.486 0.107

(1.508)** (0.421)* (0.331)*** (1.757) (0.263)* (0.234)

Crops price index 2.486 -0.300 0.428 -0.872 0.641 -0.215

(3.938) (0.705) (2.458) (3.708) (0.529) (0.696)

Crops price index (lagged) 16.342 -1.823 0.951 1.294 0.232 3.368

(4.220)*** (0.805)** (0.735) (3.626) (0.760) (1.599)**

Cattle price index (1

sem) 0.286 -0.053 -0.344 0.342 -0.114 12.650

(0.148)* (0.055) (0.658) (0.173)** (0.041 )*** (2.959)***

Cattle price index (lagged) 0.295 0.064 -0.034 0.242 0.159 -0.029

(0.125)** (0.058) (0.059) (0.170) (0.035)*** (0.022)

Priority municipalities 1.190 -4.267 0.073 -1.142 -1.292 0.032

(4.305) (2.045)** (0.062) (6.908) (0.790) (0.030)

Observations 93,720 93,720 93,720 93,720 93,720 93,720

Number of Municipalities 781 781 781 781 781 781

Year and Municipality FE Yes Yes Yes Yes Yes Yes

Notes: Coeﬃcients shown are estimated using ﬁxed eﬀects speciﬁcations in a municipality-by-month panel data set covering the 2002 through 2011 period. The

sample includes all Legal Amazon municipalities that are not crossed by the Amazon Biome frontier. The dependent variable is the number of small, medium, and

large cattle-speciﬁc contracts only (columns 1-3) and the number of small, medium, and large crop-speciﬁc contracts only (columns 4-6). All regressions include

controls for municipality ﬁxed eﬀects, month-year ﬁxed eﬀects, cattle prices (current and lagged), crop prices (current and lagged), and priority municipality

status. Robust standard errors are clustered at the municipality level. Signiﬁcance: *** p<0.01, ** p<0.05, * p<0.10.

Table 6: Robustness Checks - Falsiﬁcation Tests, Alternative Sample and Alternative Dependent Variable

Panel A.1: Falsiﬁcation test - Panel A.2: Alternative sample - Panel A.3 : Alternative dependent variable -

Pronaf credit as dependent variable municipalities within 100km of biome frontier absolute values of rural credit

Total Cattle Crop Total Cattle Crop Total Cattle Crop

Biome * Post2008 0.036 0.014 0.064 -0.159 -0.199 -0.106 -277.415 -144.285 -133.130

(0.030) (0.032) (0.025)** (0.041)*** (0.046)*** (0.034)*** (57.051)*** (27.769)*** (49.758)***

Observations 85,440 85,440 85,440 46,320 46,320 46,320 85,440 85,440 85,440

Number of Municipalities 712 712 712 386 386 386 712 7 1 2 712

Year and Municipality FE Yes Yes Yes Yes Yes Yes Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon <100km <100km <100km Legal Amazon Legal Amazon Legal Amazon

Panel B.1: Falsiﬁcation test - Panel B.2: Alternative sample - Panel B.3: Alternative dependent variable -

subsample of Legal Amazon municipalities outside biome municipalities within 10 0km of biome frontier absolute values of rural credit

Total Cattle Crop Total Cattle Crop Total Cattle Crop

Biome * Non-Pronaf * Post2008 -0.096 -0.156 -0.047 -287.182 -186.752 -100.430

(0.057)* (0.060)*** (0.045) (57.198)*** (26.928)*** (50.230)**

Non-Pronaf * Post2008 0.064 0.116 -0.005 0.064 0.116 -0.005 237.473 158.671 78.801

(0.030)** (0.031)*** (0.025) (0.030 )* * (0.031)*** (0.025) (53.1 1 2)*** (21.711)*** (48.697)

Biome * Non-Pronaf 0.038 0.062 0.019 -626.448 -5.130 -621.319

(0.023)* (0.024)*** (0.018) (157.541)*** (26.126) (147.809)***

Biome * Post2008 -0.060 -0.057 -0.021 4.883 21.233 -16.350

(0.043) (0.046) (0.034) (12.823) (9.166)** (8.443)*

Non-Pronaf -0.025 -0.046 0.002 -0.025 -0.046 0.002 876 .9 7 5 159.534 717.441

(0.012)** (0.012)*** (0.010) (0.012)** (0.012)*** (0.010) (154.199)*** (20.239)*** (145.679)***

Observations 61,200 61,200 61,200 92,640 92,640 92,640 170,880 170,880 170,880

Number of Municipalities 712 712 712 386 386 386 712 712 712

Year and Municipality FE Yes Yes Yes Yes Yes Yes Yes Yes Yes

Sample Outside Biome Outside Biome Outside Biome <100km <100km <100km Legal Amazon Legal Amazon Legal Amazon

Notes: Coeﬃcients shown are estimated using ﬁxed eﬀects speciﬁcations in a municipality-by-month panel data set covering the 2002 through 2011 period. The sample includes all Legal Amazon municipalities

that are not crossed by the Amazon Biome frontier, unless otherwise stated. Falsiﬁcation tests are performed using normalized Pronaf rural credit as the dependent variable (Panel A.1) and a subsample of Legal

