Limits of Brazil’s Forest Code as a means to end

illegal deforestation
Andrea A. Azevedoa, Raoni Rajãob, Marcelo A. Costab, Marcelo C. C. Stabilea,1, Marcia N. Macedoa,c, Tiago N. P. dos Reisa,
Ane Alencara, Britaldo S. Soares-Filhod, and Rayane Pachecob
a
Instituto de Pesquisa Ambiental da Amazônia, Lago Norte, Brasilia, DF 71503-505, Brazil; bLaboratório de Gestão de Serviços Ambientais, Universidade
Federal de Minas Gerais, 6627-Pampulha, Belo Horizonte, MG 31270-901, Brazil; cWoods Hole Research Center, Falmouth, MA 02450; and dCentro de
Sensoriamento Remoto, Universidade Federal de Minas Gerais, 6627-Pampulha, Belo Horizonte, MG 31270-901, Brazil

Edited by Emilio F. Moran, Michigan State University, East Lansing, MI, and approved May 24, 2017 (received for review March 23, 2016)

The 2012 Brazilian Forest Code governs the fate of forests and Environmental Registry (CAR, Portuguese acronym) in 2008 (MT)
savannas on Brazil’s 394 Mha of privately owned lands. The govern- and 2009 (PA). To join CAR, landowners must georeference their
ment claims that a new national land registry (SICAR), introduced un- property boundaries and remaining forests using satellite images (Fig.
der the revised law, could end illegal deforestation by greatly reducing 1) (17, 18). For the first time, CAR made it possible for government
the cost of monitoring, enforcement, and compliance. This study eval- agencies to identify the perpetrators of deforestation and monitor
uates that potential, using data from state-level land registries (CAR) in whether individual landowners were complying with the Forest Code.
Pará and Mato Grosso that were precursors of SICAR. Using geospatial These state land registries served as models for the National Rural
analyses and stakeholder interviews, we quantify the impact of CAR Environmental Registry System (SICAR), which today is the main
on deforestation and forest restoration, investigating how landowners instrument for implementing the new Forest Code. The 2012 Forest
adjust their behaviors over time. Our results indicate rapid adoption of Code stipulates that landowners in the Amazon biome should con-
CAR, with registered properties covering a total of 57 Mha by 2013. serve 80% of their property (land area) in native vegetation, whereas
This suggests that the financial incentives to join CAR currently exceed
those in the Cerrado should conserve 20–35% (19).
the costs. Registered properties initially showed lower deforestation
SICAR aimed to register roughly 5,000,000 rural properties
rates than unregistered ones, but these differences varied by property
throughout Brazil by May 2016. This target date was postponed to
size and diminished over time. Moreover, only 6% of registered pro-
December 2017 by Law No. 13.295 on June 14, 2016 (20). By
ducers reported taking steps to restore illegally cleared areas on their
August 2016, it had registered 3,700,000 properties spanning
properties. Our results suggest that, from the landowner’s perspective,
full compliance with the Forest Code offers few economic benefits.
387 Mha (21). The GIS-based environmental registry promises to

ENVIRONMENTAL
Achieving zero illegal deforestation in this context would require the make landowners accountable for illegal deforestation and resto-
ration requirements, while reducing the cost of monitoring for the

SCIENCES
private sector to include full compliance as a market criterion, while
state and federal governments develop SICAR as a de facto enforce- government, landowners, and the private sector (22). Commodities
ment mechanism. These results are relevant to other tropical countries buyers currently face high monitoring and transaction costs to en-
and underscore the importance of developing a policy mix that creates sure deforestation-free supply chains (e.g., the soy moratorium)
lasting incentives for sustainable land-use practices. (23). If the national CAR system were fully implemented—together
with complementary public policies—it has the potential to replace
deforestation | Forest Code | tropical forests | governance | Amazon these initiatives, reduce deforestation, and lower costs (11, 24).

SUSTAINABILITY
Although an important first step, registering with CAR does

SCIENCE
not guarantee that landowners will comply with the law or reduce
H istorically, deforestation has accounted for the majority of
greenhouse gas (GHG) emissions from developing countries
(1, 2). In Brazil, this trend changed dramatically when annual
deforestation. Full compliance involves very high restoration

deforestation rates in the Amazon dropped by 76% from 2005 to Significance

2012 (3–5). Avoided deforestation during this period generated
emissions reductions on the order of 3.2 Gt CO2, compared with a Brazil’s new Forest Code has the potential to halt illegal de-
historical baseline (5–7). There are several potential explanations for forestation in the country’s native forests and savannas through
the observed decline in deforestation. These include the establish- implementation of a federal land registry—along with powerful
ment of new protected areas (7), restrictions on credit available to tools that facilitate enforcement and give landowners a pathway to
illegal deforesters (8, 9), public blacklists of properties and munici- restoring or compensating their “forest deficits.” This study suggests
palities that deforest illegally (10), moratoria to eliminate deforesters that these tools fall short of their promise. Although landowners in
from soy and beef supply chains (5, 11), and command-and-control eastern Amazonia have been motivated to join state land registries,
enforcement actions by state and federal agencies (12–15). many continue to deforest and few have restored their illegally
Despite advances, Brazil still faces two key barriers to effective cleared areas. Results indicate that the economic benefits of full
enforcement of deforestation. First, the lack of a comprehensive compliance with the Forest Code remain scant. To end deforestation,
national database of property boundaries (i.e., a land registry) has Brazil must realign its financial and policy incentives to encourage
made it difficult to link new deforestation to specific land owners. this outcome. The fate of the country’s forests hangs in the balance.
Second, deforestation patches have decreased in size, making
Author contributions: A.A.A., R.R., M.A.C., M.C.C.S., and A.A. designed research; A.A.A.,
them increasingly difficult to detect (16). Both pose substantial R.R., M.A.C., and M.C.C.S. performed research; B.S.S.-F. contributed new analytic tools;
challenges for forest monitoring, effective enforcement, and res- R.R., A.A., and R.P. collected data in the field; A.A.A., R.R., M.A.C., M.C.C.S., and T.N.P.R.
toration of illegally deforested areas (i.e., forest “deficits”) man- analyzed data; and A.A.A., R.R., M.A.C., M.C.C.S., and M.N.M. wrote the paper.
dated by the Forest Code. This is illustrated by the fact that the The authors declare no conflict of interest.
majority (∼69%) of deforestation from 2002 to 2009 occurred on This article is a PNAS Direct Submission.
properties whose boundaries were not publicly registered. Freely available online through the PNAS open access option.
In the face of these difficulties, the Amazon states of Mato Grosso 1
To whom correspondence should be addressed. Email: marcelo.stabile@ipam.org.br.
(MT) and Pará (PA) invested in systems to control and monitor This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
deforestation, implementing a land registry known as the Rural 1073/pnas.1604768114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1604768114 PNAS | July 18, 2017 | vol. 114 | no. 29 | 7653–7658

 status of CAR properties and estimate the economic costs of being
compliant (19, 32–34). We also used primary data from question-
naires with farmers and GIS professionals to evaluate costs and
benefits at each stage of Forest Code compliance, using a sample of
20 municipalities and 33 in-depth interviews with state officials.
Results and Discussion
Incentives for Joining CAR. By 2013, registered CAR properties
covered roughly 32% (23 Mha) of the areas eligible for regis-
tration in Mato Grosso and 57% (34 Mha) in Pará. Registered
properties were distributed uniformly in both states, suggesting
that a broad cross section of producers have joined (Fig. 2).
Among the surveyed producers outside CAR, 30% in Pará and
36% in Mato Grosso declared that they would join only if forced
to by government or market sanctions.
Both the rapid adoption of CAR and data from our ques-
tionnaires suggest that the incentives to join CAR outweighed
the costs of remaining outside the system. The most immediate
benefit of joining was a lower chance of receiving fines for not
complying with state laws in Mato Grosso and Pará, where CAR
Fig. 1. Example of a CAR (Rural Environmental Registry) property registered membership is mandatory for all rural properties. To encourage
according to the national registry standard (SICAR). The property boundary is adherence to the system, state officials reported having ignored
shown by the dashed yellow line. Legal Reserves are designated by the green legal infractions within CAR properties to avoid “scaring off”
hatching, whereas Legal Reserve deficits are marked by red hatching. Buffer zones new registrants from joining the system.
around rivers and dams (areas of permanent protection) are included in the Legal
A second (and likely stronger) incentive to join CAR was access to
Reserve area. Blue hatched areas represent agriculture and cattle-ranching areas.
additional lines of credit for farmers. Resolution No. 3545/2008 of
Brazil’s Central Bank made it mandatory for producers to present a
costs, opportunity costs of foregone production, and negligible “license, certificate, or equivalent evidence of environmental com-
benefits, given the relatively low risk of receiving fines due to pliance” to qualify for public loans (35). Because public loans are
poor enforcement. This also reflects a lack of market demand for Brazil’s main instrument for subsidizing the agricultural sector, their
legality as a criterion for purchase of commodities (24). interest rates are much lower than those of private banks (28).
Despite the great potential of public land registries, few studies The third incentive to join CAR stemmed from the intervention
have quantified their effects on deforestation in the Amazon (but of public prosecutors. To control growing deforestation rates in
see 17, 25, 26) or their role in ensuring compliance on private Pará, in 2009 public prosecutors pressed the state’s large slaughter-
properties. Studies of other deforestation-control measures— houses to stop buying cattle from ranches that did not comply with
including payments for environmental services (27, 28) and pro- environmental and labor laws. That same year, Greenpeace proposed
tected areas (29, 30)— suggest that these programs do not always an agreement urging the Amazon’s four biggest slaughterhouses to
yield the expected conservation outcomes. The effectiveness of these boycott cattle from ranches with illegal deforestation after July 2009
policies in improving forest governance in the tropics remains an
open question, including CAR, which has yet to be fully imple-
mented. To address this gap, we analyzed the recent experiences of
Mato Grosso and Pará, with the goal of improving implementation
of SICAR in Brazil and similar systems in other countries.
This study addresses three central questions: (i) What motivates
producers to join CAR? (ii) Are registered producers less likely to
deforest? (iii) Are registered producers more likely to comply with
Forest Code restoration requirements? These questions are essential
to understanding how individual farmers perceive the incentives at
each stage of CAR implementation. To address them, we evaluated
costs and benefits of environmental compliance from the producer’s
point of view. We quantified deforestation in 49,669 rural properties
in Mato Grosso and Pará that joined CAR from 2008 to 2013. The
control group included properties before CAR registration, whereas
the treatment group included properties after registration (31) (SI
Appendix). To control for exogenous factors (other than CAR) that
might influence deforestation, we evaluated a series of models in-
cluding several potential explanatory variables. Deforestation prob-
ability was modeled based on distance to markets, infrastructure,
agricultural suitability, and slope (7). Forest patch size represented
the supply of forests available for deforestation in a given property.
We also assessed the impact of public policies such as the Green
Municipalities Program (GMP) in Pará (a federal blacklist restricting
credit to municipalities with high rates of illegal deforestation) and Fig. 2. Spatial distribution of properties enrolled in CAR (Rural Environ-
the number of fines related to environmental infractions over time. mental Registry) by 2013, grouped according to their size class. Properties in
To estimate the incentives to comply with the Forest Code, we yellow are small properties of up to four Rural Modules (<4 RM); orange are
used secondary data and published studies to evaluate the legal medium-sized properties (4–15 RM); and red are large properties (> 15 RM).