Amazon municipalities located outside the Amazon Biome (Panel B.1). Alternative speciﬁcations tested include using a restricted sample of municipalities within 100 km of biome frontier (Panels A.2 and B.2) and

absolute value of rural credit as the dependent variables (Panels A.3 and B.3). The dependent variables are calculated using all contracts (columns 1, 4 and 7), cattle-speciﬁc contracts only (columns 2, 5 and 8),

and crop-speciﬁc contracts only (columns 3, 6 and 9). All regressions include controls for municipality ﬁxed eﬀects, month-year ﬁxed eﬀects, cattle prices (current and lagged), crop prices (current and lagged), and

priority municipality status. Robust standard errors are clustered at the municipality level. Signiﬁcance: *** p<0.01, ** p<0.05, * p<0.10.

Table 7: The Eﬀect of Resolution 3,545 on Deforestation in the Amazon Biome (ﬁxed eﬀects and

2SLS regressions)

Panel A: Fixed eﬀects and IV second stage

Fixed eﬀects IV: Biome * Post IV: triple interaction

Total rural credit 0.026 0.700 0.148

(0.013)** (0.238)*** (0.036)***

Observations 5,750 5,750 10,571

Number of Municipalities 575 575 575

Year and Municipality FE Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon

Panel B: IV ﬁrst stage

IV: Biome * Post IV: triple interaction

Biome * Post2009 -0.385 0.126

(0.059)*** (0.061)**

Non-Pronaf * Post2009 0.196

(0.074)***

Biome * Post2009 * Non-Pronaf -0.949

(0.085)***

Observations 5,750 10,571

Number of Municipalities 575 575 575

Year and Municipality FE Yes Yes

Sample Legal Amazon Legal Amazon

Notes: Coeﬃcients shown are estimated using ﬁxed eﬀects speciﬁcations in a municipality-by-year panel

data set covering the 2002 through 2011 period. The sample includes all Legal Amazon municipalities

that are not crossed by the Amazon Biome frontier and that showed variation in forest cover during the

period. Fixed eﬀects OLS and second-stage 2SLS estimations are shown in Panel A, and ﬁrst-stage 2SLS

estimations are shown in Panel B. IV models instrument rural credit by the policy variable from model 13

(column 2) or model 14 (column 3). All regressions include controls for municipality ﬁxed eﬀects, year ﬁxed

eﬀects, cattle prices (current and lagged), crop prices (current and lagged), priority municipality status,

clouds and non-observable areas during period of remote sensing. Robust standard errors are clustered at

the municipality level. Signiﬁcance: *** p<0.01, ** p<0.05, * p<0.10.

Table 8: Counterfactual Simulations - The Eﬀect of Resolution 3,545 on Deforestation in the

Amazon Biome

Year

IV: Biome * Post IV: triple interaction

Observed Estimated Diﬀerence Observed Estimated Diﬀerence

2002 21,549 21,549 18,810

2003 25,686 25,686 22,844

2004 23,087 23,087 19,176

2005 20,087 20,087 19,847

2006 9,946 9,946 10,717

2007 10,565 10,565 10,825

2008 11,295 11,295 12,252

2009 5,220 5,688 468 5,220 6,446 1,226

2010 5,657 7,398 1,741 5,657 7,390 1734

2011 5,119 5,693 574 5,119 5,219 101

2009-2011 Total 15,995 18,778 2,783 15,995 19,056 3,061

Notes: Counterfactual simulations are conducted using the sample, speciﬁcations and estimated co-

eﬃcients from Table 7. Instruments used are Biome * Post (columns 1-3) and Biome * Post, Non-

Pronaf * Post, and Non-Pronaf * Biome * Post (columns 4-6). Observed deforestation is obtained

from PRODES/INPE and estimated deforestation is calculated by replacing the coeﬃcient of the policy

variable by zero.

Table 9: The Eﬀect of Resolution 3,545 on Defo restation in the Amazon Biome Using Cattle and Crop-Oriented Subsamples (ﬁxed eﬀects and 2SLS

regressions)

Panel A: Fixed eﬀects and IV second stage

Panel A.1: Cattle-oriented municipalities Panel A.2: Crop-oriented municipalities

Fixed eﬀects IV: Biome*Post IV: triple interaction Fixed eﬀects IV: Biome*Post IV: triple interaction

Total rural credit 0.023 1.033 0.225 0.040 -0.212 -0.027

(0.015) (0.334)*** (0.051)*** (0.026) (0.308) (0.034)

Observations 4,120 4,120 7,904 1,410 1,410 2,593

Number of Municipalities 412 412 412 141 141 141

Year and Municipality FE Yes Yes Yes Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon Legal Amazon Legal Amazon Legal Amazon

Panel B: IV ﬁrst stage

Panel B.1: Cattle-oriented municipalities Panel B.2: Crop-oriented municipalities