7654 | www.pnas.org/cgi/doi/10.1073/pnas.1604768114 Azevedo et al.

(36). These initiatives required cattle ranchers in both states to join and explanatory variables (Methods and SI Appendix, Table S5).
CAR to sell their product to large slaughterhouses. The final model included deforestation probability (7), remain-
Governmental and nongovernmental organizations provided a ing forest area, and whether the municipality was part of a fed-
fourth incentive by subsidizing the GIS surveys needed to register eral blacklist to combat high deforestation (37).
with CAR. Surveys of GIS professionals indicate that the cost of Our results indicate that registering with CAR did not necessarily
joining CAR averages US $549 for small and US $1686 for large reduce illegal deforestation. Despite controlling for other spatial,
properties in Mato Grosso, compared with US $307 for small and economic, and policy factors, we observed substantial variation in the
US $845 for large properties in Pará. Estimates are based on the effectiveness of CAR over time and across property sizes (Table 1 and
2013 average exchange rate between the Brazilian Real and US SI Appendix, Fig. S1). Small properties (<400 ha) in Mato Grosso and
Dollar. These upfront costs represent half the monthly income for Pará had lower deforestation immediately after entering CAR, but this
some small farmers, making them a significant barrier to entering effect decreased over time and, in the case of Pará, disappeared en-
CAR. NGO programs to cover these costs are an important in- tirely by 2012. Medium and large properties in both states showed no
centive to join and exist in at least 64 municipalities where such work consistent pattern. For instance, in medium properties (400–1,500 ha)
is undertaken by the following organizations: The Nature Conser- in Mato Grosso, deforestation was higher inside CAR for 2009–2010,
vancy (TNC), Instituto Socioambiental (ISA), Instituto do Homem e but lower for 2011. The inverse was true for large properties
Meio Ambiente da Amazônia (IMAZON), Instituto Centro de Vida (>1500 ha) in Mato Grosso, where deforestation was lower inside
(ICV), and Instituto de Pesquisa Ambiental da Amazônia (IPAM). CAR in 2009, but higher in 2011 (Table 1 and SI Appendix, Fig. S1).
Since its introduction, CAR has shifted from an instrument focused Including the effect of the municipalities blacklist improved our
exclusively on environmental sustainability to one that is vital for the model, but other public policies had no clear effect. Including Pará’s
economic sustainability of rural producers. Our results suggest that green municipalities program (38) and the number of fines at the
subsidies to decrease the cost of entering CAR, combined with credit municipal level issued by IBAMA (Instituto Brasileiro do Meio
and market restrictions that increase the costs of production outside Ambiente e dos Recursos Naturais Renováveis) (13, 15) did not
CAR, have made it relatively costly to remain outside the registry. change the general conclusions described above and, in some cases,
reduced the overall predictive power of the models. This suggests
Deforestation Within CAR. A key assumption of CAR supporters at that these policies either (i) did not reduce deforestation beyond the
the federal level is that registering rural properties in the system CAR effect or (ii) covaried with CAR at the property level.
will substantially decrease illegal deforestation. This is rooted in Interviews with local farmers in Pará and Mato Grosso confirm
the idea that the land registry could radically reduce the cost of the results presented above. Some small farmers reported feeling
property-level monitoring and enforcement. To understand how like they were being watched more closely by the state after joining
CAR affected deforestation decisions, we compared annual de- CAR, supporting the idea that CAR could lower monitoring costs

ENVIRONMENTAL
forestation rates in registered and unregistered properties, and improve enforcement. However, this initial perception of risk

SCIENCES
stratifying by property size. To control for other factors that has decreased over time—in some cases to the point where the
might influence deforestation, we considered a series of models benefits of increasing deforestation (e.g., increased land value)

Table 1. Average property-level deforestation (ha) within CAR and Control groups
Size Class (RM) Year Before CAR (Control) After CAR CAR effect (%) CAR effect (P value)

SUSTAINABILITY
Mato Grosso
Up to 4 RM 2009 0.0772 0.0556 −3.25% 0.6513

SCIENCE
2010 0.1293 0.0374 −7.21% 0.0051
2011 0.1925 0.0987 −6.12% 0.6445
4–15 RM 2009 0.0949 0.1253 7.41% 0.4346
2010 0.1204 0.1182 8.50% 0.0737
2011 0.2707 0.1323 −4.93% 0.8112
Over 15 RM 2009 0.1780 0.0268 −10.12% 0.2944
2010 0.2206 0.1609 0.21% 0.9700
2011 0.1733 0.2294 10.75% 0.6589
Pará
Up to 4 RM 2008 0.4430 0.1702 −27.47% 0.0108
2009 0.3133 0.0803 −21.34% 0.0000
2010 0.3573 0.2653 −10.89% 0.0000
2011 0.2596 0.2200 −5.29% 0.0000
2012 0.1836 0.1592 −3.19% 0.2785
4–15 RM 2008 0.8486 0.0277 −34.71% 0.0956
2009 0.5589 0.4379 18.54% 0.0177
2010 0.5400 0.3460 −4.72% 0.1933
2011 0.3174 0.2612 0.96% 0.7498
2012 0.3616 0.1728 −14.61% 0.0173
Over 15 RM 2008 0.9885 0.9923 13.78% 0.5111
2009 0.5416 0.5416 21.37% 0.0079
2010 0.6480 0.4546 −8.13% 0.0416
2011 0.4379 0.3288 −2.91% 0.4065
2012 0.1614 0.2380 14.09% 0.1277

The estimated CAR effect (model 3) is adjusted for forest size, an index of deforestation risk (developed using
the Dinamica EGO modeling platform), and presence of the blacklist within the municipality. Bold numbers
indicate P values that are significant (P ≤ 0.1).

Azevedo et al. PNAS | July 18, 2017 | vol. 114 | no. 29 | 7655
outweigh the potential costs (e.g., fines). Some farmers confessed to 3,951,664 ha to be restored in Mato Grosso and Pará, respectively
clearing small areas (<10 ha) on their properties, hoping that this (SI Appendix, Tables S1 and S2). Considering that restoration costs
small-scale deforestation would escape detection by satellites or be range from US $536 to 1,327 ha−1, depending on the property’s
overlooked by state prosecutors. Satellite observations confirmed land-use history and adjacent land uses (19, 32), we estimated a
that a large proportion (63% for PA and 51% for MT) of clearings total restoration cost from US $0.5 to 1.1 billion in Mato Grosso
inside CAR were smaller than 10 ha. Recent studies indicate that and from US $2.1 to 5.2 billion in Pará.
this decrease in the size of deforestation patches is widespread (16, There are also substantial opportunity costs associated with
39). Officials from both federal and state agencies confirmed that, (i) forgoing production on a given land parcel to begin restoration
in practice, small clearings are systematically ignored due to the and (ii) maintaining surplus Legal Reserves—i.e., forest assets
logistical difficulty of inspecting deforestation events in situ. Al- that could legally be converted for production. Stickler et al. (32)
though federal and state environmental agencies have started to use estimated that the first incurred a cost of US $673 ha−1, and the
CAR data to issue fines remotely, officials report that this requires latter incurred a cost of US $500 ha−1. Combining these figures
substantial labor and that personnel limitations make it impractical with our sample, we estimate that the total opportunity cost of
to prosecute small deforestation events. This suggests that most forgoing production for restoration in 2008 was about US
landowners deforesting within CAR do so with the expectation of $0.5 billion in Mato Grosso and US $2.6 billion in Pará. The
impunity because small deforestation patches are not being de- costs of maintaining surplus forests were excluded from our es-
tected or prosecuted by the control agencies. timate of total compliance costs because they reflected both di-
rect and indirect costs of restoration, as described below.
Compliance with Forest Restoration Requirements Within CAR. The combined direct cost of restoration and opportunity cost
Attaining zero illegal deforestation within CAR is an important of forgone production ranges from US $1.0 to 1.6 billion in Mato
target, but is not enough to guarantee Forest Code compliance. The Grosso (2008) and from US $4.7 to 7.9 billion in Pará (2007),
law requires landowners who have deforested illegally to restore or considering a sample area of 57.2% in PA and 31.7% in MT.
compensate these clearings to fulfill the minimum Legal Reserve Considering the total productive area (In our sample, 1.3 Mha in
requirement (19). To weigh the costs and benefits of complying, the Mato Grosso and 4.3 Mha in Pará are productive lands), we
farmer must consider (i) incentives reserved for farmers that are estimate that the average cost of Forest Code compliance ranges
fully compliant, (ii) the cost of forest restoration or compensation, from US $768 to 1,270 ha−1 in Mato Grosso and from US
(iii) the opportunity cost of foregone rents from agricultural pro- $1,099 to 1,818 ha−1 in Pará. Although the cost of compliance
duction, (iv) the potential for future changes in the law, and (v) the can be reduced significantly through compensation mechanisms
probability of getting caught and punished for noncompliance. such as the Environmental Reserve Quota (CRA), in most cases
Using our sample and the survey data, we assessed the influence of this remains prohibitively expensive (41). For this reason, the
most of these factors on producer decisions to maintain or restore level of Forest Code implementation in Brazil is low (34), and
Legal Reserves and riparian areas on their properties. the forest debt in states like Mato Grosso is massive (19, 24).
Incentives for compliance. At the moment, the economic benefits of full Changes in the Forest Code. Revisions to the Forest Code have created
compliance with the Forest Code are scant. Officials from both a substantial disincentive for compliance. The latest of these occurred
states report that compliance with these obligations is rarely verified in 2012, with approval of a new Forest Code that lowered standards
on the ground. Farmers need only present a report stating that they for environmental compliance. The 2012 Forest Code not only for-
have taken steps to restore their forest debts, but only a fraction of gave fines for areas deforested illegally before 2008, but also reduced
CAR participants provide these reports on a regular basis. restoration requirements. The revised law decreased the total area
Results from the questionnaires corroborate these findings. Only requiring restoration by 41% and 68% in Mato Grosso and Pará,
6% of landowners with forest debts in Pará and Mato Grosso respectively (19). Considering only properties inside CAR, the land
reported that they were taking the necessary measures to compensate area to be restored dropped by 21% in Mato Grosso and 15% in
or restore their Legal Reserves, whereas 76% affirmed that they Pará. These reductions affected 55% of the properties in Mato
would only compensate or restore if coerced to do so through gov- Grosso and 70% in Pará—primarily due to changes in the rules for
ernment fines or market incentives. Even faced with a scenario in smallholders. The only prerequisite for this benefit was to join CAR
which strong restrictions were imposed by private and public actors, and commit to an official management plan (Portuguese acronym,
18% said they would never compensate or restore their forest debts. PRA) to achieve environmental compliance. The new Forest Code
Aside from a lower probability of receiving fines, the only economic thus provided substantial economic payoffs to producers who defor-
incentive currently applicable to forest restoration is a 15% increase ested illegally before 2008, while punishing those that refrained from
in the total amount of subsidized loans available to farmers who can clearing or invested in forest restoration to comply with the law. The
demonstrate a commitment to full compliance with the Forest Code amnesty provided by the new Forest Code increased the perceived
(40). No market initiative targets the forest debts of individual risk of compliance by setting a precedent that future changes in the
farmers under the Forest Code; they focus instead on eliminating law might benefit farmers who deforest illegally.
newly deforested areas from commodity supply chains (11, 24). From
a market perspective, there is still no difference between a landowner Cost–Benefit Analysis of CAR Compliance. The empirical data pre-
with an 80% Legal Reserve (compliant) and one with only 2% sented here suggest four stages of compliance (Fig. 3): (i) outside
(noncompliant). Nevertheless, compliant and noncompliant land- CAR (BAU, business as usual), (ii) joining CAR (GOV1, gover-
owners will obtain very different economic returns and environ- nance 1), (iii) inside CAR and reducing deforestation (GOV2), and
mental outcomes from properties of the same size. (iv) inside CAR and fully compliant with the Forest Code (GOV3).
Costs of compliance. The economic benefits of fully complying with the From the farmer’s point of view, each stage carries potential costs
Forest Code are very low, whereas the costs are substantial. Illegally and benefits that may or may not provide incentives to follow the
deforested areas provide a sizable portion of the income of Amazon rules. Our results suggest that there is a clear incentive for land-
farmers. In addition to forgoing this income, farmers are faced with owners to join CAR. This move from business as usual (BAU) to
the costs of restoration, which may be high depending on the the first stage of governance (GOV1) has a relatively low trans-
method used. To estimate the potential costs, we first quantified the action cost; a minimal increase in the risk of being fined; and sub-
environmental deficit in our sampled properties. We found that stantial financial benefits, such as access to subsidized loans (Fig. 3).
2,944 (82.6%) properties in Mato Grosso and 15,170 (76.6%) The incentives are less clear when we consider the transition
properties in Pará were not compliant with the Forest Code before from joining CAR (GOV1) to stopping illegal deforestation
joining the CAR system. This represented 841,564 ha and (GOV2) and restoring illegally cleared areas (GOV3). Producers