IV: Biome*Post IV: triple interaction IV: Biome*Post IV: triple interaction

Biome * Post2009 -0.389 0.161 -0.353 0.041

(0.067)*** (0.069)** (0.129)*** (0.130)

Non-Pronaf * Post2009 0.279 -0.084

(0.083)*** (0.159)

Biome * Post2009 * Non-Pronaf -0.954 -0.894

(0.096)*** (0.180)***

Observations 4,120 7,904 1,410 2,593

Number of Municipalities 412 412 141 141

Year and Municipality FE Yes Yes Yes Yes

Sample Legal Amazon Legal Amazon Legal Amazon Legal Amazon

Notes: Coeﬃcients shown are estimated using ﬁxed eﬀects speciﬁcations in a municipality-by-year panel data set covering the 2002 through 2011 period. The sample

includes all Legal Amazon municipalities that are not crossed by the Amazon Biome frontier and that showed variation in forest cover during the period. Samples are

restricted to cattle (Panels A.1 and B.1) and crop-oriented (Panels A.2 and B.2) municipalities. Fixed eﬀects OLS and second-stage 2SLS estimations are shown in Panel

A, and ﬁrst-stage 2SLS estimations are shown in Panel B. IV models instrument rural credit by the p olicy variable from model 13 (columns 2 and 5) or model 14 (columns

3 and 6). All regressions include controls for municipality ﬁxed eﬀects, year ﬁxed eﬀects, cattle prices (current and lagged), crop prices (current and lagged), priority

municipality status, clouds and non-observable areas during period of remote sensing. Robust standard errors are clustered at the municipality level. Signiﬁcance: ***

p<0.01, ** p<0.05, * p<0.10.

Table 10: Robustness Checks - IV Regressions with Alternative Sample, Alternative Dependent Variable and Controlling for the Interaction Between a Non-Linear Trend and Initial Deforestation

Panel A - IV Seco nd Stage

Panel A.1: Alternative sample - Panel A.2: Alternative variables - Panel A.3: Controlling for the interaction

municipalities within 100 km of biome frontier absolute values of total rural credit and deforestation between a non-linear trend and initial deforestation

IV-Biome*Post IV- Triple interaction IV-Biome*Post IV- Triple interaction IV-Biome*Post IV- Triple interaction

Total Rural Credit 1.514 1.242 3.621 0.131 0.619 0.127

(0.602)** (0.413)*** (1.300)*** (0.039)*** (0.224)*** (0.034)***

Observations 2,510 4,868 5,750 10,610 5,750 10,571

Number of Municipalities 251 251 575 575 575 575

Year and Municipality FE Yes Yes Yes Yes Yes Yes

Sample <100km <100km Legal Amazon Legal Amazon Legal Amazon Legal Amazon

Panel B - IV First Stage

Panel B.1: Alternative sample - Panel B.2: Alternative variables - Panel B.3: Controlling for the interaction

municipalities within 100 km of biome frontier absolute values of total rural credit and deforestation between a non-linear trend and initial deforestation

IV- Biome*Post IV- Triple interaction IV- Biome*Post IV- Triple interaction IV- Biome*Post IV- Triple interaction

Biome * Post2009 -0.29 -0.081 -3.6 67 0.015 -0.376 0.131

(0.072)*** (0.076) (0.638)*** (0.833) (0.081)*** (0.085)

Non-Pronaf * Post 2009 0.196 2.594 0.196

(0.071)*** (1.014)** (0.092)**

Biome * Post2009 * N on-Pronaf -0.283 -3.615 -0.949

(0.105)*** (1.169)*** (0.111)***

Observations 2,510 4,868 5,750 10,610 5,750 10,571

Number of Municipalities 251 251 575 575 575 575

Year and Municipality FE Yes Yes Yes Yes Yes Yes

Sample <100km <100km Legal Amazon Legal Amazon Legal Amazon Legal Amazon

Notes: Co eﬃcients shown are estimated using ﬁxed eﬀects sp eciﬁcations in a municipality-by-month panel data set covering the 2002 through 2011 period. Second-stage 2SLS estimations are shown in Panel A,

and ﬁrst-stage 2SLS estimations are shown in Panel B. IV models instrument rural credit by the policy variable from model 13 (columns 1, 3 and 5) or model 14 (columns 2, 4 and 6). Alternative speciﬁcations

tested include using a restricted sample of municipalities within 100 km of biome frontier (Panels A.1 and B.1); absolute value (instead of normalized value) of rural credit and deforestation variables (Panels A.2 and

B.2); and controlling for the interaction between deforestation initial values and a non-linear trend (Panels A.3 and B.3). All regressions include controls for municipality ﬁxed eﬀects, year ﬁxed eﬀects, cattle prices

(current and lagged), crop prices (current and lagged), priority municipality status, clouds and non-observable areas during period of remote sensing. Robust standard errors are clustered at the municipality level.

Signiﬁcance: *** p< 0.01, ** p<0.05, * p< 0.10.