7656 | www.pnas.org/cgi/doi/10.1073/pnas.1604768114 Azevedo et al.

who do not deforest earn less by not expanding agriculture, but landowners within CAR act as a safeguard for registered producers
may benefit from fewer fines and access to green markets (11, 23). who continue deforesting. The resulting perception of impunity
Our finding that some CAR properties had lower deforestation severely weakens environmental policies to control deforestation.
than the control group suggests that the perceived financial risks CAR’s biggest potential stems from the fact that it drastically
outweighed the benefits of deforesting. On the other hand, the fact reduces the cost of monitoring and enforcement, but these savings
that many producers maintained or increased deforestation after have yet to materialize in Mato Grosso and Pará. To remedy this,
joining CAR suggests that the incentives to avoid deforestation public and private actors will need to shift the costs and benefits
vary in space and time (Fig. 3 and Table 1). Finally, our results related to each of the four stages of compliance outlined above. At
indicate that landowners in all size classes are unlikely to invest in a minimum, the government must increase the likelihood of
forest restoration (GOV3) under current conditions. Because most prosecution of illegal deforesters. To accomplish this, the Ministry
of the benefits can be accrued by joining CAR, achieving full of Environment could use SICAR to develop mechanisms to au-
compliance would require additional government or market in- tomatically detect illegal deforestation, identify the responsible
terventions to realign the incentives for Forest Code compliance. parties, and levy fines. Restoration agreements signed by pro-
ducers registered with CAR should also be monitored and eval-
Conclusion and Policy Considerations uated using a combination of remote-sensing technologies and
The CAR system will play an increasingly central role in the field sampling to increase compliance.
implementation of the Forest Code and climate policy in Brazil On the market side, public and private actors must increase
(19). This study shows that credit and market restrictions provide the benefits of complying with the Forest Code beyond reducing
strong incentives for producers to join CAR. However, results the risk of fines. The 2012 Forest Code presents an opportunity
suggest that its implementation has not contributed significantly to to do this by creating new market mechanisms that allow land-
the observed reductions in deforestation from 2008 to 2012. Fur- owners with forest surpluses to trade with farmers that need to
thermore, the cost of restoring Legal Reserves and riparian areas compensate their forest debts (41, 42). This offset mechanism
remains prohibitively high relative to the benefits of joining CAR. can be used to avoid legal deforestation and provide incentives to
This study demonstrates that CAR membership does not yet restore forests in highly degraded areas, particularly if integrated
provide the full suite of financial incentives (or command-and- into the Brazilian REDD+ strategy and the Amazon Fund (43).
control disincentives) needed to prevent deforestation and ensure New sustainable supply-chain initiatives (e.g., for beef and soy)
full compliance with Forest Code restoration requirements. The should strive to adopt more stringent environmental compliance
existence of incentives like the soy and beef moratoria has helped standards for purchase from industry retailers. Aside from requiring
to inhibit deforestation, but no comparable incentives exist to CAR, companies could build a network of suppliers who use Forest
encourage restoration. Inconsistent monitoring and enforcement Code compliance as a criterion for purchasing products and pro-
and the reluctance of state and municipal managers to punish viding financial incentives (18, 24). This would ultimately increase

ENVIRONMENTAL
SCIENCES
SUSTAINABILITY
SCIENCE

Fig. 3. Theoretical cost–benefit curve of CAR. In the BAU scenario, the costs are higher than the benefits of being outside CAR. The inverse is true in scenarios GOV1 and
GOV2. The curves overlap again in GOV3, where the costs are higher than the benefits because of legal reserve restoration costs and possible reductions in productive area.

Azevedo et al. PNAS | July 18, 2017 | vol. 114 | no. 29 | 7657
awareness and trust by buyers throughout the supply chains, reduce the likelihood of deforestation decreases as forest becomes scarce; (ii) a dummy
the risk of contamination with noncompliant products, and lower variable to indicate whether the property was enrolled in CAR; (iii) an index of
the reputational risk for large national and international buyers. deforestation risk (SI Appendix) (7); (iv) a variable indicating whether the property
In theory, CAR can increase the government’s ability to mon- was in a blacklisted municipality (37); (v) the change in the number of fines in a
itor environmental performance, prosecute illegal deforestation, municipality (an indicator of enforcement); and (vi) a dummy variable indicating
and distribute the economic benefits of compliance. In practice, whether the municipality participated in Pará’s Green Municipality Program
(GMP). We developed a series of models, containing combinations of these
this potential has not yet been realized due to incomplete imple-
covariates, and compared their effectiveness in predicting deforestation at the
mentation of CAR and supporting public policies. Nevertheless,
property level (SI Appendix, S1. Methods and Tables S5–S14).
many commodity companies in the world have pledged zero de-
To estimate the incentives for full Forest Code compliance, we evaluated the
forestation (and illegality) within their supply chains by 2020. The
legal status of CAR properties (SI Appendix, Fig. S2 and Tables S1 and S2) and
experiences of CAR in Pará and Mato Grosso provide valuable
estimated the economic costs of compliance using secondary data and published
lessons that could help federal and state governments make studies (19, 32–34). We used qualitative methods to estimate the incentives for
SICAR a more effective instrument for ending illegal deforesta- the BAU and GOV institutional frameworks, administering questionnaires to
tion and promoting forest restoration. The lessons learned from 92 farmers and GIS professionals in 20 randomly selected municipalities in Mato
this study are relevant to the rest of Brazil and other tropical re- Grosso and Pará. We conducted 33 semistructured interviews to comprehend the
gions trying to balance food production and forest conservation. historical and political context of forest governance in the region. The authors
were responsible for discussing and approving the methods for the interviews, as
Methods well as for obtaining consent for publishing interview results. Details on the
We used a BACI (Before-After-Control-Impact) design to evaluate whether in- conceptual approach to this analysis are provided in SI Appendix.
centives to obey the law and reduce deforestation outweighed incentives to
deforest within CAR. We compared areas that had not yet joined CAR (control) ACKNOWLEDGMENTS. We thank Paulo Moutinho, Vivian Ribeiro, and two
with those that had (treatment) to quantify the influence of the policy in- anonymous reviewers for helpful comments on earlier drafts of this paper. We
tervention (31). We used ordinary least squares regression models to evaluate are also grateful to the interview participants for sharing their experiences and
the differences in deforestation rates between the “CAR” and “control” groups. key insights that made this work possible. Funding for this work came from the
The dependent variable was the logarithm of the deforestation rate plus one (to Climate and Land use Alliance, Gordon and Betty Moore Foundation, Norwegian
account for records with zero deforestation) (SI Appendix, S1. Methods). Agency for Development Cooperation (NORAD), Conselho Nacional de Desen-
We used different sets of covariates to test the importance of exogenous factors volvimento Científico e Tecnológico (CNPq), and Fundação de Amparo à Pesquisa
that might influence deforestation and confound our interpretation of CAR’s de Minas Gerais (FAPEMIG). Various institutions generously shared data, includ-
performance, including (i) the logarithm of the remaining forest area, because ing Secretaria Estadual de Meio Ambiente (SEMA) in Mato Grosso and Pará.

1. Angelsen A, et al. (2009) Realising REDD+: National Strategy and Policy Options 23. Gibbs HK, et al. (2015) Environment and development. Brazil’s soy moratorium.
(CIFOR, Bogor, Indonesia). Science 347:377–378.
2. Gibbs HK, Herold M (2007) Tropical deforestation and greenhouse gas emissions. 24. Azevedo AA, Stabile MCC, Reis TNP (2015) Commodity production in Brazil: Com-
Environ Res Lett 2:045021. bining zero deforestation and zero illegality. Elementa 3:000076.
3. Macedo MN, et al. (2012) Decoupling of deforestation and soy production in the 25. Azevedo AA, Saito CH (2013) Deforestations profile in Mato Grosso, after im-
southern Amazon during the late 2000s. Proc Natl Acad Sci USA 109:1341–1346. plementation of the environmental licensing in rural properties. Cerne 19:111–122.
4. INPE (2014) Satellite Monitoring of Brazil’s Amazon Forest (PRODES). Available at www. 26. L’Roe J, Rausch L, Munger J, Gibbs HK (2016) Mapping properties to monitor forests:
obt.inpe.br/prodes/. (Brazilian Natl Agency for Space Res, São José dos Campos, SP). Landholder response to a large environmental registration program in the Brazilian
5. Nepstad D, et al. (2014) Slowing Amazon deforestation through public policy and Amazon. Land Use Policy 57:193–203.
interventions in beef and soy supply chains. Science 344:1118–1123. 27. Alix-Garcia JM, Sims KRE, Yañez-Pagans P (2015) Only one tree from each seed? En-
6. Lapola DM, et al. (2014) Pervasive transition of the Brazilian land-use system. Nat Clim vironmental effectiveness and poverty alleviation in Mexico’s payments for ecosystem
Change 4:27–35. services program. Am Econ J Econ Policy 7:1–40.
7. Soares-Filho B, et al. (2010) Role of Brazilian Amazon protected areas in climate 28. Robalino J, Pfaff A (2013) Ecopayments and deforestation in Costa Rica: A nationwide
change mitigation. Proc Natl Acad Sci USA 107:10821–10826. analysis of PSA’s initial years. Land Econ 89(3):432–448.
8. Arima EY, Barreto P, Araújo E, Soares-Filho B (2014) Public policies can reduce tropical 29. Pfaff A, Robalino J, Sanchez-Azofeifa GA, Andam Kwaw S, Ferraro Paul J (July 13,
deforestation: Lessons and challenges from Brazil. Land Use Policy 41:465–473. 2009) Park location affects forest protection: Land characteristics cause differences in
9. Assunção J, Gandour C, Rocha R, Rocha R (2013) Does Credit Affect Deforestation? park impacts across Costa Rica. B.E. J Econ Anal Policy, 10.2202/1935-1682.1990.
Evidence from a Rural Credit Policy in the Brazilian Amazon (Climate Policy Initiative, 30. Pfaff A, Robalino J, Herrera D, Sandoval C (2015) Protected areas’ impacts on Brazilian
Rio de Janeiro, Brazil), pp 1–48. Amazon deforestation: Examining conservation-development interactions to inform
10. Ministério de Meio Ambiente (2015) List of Priority Municipalities in Amazonia (Brazilian planning. PLoS One 10:e0129460.
Min of Environ, Brasília-DF, Brazil). Available at bit.ly/1meJVWw. Accessed March 12, 2016. 31. Ferraro PJ, Pattanayak SK (2006) Money for nothing? A call for empirical evaluation
11. Gibbs HK, et al. (2016) Did ranchers and slaughterhouses respond to zero-deforestation of biodiversity conservation investments. PLoS Biol 4:e105.
agreements in the Brazilian Amazon? Conserv Lett 9:32–42. 32. Stickler CM, Nepstad DC, Azevedo AA, McGrath DG (2013) Defending public interests
12. Börner J, et al. (2013) Promoting forest stewardship in the Bolsa Floresta Programme: in private lands: Compliance, costs and potential environmental consequences of the
Local livelihood strategies and preliminary impacts. (CIFOR, Rio de Janeiro, Brazil). Brazilian Forest Code in Mato Grosso. Philos Trans R Soc Lond B Biol Sci 368:20120160.
13. Börner J, Wunder S, Wertz-Kanounnikoff S, Hyman G, Nascimento N (2014) Forest law en- 33. Nepstad D, et al. (2009) Environment. The end of deforestation in the Brazilian
forcement in the Brazilian Amazon: Costs and income effects. Glob Environ Change 29:294–305. Amazon. Science 326:1350–1351.
14. Börner J, Kis-Katos K, Hargrave J, König K (2015) Post-crackdown effectiveness of 34. Sparovek G, Berndes G, Klug ILF, Barretto AG (2010) Brazilian agriculture and environ-
field-based forest law enforcement in the Brazilian Amazon. PLoS One 10:e0121544. mental legislation: Status and future challenges. Environ Sci Technol 44:6046–6053.
15. Hargrave J, Kis-Katos K (2011) Economic causes of deforestation in the Brazilian 35. Conselho Monetário Nacional (2008) Resolução 3.545. Available at www.bcb.gov.br/
Amazon: A panel data analysis for the 2000s. Discussion Paper Series (Dept of Intl pre/normativos/busca/normativo.asp?tipo=Res&ano=2008&numero=003545. Ac-
Econ Policy, University of Freiburg, Germany). cessed March 12, 2016.
16. Godar J, Gardner TA, Tizado EJ, Pacheco P (2014) Actor-specific contributions to the de- 36. Greenpeace (2006) Eating up the Amazon. (Greenpeace Int, Amsterdam), pp 1–64.
forestation slowdown in the Brazilian Amazon. Proc Natl Acad Sci USA 111:15591–15596. 37. Cisneros E, Zhou SL, Börner J (2015) Naming and shaming for conservation: Evidence
17. Rajão R, Azevedo A, Stabile MC (2012) Institutional subversion and deforestation: from the Brazilian Amazon. PLoS One 10:e0136402.
Learning lessons from the system for the environmental licensing of rural properties 38. Sills EO, et al. (2015) Estimating the impacts of local policy innovation: The synthetic
in Mato Grosso. Public Adm Dev 32:229–244. control method applied to tropical deforestation. PLoS One 10:e0132590.
18. Azevedo AA, et al. (2014) Cadastro Ambiental Rural e sua influência na dinâmica do 39. Assunção J, Gandour C, Pessoa P, Rocha R (2015) Strengthening Brazil’s forest protection in a
desmatamento na Amazônia Legal. Boletim Amazônia em Pauta, ed Azevedo AA changing landscape. in Policy Brief (Clim Policy Initiative, PUC-Rio, Rio de Janeiro, Brazil), pp 1–4.
(Inst de Pesquisa Ambiental da Amazônia, Brasília). 40. Conselho Monetário Nacional (2012) Resolução 4.106. Available at www.bcb.gov.br/
19. Soares-Filho B, et al. (2014) Land use. Cracking Brazil’s Forest Code. Science 344:363–364. pre/normativos/busca/normativo.asp?tipo=res&ano=2012&numero=4106. Accessed
20. Casa Civil (2016) Lei 13.295 de 14 de Junho de 2016. Available at www.planalto.gov. March 12, 2016.
br/ccivil_03/_ato2015-2018/2016/lei/L13295.htm. Accessed March 12, 2016. 41. Rajão R, Soares-Filho B (2015) Policies undermine Brazil’s GHG goals. Science 350(6260):519.
21. Ministério de Meio Ambiente (2015) Rural Environmental Registry (CAR) Informative 42. Rajão R, Soares-Filho BS, Santiago L (2015) Estudo de viabilidade econômica do
Bulletin (Brazilian Min of Environ, Brasilia-DF, Brazil). Avaliable at bit.ly/2sHaMlk. potencial mercado de Cotas de Reserva Ambiental (CRA) no Brasil (Federal Univer-
Accessed March 12, 2016. sity of Minas Gerais, Belo Horizonte, Brazil), pp 1–70.
22. Fearnside PM (2003) Deforestation control in Mato Grosso: A new model for slowing 43. Soares-Filho B, et al. (2016) Brazil’s market for trading forest certificates. PLoS One
the loss of Brazil’s Amazon forest. Ambio 32:343–345. 11:e0152311.

7658 | www.pnas.org/cgi/doi/10.1073/pnas.1604768114 Azevedo et al.

Supplemental Information Appendix

S1. Methods

CAR database 
The CAR land registry was implemented in the states of Mato Grosso and Pará in 2009 and 2008,
respectively. Its goal was to encourage landowner compliance with Brazil’s Forest Code and improve the
monitoring capacity of each state – one of a suite of land-use policies introduced to reduce illegal
deforestation in the Brazilian Amazon. The registry contains spatial data on property boundaries,
hydrography, and land use on individual rural properties. Our analysis combines that CAR database with
deforestation data from PRODES, a deforestation monitoring system developed by the Brazilian Institute
for Space Research (1). The combined dataset summarizes land-use dynamics in 49,669 rural properties
enrolled in the CAR program, allowing us to quantify annual deforestation within each property from 2008-
2012.
 
To ensure the robustness of our statistical analyses, we excluded a number of properties from this study.
To account for limitations of the PRODES data (which cannot detect clearings under 6.5 ha), we
eliminated all properties smaller than 10 ha, as well as those outside the Amazon biome. We excluded
properties with cumulative deforestation higher than 95% to eliminate the possibility of deforestation rates
being biased by the absence of forest in a specific group of properties. Properties participating in other
environmental programs (e.g. land reform) were excluded from the analysis to isolate the impact of CAR
on deforestation. More specifically, we excluded rural settlements under the jurisdiction of the National
Institute for Agrarian Reform (INCRA), as well as properties certified1 by INCRA. In the case of Mato
Grosso, we also excluded rural properties that had begun the environmental licensing process (LAU) prior
to the creation of CAR, since the LAU is more comprehensive and allows licenses for legal deforestation.

To ensure the spatial consistency of the dataset, we excluded any properties with more than 70% of their
area overlapping, because it was impossible to determine which boundary was correct. In cases where
the overlap was smaller than 70%, we visually inspected each property and excluded the property with
the oldest CAR date. We also excluded all properties whose CAR registration lacked a date. Finally, since
our goal was to estimate CAR’s effect on illegal deforestation, we excluded properties that still had
surplus forests that could be legally deforested. We did not have access to deforestation authorizations
and could not evaluate the legality of clearings in these areas. The final filtered database excluded 52.6%
of the properties and 29.6% of the area from Pará, leaving 18,738 CAR properties covering 9.4 million ha.
In Mato Grosso, 54.8% of the properties and 45.7% of the area were excluded, leaving 3,513 CAR
properties spanning 3.0 million ha. Below is a detailed description of the final dataset for each state.

In Pará, there were 13,028 small properties (< 4 rural modules, RM), representing 69.5% of the properties
and 13.1% of the land area sampled. Medium properties (4-15 RM) totaled 3,320 (17.7%) and
occupied 20.7% of the sample area, whereas large properties (>15 RM) totaled 2,390 (12.8%) and
occupied 66.2% of the sample area. Small properties occupied a total of 1.2 million ha, medium
properties 1.95 million ha, and large properties 6.2 million ha. In 2008, less than 3% of all properties had
enrolled in CAR; by 2012, more than 96.5% of the properties had enrolled. Table S3 illustrates the
temporal dynamics of CAR enrollment for properties in Pará. Approximately 84% of the properties had
zero deforestation during the period 2008-2012 (a zero-inflated distribution).

                                                            
1
 This is a requirement for all properties, dictated by INCRA under law #10.267/01. The document is mandatory to 
sell the property or change its size.  

1 
 
In Mato Grosso the number of small properties (< 4 RM) was 1,909, representing 54.3% of the properties
and 9.2% of the land area sampled. Medium properties (4-15 RM) totaled 1,027 (29.2%) and
occupied 24.0% of the sample area, whereas large properties (>15 RM) totaled 577 (16.4%) and
occupied 66.8% of the sample area. Small properties accounted for 0.3 million ha; medium properties 0.7
million ha; and large properties 2.0 million ha. In 2009, less than 11% of the properties in these size
classes had enrolled in CAR. By 2011, more than 97% of the properties had enrolled in CAR. Table S4
illustrates the temporal dynamics of CAR enrollment for properties in Mato Grosso. Approximately 96% of
the records had zero deforestation from 2009-2011 (a zero-inflated distribution).

Analysis of deforestation before and after CAR

Finding appropriate “controls” to evaluate the impact of CAR on land-use dynamics presents a challenge.
It is not appropriate, for example, to compare the properties inside CAR (which tend to be
active/productive farms) with randomly chosen areas outside CAR, which may encompass undesignated
public lands or other land uses that have different factors affecting their deforestation probability. To
match CAR properties with comparable areas outside CAR, we adopted a BACI (Before-After-Control-
Impact) design that involves classifying properties with known boundaries into two groups – registered
and unregistered – during each year of the study.

The “CAR group” included properties that were registered in a given year and were therefore considered
to be under the influence of the CAR policy. For comparison, we selected properties with known
boundaries (the “control group”) that were not registered in CAR during that year. The properties
classified as part of the CAR group were matched to comparable properties in the control group to
evaluate the rate of deforestation before and after CAR in each year. For example, a property registered
with CAR in 2010 was considered part of CAR in subsequent years, but belonged to the control group
prior to 2010. By comparing these two groups, we quantified the effect of CAR on deforestation from 2008
to 2012. Due to the restricted number of CAR properties in Mato Grosso at the beginning and end of the
study period (2008 and 2012), we restricted our analysis to the period from 2009 to 2011.

To control for the effect of property size on deforestation rates, we divided the dataset into three size
classes defined by the number of rural modules (RM) – a legal designation whose absolute area varies by
municipality. The Forest Code uses the number of RMs as a criterion to define the legal rights and
obligations of a given property. The first group consists of small properties with less than 4 RM (under
400 ha in most Amazon municipalities); the second includes medium properties ranging from 4 to 15 RM;
and the third consists of large properties with more than 15 RM (usually larger than 6,000 ha).

We used standard ordinary least squares (OLS) regression models to estimate average deforestation
rates, adjusted for property size, the remaining forest area, CAR group, and year effects. A detailed
mathematical description of these analyses is provided below (see “Statistical Methods”). To evaluate the
effect of exogenous factors (i.e. factors other than CAR) on deforestation, we quantified exogenous
deforestation risk using the Dinâmica EGO modeling platform (SimAmazon model), described in Soares-
Filho et. al. (2). This spatially explicit model uses a “weights of evidence” approach to estimate the overall
probability of deforestation for each property. This Bayesian method accounts for the individual and
combined effects of different drivers of deforestation. Our analysis considered the following exogenous
drivers of Amazon deforestation: 1) distance to rivers; 2) distance to major roads; 3) maximum net
present value of soy and cattle rents; 4) edaphic suitability for mechanized crops; 5) elevation; 6) slope;
and 7) distance to urban centers. Based on these factors, we calculated an index of deforestation
probability (EGO index), which was included as a covariate in the regression models.

2 
 
Following Cisneros et al.  (3), we included the blacklist covariate, , which indicates whether property
belongs to a municipality blacklisted at time . We also evaluated the change in the number of fines (i.e.
for environmental infractions) as a covariate, , which serves as a proxy for the strength of enforcement
in a given municipality. We also used data on the Green Municipality Program (GMP)— available for 2011
and 2012 in the state of Pará only— to create a dummy variable considered as a potential covariate. We
then developed several regression models, comprised of different subsets of these covariates (Table S5)
and evaluated their effectiveness in predicting deforestation rates at the property level, within different
size classes (see also “Statistical methods”).

Statistical methods
Below is a brief mathematical description of the statistical model. To account for covariates, let be a
continuous random variable representing the deforested area (ha), of property at time . represents
the area of forest (ha) remaining in property at time . Each property can be classified into one of three
groups, related to size. Let , ∈ 1, 2, 3 , be the index related to each size group, or simply . For
properties up to 4 RM, then 1; for properties between 4 and 15 RM, then 2; and for properties
greater than 15 RM, then 3. Furthermore, each property can be classified through time into the
CAR or Control groups. Thus, let , be the index representing the CAR ( 1) or Control ( 2)
groups of property at time .

The response variable has a zero-inflated distribution. As previously described, 84% of the records for
Pará state and 95.6% of those for Mato Grosso state had zero deforestation ( 0). In 2009, for
example, 540 (93.6%) of the 577 properties in Mato Grosso larger than 15 RM had zero deforestation.
Observing 32 properties with non-zero deforestation would thus require 500 properties, on average. Non-
zero deforestation data ( 0) occurred in small subsets of properties and were highly asymmetric, with
extreme values. To account for this empirical distribution of the response variable, we wrote the base
OLS statistical model (hereafter named model 1) as follows:

log 1

where the dependent variable is the logarithm of the deforestation rate plus one (to account for records
with zero deforestation). The logarithm function also reduces the effect of highly asymmetric deforestation
values on the dependent variable. The logarithm of the remaining forest area was included in the model
as a covariate, , following a standard form used in epidemiological statistical regression models (4).
The dummy variable, , indicates whether property at time is enrolled in CAR ( 1) and is a
random variable representing the error.

We evaluated seven different OLS models, considering different subsets of covariates as described
below.

Model Description
1 log 1
2 log 1
3 log 1
4 log 1
5 log 1
6 log 1
7 log 1

3 
 
The variable indicates whether property belongs to a blacklisted municipality at time , described in
Cisneros et al (3); is the deforestation index (EGO), described earlier; is the change in the number
of fines over time; and is a dummy variable indicating the presence of the Green Municipality Program
(GMP) ( 1). Tables S5-S14 summarize the model results. The variables included in each model are
in Table S5. A comparison of the adjusted R2 for each model is presented in Table S6. The CAR effect of
each model is presented in Table S7, while the coefficients and model estimates from models 1-7 are
presented in Tables S8-S14. The “best” model was selected by maximizing the adjusted R-squared
(R2adj.), which includes a penalty for additional variables (see Table S6). This model selection was
corroborated by the Akaike Information Criteria (AIC) metric, which yielded the same results.

Overall, Model 3 (Table S10) did the best job of predicting deforestation at the property level across both
states (see also Table S6). Predictors included in that model were: the (log) area of remaining forest, risk
of deforestation as indicated by the EGO index, participation in the blacklist, and membership in CAR.
Although the other two variables (GMP and Fines) had an effect on deforestation in some models (and
also independently), they likely covaried with membership in CAR and did not add sufficient explanatory
power to warrant inclusion in the final model (see Table S7 for a comparison of the CAR effect in each
model). Although the municipal blacklist had an effect in Model 3, its coefficient suggests that being
blacklisted did not necessarily reduce deforestation in CAR properties (Table S10). This may be because
the federal policy of blacklisting occurs in municipalities with high deforestation rates. Municipal and state
governments have responded to these federal blacklists by creating programs such as Pará’s Green
Municipalities Program, but there is likely a time lag between being blacklisted and any observed effect
on deforestation.

Incentives for institutional change

Our analysis of farmers’ behavior and preferences relied on an analytical framework derived from
institutional theory (5-7). We evaluated farmer incentives for institutional change (IC) as follows:

IC = IN – (OC + TC + RB) 

Where IN is the net income (considering production costs and revenue); OC is the opportunity cost due to
institutional restrictions on resource use, TC is the transaction cost related to monitoring and enforcement
and RB is the risk of breaking the rules by deforesting illegally. These consequences include the payment
of fines, legal fees, losses from market restrictions, and other issues perceived by the farmer (8).
Together, these elements represent IC, the incentive associated with a given institutional framework.
Using this framework, we evaluated the likelihood that an economic agent (producer) would adhere to a
given set of rules by considering the tradeoffs among costs and benefits for any given scenario. Results
of these qualitative analyses are summarized in Figure 3.

4 
 
References

1. INPE (2016) Satellite Monitoring of Brazil's Amazon Forest (PRODES). Available at

http://www.obt.inpe.br/prodes/. (Brazilian National Agency for Space Research, São José dos
Campos, SP).
2. Soares-Filho B, et al. (2010) Role of Brazilian Amazon protected areas in climate change
mitigation. Proceeding of the National Academy of Sciences 107(24):10821-10826.
3. Cisneros E, Zhou SL, & Borner J (2015) Naming and Shaming for Conservation: Evidence from
the Brazilian Amazon. PLoS ONE 10(9):e0136402.
4. Clayton D & Hills M (2013) Statistical models in epidemiology (OUP Oxford).
5. Rajão R & Soares-Filho B (2015) Policies undermine Brazil's GHG goals. Science
350(6260):519-519.
6. Soares-Filho B, et al. (2016) Brazil's market for trading forest certificates. PLoS ONE 11(4):1-17.
7. Rajão R, Soares-Filho BS, & Santiago L (2015) Estudo de viabilidade econômica do potencial
mercado de Cotas de Reserva Ambiental (CRA) no Brasil. (Federal University of Minas Gerais,
Belo Horizonte, Brazil), pp 1-70.
8. Börner J, Wunder S, Wertz-Kanounnikoff S, Hyman G, & Nascimento N (2014) Forest law
enforcement in the Brazilian Amazon: Costs and income effects. Global Environmental Change
29:294-305.

5 
 
S2. SI Figures

Figure S1. Annual deforestation rates (per property) within the CAR and Control groups in Mato Grosso
(MT) and Pará (PA). The groups were divided into (a) small properties (< 4 RM); (b) medium-sized
properties (4-15 RM); and (c) large properties (> 15 RM). Blue lines represent deforestation in the Control
group. Dashed red lines represent deforestation in the CAR group.

6 
 
Figure S2: Farm-scale requirements to achieve environmental compliance under Brazil’s Forest Code
(FC). To determine whether a property had Legal Reserve deficits or surpluses, we conducted a spatial-
temporal analysis using CAR property boundaries and annual deforestation data from PRODES/INPE.
The analysis accounted for remaining forestland within properties in 2007 (Pará) and 2008 (Mato
Grosso). Properties with over 80% forest cover were considered to have a surplus – meaning they
exceeded FC conservation requirements. Those with le

7 
 
 S3. SI Tables
Table S1. Area of original and remaining forestland (including forest deficits and surpluses) in rural
properties of Pará state (2007), grouped according to size classes.
State of Para (2007)
Properties with forest deficit (Legal Reserve deficit)
Original  Remaining 
Property size  No. of  Remaining  Forest deficit  Forest deficit  
forest area  forest area 
classes  properties  forestland (%)  (below 80%)  (ha) 
(ha)  (ha) 
under 4 RM  10,999  1,027,431 404,818 39.40% ‐ 40.60%  417,127
4 to 15 RM  2,581  1,468,735 584,059 39.77% ‐ 40.23%  590,929
over 15 RM  1,590  5,803,471 1,699,169 29.28% ‐ 50.72%  2,943,608
Total  15,170  8,299,637 2,688,046 32.39% ‐ 47.61%  3,951,664
Properties with forest asset (Legal Reserve surplus)
Original  Remaining 
Property size  No. of  Remaining  Forest surplus  Forest surplus 
forest area  forest area 
classes  properties  forestland (%)  (above 80%)  (ha) 
(ha)  (ha) 
under 4 RM  2,787  262,986 243,445 92.57% 12.57%  33,056
4 to 15 RM  903  582,851 548,855 94.17% 14.17%  82,574
over 15 RM  945  2,023,231 1,919,872 94.89% 14.89%  301,287
Total  4,635  2,869,068 2,712,172 94.53% 14.53%  416,918
Total sample: 19,805 properties
Original  Remaining 
Property size  No. of  Remaining 
forest area  forest area 
classes  properties  forestland (%)     
(ha)  (ha) 
under 4 RM  13,786  1,290,417 648,263 50.24%
4 to 15 RM  3,484  2.051.586 1,132,914 55.22%
over 15 RM  2,535  7,826,702 3,619,041 46.24%
Total  19,805  11,168.705 5,400,218 48.35%
 

8 
 
 Table S2. Area of original and remaining forestland (including forest deficits and surpluses) in rural
properties of Mato Grosso state (2008), grouped according to sizes classes.
State of Mato Grosso (2008)
Properties with forest deficit (Legal Reserve deficit)
Remaining 
Property size  No. of  Original forest  Remaining  Forest deficit  Forest deficit 
forest area 
classes  properties  area (ha)  forestland (%)  (below 80%)  (ha) 
(ha) 
under 4 RM  1,762  255,316 78,666 30.81% ‐ 49.19%  125,587
4 to 15 RM  812  570,924 209,347 36.67% ‐ 43.33%  247,392
over 15 RM  370  1,340,641 603,928 45.05% ‐ 34.95%  468,585
Total  2,944  2,166,881 891,941 41.16% ‐ 38.84%  841,564
Properties with forest asset (Legal Reserve surplus)
Remaining 
Property size  No. of  Original forest  Remaining  Forest surplus  Forest surplus 
forest area 
classes  properties  area (ha)  forestland (%)  (above 80%)  (ha) 
(ha) 
under 4 RM  164  27,156 25,623 94.35% 14.35%  3,898
4 to 15 RM  231  169,855 161,118 94.86% 14.86%  25,234
over 15 RM  225  768,938 728,439 94.73% 14.73%  113,289
Total  620  965,949 915,180 94.74% 14.74%  142,421
Total sample: 3,564 properties
Remaining 
Property sizes  No. of  Original forest  Remaining 
forest area 
groups  properties  area (ha)  forestland (%)     
(ha) 
under 4 RM  1,926  282,472 104,289 36.92%
4 to 15 RM  1,043  740,779 370,465 50.01%
over 15 RM  595  2,109,579 1,332,367 63.16%
Total  3,564  3,132,830 1,807,121 57.68%
 

9 
 
Table S3. Enrollment of properties in CAR in Pará state (2008-2012).
 Property size class   Property type  2008 2009 2010  2011  2012

under 4 RM  Control  99.40% 96.80% 75.44% 45.1%  3.48%

CAR  0.60% 3.20% 24.56% 54.9%  96.52%
4 a 15 RM  Control  99.25% 90.90% 59.58% 34.65%  3.49%
CAR  0.75% 9.10% 40.42% 65.35%  96.51%
over 15 RM  Control  97.66% 87.57% 52.13% 31.55%  2.55%
CAR  2.34% 12.43% 47.87% 68.45%  97.45%

Table S4. Enrollment of properties in CAR in Mato Grosso state (2009- 2011).
 Property size class  Property type    2009 2010  2011 

under 4 RM  Control    97.27% 41.03% 1.10% 

CAR    2.73% 58.97% 98.90% 
4 a 15 RM  Control    90.36% 35.74% 1.66% 
CAR    9.64% 64.26% 98.34% 
over 15 RM  Control    89.60% 40.21% 2.08% 
CAR    10.40% 59.79% 97.92% 

10 
 
Table S5. Variables included in each of the tested ordinary least squares regression models.

OLS regression  Included variables 
models  logForest  CAR  EGO  Blacklist  Fines  GMP 
1  X  X             
2  X  X  X 
3  X  X  X  X 
4  X  X  X  X 
5  X  X  X  X 
6  X  X  X  X  X 
7  X  X  X     X  X 

11 
 
Table S6. Adjusted coefficient of determination (R2adj) for each of the OLS models. Bold text indicates the
best value; bold red text indicates the average of the best model.

Property size class     Model 
(Rural Module, RM)  Year  1  2  3  4  5  6  7 
Mato Grosso (2009‐2011) 
under 4 RM  2009  1.19%  1.21%  1.31%  1.30%  ‐  ‐  ‐ 
   2010  1.54%  1.53%  1.81%  1.58%  ‐  ‐  ‐ 
   2011  1.21%  1.35%  1.68%  1.28%  ‐  ‐  ‐ 
4 to 15 RM  2009  1.19%  1.21%  1.31%  1.30%  ‐  ‐  ‐ 
   2010  1.54%  1.53%  1.81%  1.58%  ‐  ‐  ‐ 
   2011  1.21%  1.35%  1.68%  1.28%  ‐  ‐  ‐ 
over 15 RM  2009  1.19%  1.21%  1.31%  1.30%  ‐  ‐  ‐ 
   2010  1.54%  1.53%  1.81%  1.58%  ‐  ‐  ‐ 
   2011  1.21%  1.35%  1.68%  1.28%  ‐  ‐  ‐ 
Mean value   1.31%  1.37%  1.60%  1.38%  ‐  ‐  ‐ 
Pará (2008‐2012) 
under 4 RM  2008  7.36%  8.76%  9.41%  8.83%  8.76%  9.41%  8.83% 
   2009  5.52%  5.84%  7.70%  5.95%  5.84%  7.70%  5.95% 
   2010  3.56%  3.72%  4.21%  4.05%  3.72%  4.21%  4.05% 
   2011  2.29%  2.62%  2.87%  2.63%  2.62%  2.88%  2.63% 
   2012  1.78%  2.02%  2.13%  2.14%  2.02%  2.11%  2.15% 
4 to 15 RM  2008  7.36%  8.76%  9.41%  8.83%  8.76%  9.41%  8.83% 
   2009  5.52%  5.84%  7.70%  5.95%  5.84%  7.70%  5.95% 
   2010  3.56%  3.72%  4.21%  4.05%  3.72%  4.21%  4.05% 
   2011  2.29%  2.62%  2.87%  2.63%  2.62%  2.88%  2.63% 
   2012  1.78%  2.02%  2.13%  2.14%  2.02%  2.11%  2.15% 
over 15 RM  2008  7.36%  8.76%  9.41%  8.83%  8.76%  9.41%  8.83% 
   2009  5.52%  5.84%  7.70%  5.95%  5.84%  7.70%  5.95% 
   2010  3.56%  3.72%  4.21%  4.05%  3.72%  4.21%  4.05% 
   2011  2.29%  2.62%  2.87%  2.63%  2.62%  2.88%  2.63% 
   2012  1.78%  2.02%  2.13%  2.14%  2.02%  2.11%  2.15% 
Mean value   4.10%  4.59%  5.26%  4.72%  4.59%  5.26%  4.72% 

12 
 
Table S7. Estimated CAR effect for each of the proposed OLS models. Statistically significant results are
shown in bold red text ( = 0.10).
Property size     Model 
class (Rural 
Module, RM)  Year  1  2  3  4  5  6  7  Mean value 
Mato Grosso (2009‐2011)
under 4 RM  2009  ‐3.77%  ‐3.54% ‐3.25% ‐3.47% ‐ ‐ ‐  ‐3.50%
   2010  ‐7.38%  ‐7.23% ‐7.21% ‐7.27% ‐ ‐ ‐  ‐7.27%
   2011  ‐5.87%  ‐5.89% ‐6.12% ‐5.89% ‐ ‐ ‐  ‐5.94%
4 to 15 RM  2009  8.87%  8.62% 7.41% 8.85% ‐ ‐ ‐  8.44%
   2010  8.37%  8.33% 8.50% 8.40% ‐ ‐ ‐  8.40%
   2011  ‐5.70%  ‐5.12% ‐4.93% ‐5.02% ‐ ‐ ‐  ‐5.19%
over 15 RM  2009  ‐9.54%  ‐9.70% ‐10.12% ‐10.89% ‐ ‐ ‐  ‐10.06%
   2010  1.81%  2.31% 0.21% 1.69% ‐ ‐ ‐  1.50%
   2011  11.56%  13.84% 10.75% 14.11% ‐ ‐ ‐  12.56%
Mean value  ‐0.18%  0.18% ‐0.53% 0.06% ‐ ‐ ‐  ‐0.12%
Pará (2008‐2012)
under 4 RM  2008  ‐31.03%  ‐31.35% ‐27.47% ‐31.36% ‐31.35% ‐27.47%  ‐31.36%  ‐30.20%
   2009  ‐25.55%  ‐26.11% ‐21.34% ‐25.93% ‐26.11% ‐21.34%  ‐25.93%  ‐24.62%
   2010  ‐11.49%  ‐11.45% ‐10.89% ‐11.11% ‐11.45% ‐10.89%  ‐11.11%  ‐11.20%
   2011  ‐5.24%  ‐5.41% ‐5.29% ‐5.40% ‐5.42% ‐5.23%  ‐5.40%  ‐5.34%
   2012  ‐3.00%  ‐3.01% ‐3.19% ‐3.09% ‐3.18% ‐3.27% ‐3.28%  ‐3.15%
4 to 15 RM  2008  ‐30.24%  ‐33.81% ‐34.71% ‐33.86% ‐33.81% ‐34.71%  ‐33.86%  ‐33.57%
   2009  26.87%  25.56% 18.54% 24.58% 25.56% 18.54%  24.58%  23.46%
   2010  ‐5.04%  ‐5.18% ‐4.72% ‐5.99% ‐5.18% ‐4.72% ‐5.99%  ‐5.26%
   2011  0.72%  0.96% 0.96% 0.87% 1.21% 1.28% 1.14%  1.02%
   2012  ‐14.62%  ‐14.48% ‐14.61% ‐14.37% ‐14.61% ‐14.66%  ‐14.47%  ‐14.54%
over 15 RM  2008  32.05%  16.98% 13.78% 15.95% 16.98% 13.78%  15.95%  17.92%
   2009  31.71%  27.56% 21.37% 26.43% 27.56% 21.37%  26.43%  26.06%
   2010  ‐6.76%  ‐6.75% ‐8.13% ‐6.20% ‐6.75% ‐8.13%  ‐6.20%  ‐6.99%
   2011  ‐5.24%  ‐3.40% ‐2.91% ‐3.53% ‐3.40% ‐2.93% ‐3.56%  ‐3.56%
   2012  12.48%  13.48% 14.09% 12.56% 13.50% 14.19%  12.77%  13.30%
Mean value  ‐2.29%  ‐3.76% ‐4.30% ‐4.03% ‐3.76% ‐4.28% ‐4.02%  ‐3.78%

13 
 
Table S8. Estimated coefficients and statistical inference results for model 1. Bold text indicates statistical
significance; bold red text indicates significant CAR effect.

Property size class     logForest  CAR    

(Rural Module, RM)  Year  estimate  Pr(>|t|)  CAR effect  CAR effect(%)  Pr(>|t|)  R2adj 
Mato Grosso (2009‐2011) 
under 4 RM  2009  1.0469  0.0000  0.9623  ‐3.77%  0.5984  1.19% 
   2010  1.0469  0.0000  0.9262  ‐7.38%  0.0040  1.54% 
   2011  1.0646  0.0000  0.9413  ‐5.87%  0.6594  1.21% 
4 to 15 RM  2009  1.0458  0.0078  1.0887  8.87%  0.3526  1.19% 
   2010  1.0595  0.0016  1.0837  8.37%  0.0776  1.54% 
   2011  1.0381  0.0703  0.9430  ‐5.70%  0.7818  1.21% 
over 15 RM  2009  1.0424  0.0679  0.9046  ‐9.54%  0.3244  1.19% 
   2010  1.0452  0.0743  1.0181  1.81%  0.7443  1.54% 
   2011  0.9730  0.3262  1.1156  11.56%  0.6366  1.21% 
Pará (2008‐2012) 
under 4 RM  2008  1.2420  0.0000  0.6897  ‐31.03%  0.0035  7.36% 
   2009  1.1923  0.0000  0.7445  ‐25.55%  0.0000  5.52% 
   2010  1.1571  0.0000  0.8851  ‐11.49%  0.0000  3.56% 
   2011  1.1114  0.0000  0.9476  ‐5.24%  0.0000  2.29% 
   2012  1.0803  0.0000  0.9700  ‐3.00%  0.3094  1.78% 
4 to 15 RM  2008  1.2378  0.0000  0.6976  ‐30.24%  0.1634  7.36% 
   2009  1.2070  0.0000  1.2687  26.87%  0.0010  5.52% 
   2010  1.0824  0.0000  0.9496  ‐5.04%  0.1637  3.56% 
   2011  1.0494  0.0001  1.0072  0.72%  0.8119  2.29% 
   2012  1.0274  0.0085  0.8538  ‐14.62%  0.0173  1.78% 
over 15 RM  2008  1.2352  0.0000  1.3205  32.05%  0.1604  7.36% 
   2009  1.1923  0.0000  1.3171  31.71%  0.0002  5.52% 
   2010  1.1149  0.0000  0.9324  ‐6.76%  0.0905  3.56% 
   2011  1.0837  0.0000  0.9476  ‐5.24%  0.1276  2.29% 
   2012  1.0909  0.0000  1.1248  12.48%  0.1746  1.78% 

14 
 
Table S9. Estimated coefficients and statistical inference results for model 2. Bold text indicates statistical
significance; bold red text indicates significant CAR effect.

Property size class     logForest  EGO  CAR    

(Rural Module, RM)  Year  estimate  Pr(>|t|) estimate Pr(>|t|) CAR effect CAR effect(%)  Pr(>|t|) R2adj 
Mato Grosso (2009‐2011) 
under 4 RM  2009  1.1966  0.0000  1.0010  0.0020  0.9646  ‐3.54%  0.6215  5.84%
   2010  1.1567  0.0000  0.9999  0.8162  0.9277  ‐7.23%  0.0050  3.72%
   2011  1.1140  0.0000  1.0009  0.0003  0.9411  ‐5.89%  0.6578  2.62%
4 to 15 RM  2009  1.2430  0.0000  1.0027  0.0000  1.0862  8.62%  0.3655  5.84%
   2010  1.0862  0.0000  1.0003  0.6033  1.0833  8.33%  0.0797  3.72%
   2011  1.0527  0.0001  1.0004  0.4944  0.9488  ‐5.12%  0.8040  2.62%
over 15 RM  2009  1.2352  0.0000  1.0039  0.0000  0.9030  ‐9.70%  0.3161  5.84%
   2010  1.1502  0.0000  1.0035  0.0000  1.0231  2.31%  0.6804  3.72%
   2011  1.1195  0.0000  1.0037  0.0000  1.1384  13.84%  0.5757  2.62%
Pará (2008‐2012) 
under 4 RM  2008  1.2458  0.0000  1.0009  0.0221  0.6865  ‐31.35%  0.0029  8.76%
   2009  1.1966  0.0000  1.0010  0.0020  0.7389  ‐26.11%  0.0000  5.84%
   2010  1.1567  0.0000  0.9999  0.8162  0.8855  ‐11.45%  0.0000  3.72%
   2011  1.1140  0.0000  1.0009  0.0003  0.9459  ‐5.41%  0.0000  2.62%
   2012  1.0802  0.0000  1.0000  0.8758  0.9699  ‐3.01%  0.3065  2.02%
4 to 15 RM  2008  1.3290  0.0000  1.0065  0.0000  0.6619  ‐33.81%  0.1078  8.76%
   2009  1.2430  0.0000  1.0027  0.0000  1.2556  25.56%  0.0016  5.84%
   2010  1.0862  0.0000  1.0003  0.6033  0.9482  ‐5.18%  0.1524  3.72%
   2011  1.0527  0.0001  1.0004  0.4944  1.0096  0.96%  0.7513  2.62%
   2012  1.0305  0.0066  1.0003  0.4620  0.8552  ‐14.48%  0.0184  2.02%
over 15 RM  2008  1.3630  0.0000  1.0110  0.0000  1.1698  16.98%  0.4253  8.76%
   2009  1.2352  0.0000  1.0039  0.0000  1.2756  27.56%  0.0009  5.84%
   2010  1.1502  0.0000  1.0035  0.0000  0.9325  ‐6.75%  0.0910  3.72%
   2011  1.1195  0.0000  1.0037  0.0000  0.9660  ‐3.40%  0.3281  2.62%
   2012  1.1199  0.0000  1.0030  0.0000  1.1348  13.48%  0.1437  2.02%

15 
 
Table S10. Estimated coefficients and statistical inference results for model 3. Bold text indicates statistical significance; bold red text indicates
significant CAR effect.
Property size class     logForest EGO Blacklist CAR
(Rural Module, RM)  Year  estimate  Pr(>|t|) estimate Pr(>|t|) estimate Pr(>|t|)  CAR effect CAR effect(%) Pr(>|t|) R2adj
Mato Grosso (2009‐2011)
under 4 RM  2009  1.0451  0.0000 0.9988 0.1858 1.0302 0.2439  0.9675 ‐3.25% 0.6513 1.31%
   2010  1.0460  0.0000 0.9994 0.5561 1.0179 0.5239  0.9279 ‐7.21% 0.0051 1.81%
   2011  1.0653  0.0000 1.0001 0.9207 1.0306 0.3458  0.9388 ‐6.12% 0.6445 1.68%
4 to 15 RM  2009  1.0400  0.0250 1.0002 0.8332 1.0770 0.0343  1.0741 7.41% 0.4346 1.31%
   2010  1.0523  0.0072 0.9990 0.4031 1.0304 0.4272  1.0850 8.50% 0.0737 1.81%
   2011  1.0315  0.1482 0.9985 0.2864 1.0019 0.9647  0.9507 ‐4.93% 0.8112 1.68%
over than 15 RM  2009  1.0391  0.1139 0.9986 0.2528 0.9631 0.4272  0.8988 ‐10.12% 0.2944 1.31%
   2010  1.0748  0.0067 1.0003 0.8061 0.8367 0.0006  1.0021 0.21% 0.9700 1.81%
   2011  1.0197  0.5139 1.0026 0.0786 0.8062 0.0002  1.1075 10.75% 0.6589 1.68%
Pará (2008‐2012)
under 4 RM  2008  1.2289  0.0000 1.0002 0.5439 1.5264 0.0000  0.7253 ‐27.47% 0.0108 9.41%
   2009  1.1781  0.0000 1.0000 0.8923 1.3173 0.0000  0.7866 ‐21.34% 0.0000 7.70%
   2010  1.1496  0.0000 0.9992 0.0107 1.1654 0.0000  0.8911 ‐10.89% 0.0000 4.21%
   2011  1.1116  0.0000 1.0006 0.0227 1.0998 0.0000  0.9471 ‐5.29% 0.0000 2.87%
   2012  1.0797  0.0000 0.9998 0.3968 1.0386 0.0010  0.9681 ‐3.19% 0.2785 2.13%
4 to 15 RM  2008  1.2856  0.0000 1.0055 0.0000 1.6903 0.0000  0.6529 ‐34.71% 0.0956 9.41%
   2009  1.1818  0.0000 1.0009 0.1569 1.5226 0.0000  1.1854 18.54% 0.0177 7.70%
   2010  1.0711  0.0001 0.9996 0.5749 1.1378 0.0001  0.9528 ‐4.72% 0.1933 4.21%
   2011  1.0499  0.0003 1.0002 0.6748 1.0310 0.2548  1.0096 0.96% 0.7498 2.87%
   2012  1.0259  0.0228 1.0001 0.7969 1.0472 0.0404  0.8539 ‐14.61% 0.0173 2.13%
over than 15 RM  2008  1.3485  0.0000 1.0106 0.0000 1.2404 0.0226  1.1378 13.78% 0.5111 9.41%
   2009  1.2165  0.0000 1.0032 0.0000 1.1726 0.0001  1.2137 21.37% 0.0079 7.70%
   2010  1.1566  0.0000 1.0038 0.0000 0.9395 0.1104  0.9187 ‐8.13% 0.0416 4.21%
   2011  1.1143  0.0000 1.0035 0.0000 1.0560 0.0903  0.9709 ‐2.91% 0.4065 2.87%
   2012  1.1110  0.0000 1.0026 0.0000 1.0842 0.0028  1.1409 14.09% 0.1277 2.13%

16 
 
Table S11. Estimated coefficients and statistical inference results for model 4. Bold text indicates statistical significance; bold red text indicates
significant CAR effect.
Property size class     logForest EGO Delta‐Fines  CAR
(Rural Module, RM)  Year  estimate Pr(>|t|) estimate Pr(>|t|) estimate Pr(>|t|) CAR effect CAR effect(%) Pr(>|t|) R2adj
Mato Grosso (2009‐2011)
under 4 RM  2009  1.0439 0.0000 0.9985 0.1017 0.9990 0.8232 0.9653 ‐3.47% 0.6283 1.30%
   2010  1.0456 0.0000 0.9992 0.4088 0.9941 0.6396 0.9273 ‐7.27% 0.0048 1.58%
   2011  1.0644 0.0000 0.9999 0.9115 0.9999 0.9896 0.9411 ‐5.89% 0.6581 1.28%
4 to 15 RM  2009  1.0411 0.0209 0.9997 0.7858 1.0114 0.0718 1.0885 8.85% 0.3539 1.30%
   2010  1.0537 0.0059 0.9987 0.2593 0.9969 0.7891 1.0840 8.40% 0.0771 1.58%
   2011  1.0324 0.1373 0.9986 0.2790 0.9968 0.6651 0.9498 ‐5.02% 0.8078 1.28%
over 15 RM  2009  1.0341 0.1597 0.9989 0.3474 1.0130 0.0914 0.8911 ‐10.89% 0.2585 1.30%
   2010  1.0536 0.0445 1.0015 0.2596 1.0356 0.0390 1.0169 1.69% 0.7621 1.58%
   2011  0.9958 0.8861 1.0038 0.0094 0.9967 0.7685 1.1411 14.11% 0.5690 1.28%
Pará (2008‐2012)
under 4 RM  2008  1.2460 0.0000 1.0009 0.0221 0.9995 0.8854 0.6864 ‐31.36% 0.0029 8.83%
   2009  1.1960 0.0000 1.0008 0.0069 1.0121 0.0272 0.7407 ‐25.93% 0.0000 5.95%
   2010  1.1572 0.0000 1.0000 0.8991 1.0231 0.0028 0.8889 ‐11.11% 0.0000 4.05%
   2011  1.1140 0.0000 1.0009 0.0004 1.0003 0.8893 0.9460 ‐5.40% 0.0000 2.63%
   2012  1.0806 0.0000 1.0000 0.9728 0.9968 0.2476 0.9691 ‐3.09% 0.2947 2.14%
4 to 15 RM  2008  1.3320 0.0000 1.0065 0.0000 0.9958 0.5516 0.6614 ‐33.86% 0.1069 8.83%
   2009  1.2421 0.0000 1.0030 0.0000 0.9669 0.0019 1.2458 24.58% 0.0023 5.95%
   2010  1.0918 0.0000 1.0002 0.7972 1.0820 0.0000 0.9401 ‐5.99% 0.0964 4.05%
   2011  1.0518 0.0001 1.0003 0.5600 1.0057 0.2159 1.0087 0.87% 0.7746 2.63%
   2012  1.0305 0.0065 1.0003 0.4450 0.9977 0.6374 0.8563 ‐14.37% 0.0193 2.14%
over 15 RM  2008  1.3725 0.0000 1.0107 0.0000 0.9565 0.0000 1.1595 15.95% 0.4517 8.83%
   2009  1.2345 0.0000 1.0040 0.0000 0.9591 0.0009 1.2643 26.43% 0.0014 5.95%
   2010  1.1505 0.0000 1.0035 0.0000 1.0988 0.0000 0.9380 ‐6.20% 0.1217 4.05%
   2011  1.1173 0.0000 1.0037 0.0000 1.0078 0.0948 0.9647 ‐3.53% 0.3093 2.63%
   2012  1.1168 0.0000 1.0029 0.0000 1.0372 0.0000 1.1256 12.56% 0.1713 2.14%

17 
 
Table S12. Estimated coefficients and statistical inference results for model 5. Bold text indicates statistical significance; bold red text indicates
significant CAR effect.

Property size class           logForest  EGO  GMP  CAR    

(Rural Module, RM)  Year  Estimate  Pr(>|t|) estimate Pr(>|t|) estimate Pr(>|t|) estimate  Pr(>|t|) CAR effect CAR effect(%) Pr(>|t|) R2adj 
Pará (2008‐2012) 
under 4 RM  2008  0.6977  0.0000 1.2458  0.0000  1.0009  0.0221  NA  NA  0.6865  ‐31.35%  0.0029 8.76% 
   2009  0.7059  0.0000 1.1966  0.0000  1.0010  0.0020  NA  NA  0.7389  ‐26.11%  0.0000 5.84% 
   2010  0.8985  0.0030 1.1567  0.0000  0.9999  0.8162  NA  NA  0.8855  ‐11.45%  0.0000 3.72% 
   2011  0.8653  0.0002 1.1139  0.0000  1.0009  0.0003  1.0028  0.9291  0.9458  ‐5.42%  0.0000 2.62% 
   2012  0.9238  0.0652 1.0792  0.0000  1.0000  0.8909  1.0311  0.2429  0.9682  ‐3.18%  0.2819  2.02% 
4 to 15 RM  2008  0.4580  0.0000 1.3290  0.0000  1.0065  0.0000  NA  NA  0.6619  ‐33.81%  0.1078  8.76% 
   2009  0.6397  0.0003 1.2430  0.0000  1.0027  0.0000  NA  NA  1.2556  25.56%  0.0016 5.84% 
   2010  1.1961  0.1441 1.0862  0.0000  1.0003  0.6033  NA  NA  0.9482  ‐5.18%  0.1524  3.72% 
   2011  1.2033  0.0755 1.0581  0.0000  1.0005  0.3473  0.9316  0.1399  1.0121  1.21%  0.6907  2.62% 
   2012  1.2729  0.0238 1.0271  0.0192  1.0002  0.5977  1.0495  0.2315  0.8539  ‐14.61%  0.0174 2.02% 
over 15 RM  2008  0.2342  0.0000 1.3630  0.0000  1.0110  0.0000  NA  NA  1.1698  16.98%  0.4253  8.76% 
   2009  0.4645  0.0000 1.2352  0.0000  1.0039  0.0000  NA  NA  1.2756  27.56%  0.0009 5.84% 
   2010  0.6650  0.0039 1.1502  0.0000  1.0035  0.0000  NA  NA  0.9325  ‐6.75%  0.0910 3.72% 
   2011  0.6482  0.0004 1.1202  0.0000  1.0038  0.0000  0.9899  0.8446  0.9660  ‐3.40%  0.3274  2.62% 
   2012  0.4747  0.0000 1.1173  0.0000  1.0029  0.0000  1.0404  0.3624  1.1350  13.50%  0.1437  2.02% 

18 
 
Table S13. Estimated coefficients and statistical inference results for model 6. Bold text indicates statistical significance; bold red text indicates
significant CAR effect.

Property size class     logForest  EGO  Blacklist  GMP  CAR    

CAR CAR 
(Rural Module, RM)  Year  estimate  Pr(>|t|) estimate Pr(>|t|) estimate Pr(>|t|) estimate  Pr(>|t|) effect  effect(%)  Pr(>|t|) R2adj 
Pará (2008‐2012) 
under 4 RM  2008  1.2289  0.0000  1.0002  0.5439  1.5264  0.0000  NA  NA  0.7253  ‐27.47%  0.0108  9.41% 
   2009  1.1781  0.0000  1.0000  0.8923  1.3173  0.0000  NA  NA  0.7866  ‐21.34%  0.0000  7.70% 
   2010  1.1496  0.0000  0.9992  0.0107  1.1654  0.0000  NA  NA  0.8911  ‐10.89%  0.0000  4.21% 
   2011  1.1127  0.0000  1.0005  0.0257  1.1027  0.0000  0.9660  0.2734  0.9477  ‐5.23%  0.0000  2.88% 
   2012  1.0792  0.0000  0.9998  0.4155  1.0374  0.0017  1.0152  0.5704  0.9673  ‐3.27%  0.2680  2.11% 
4 to 15 RM  2008  1.2856  0.0000  1.0055  0.0000  1.6903  0.0000  NA  NA  0.6529  ‐34.71%  0.0956  9.41% 
   2009  1.1818  0.0000  1.0009  0.1569  1.5226  0.0000  NA  NA  1.1854  18.54%  0.0177  7.70% 
   2010  1.0711  0.0001  0.9996  0.5749  1.1378  0.0001  NA  NA  0.9528  ‐4.72%  0.1933  4.21% 
   2011  1.0556  0.0001  1.0003  0.5180  1.0463  0.1044  0.9103  0.0605  1.0128  1.28%  0.6743  2.88% 
   2012  1.0245  0.0351  1.0001  0.8481  1.0427  0.0753  1.0270  0.5283  0.8534  ‐14.66%  0.0169  2.11% 
over 15 RM  2008  1.3485  0.0000  1.0106  0.0000  1.2404  0.0226  NA  NA  1.1378  13.78%  0.5111  9.41% 
   2009  1.2165  0.0000  1.0032  0.0000  1.1726  0.0001  NA  NA  1.2137  21.37%  0.0079  7.70% 
   2010  1.1566  0.0000  1.0038  0.0000  0.9395  0.1104  NA  NA  0.9187  ‐8.13%  0.0416  4.21% 
   2011  1.1163  0.0000  1.0036  0.0000  1.0646  0.0646  0.9592  0.4447  0.9707  ‐2.93%  0.4024  2.88% 
   2012  1.1112  0.0000  1.0026  0.0000  1.0849  0.0044  0.9968  0.9445  1.1419  14.19%  0.1252  2.11% 

19 
 
Table S14. Estimated coefficients and statistical inference results for model 7. Bold text indicates statistical significance; bold red text indicates
significant CAR effect.
Property size class 
(RM)     logForest  EGO  Delta_Fines  PMV  CAR    
CAR CAR 
Year  estimate  Pr(>|t|)  estimate Pr(>|t|) estimate Pr(>|t|) estimate  Pr(>|t|) effect  effect(%) Pr(>|t|)  R2adj 
Pará (2008‐2012) 
under 4 RM  2008  1.2460  0.0000  1.0009  0.0221  0.9995  0.8854  NA  NA  0.6864  ‐31.36%  0.0029  8.83%
   2009  1.1960  0.0000  1.0008  0.0069  1.0121  0.0272  NA  NA  0.7407  ‐25.93%  0.0000  5.95%
   2010  1.1572  0.0000  1.0000  0.8991  1.0231  0.0028  NA  NA  0.8889  ‐11.11%  0.0000  4.05%
   2011  1.1139  0.0000  1.0009  0.0004  1.0002  0.8960  1.0024  0.9402 0.9460  ‐5.40%  0.0000  2.63%
   2012  1.0795  0.0000  1.0000  0.9357  0.9964  0.1928  1.0353  0.1892 0.9672  ‐3.28%  0.2662  2.15%
4 to 15 RM  2008  1.3320  0.0000  1.0065  0.0000  0.9958  0.5516  NA  NA  0.6614  ‐33.86%  0.1069  8.83%
   2009  1.2421  0.0000  1.0030  0.0000  0.9669  0.0019  NA  NA  1.2458  24.58%  0.0023  5.95%
   2010  1.0918  0.0000  1.0002  0.7972  1.0820  0.0000  NA  NA  0.9401  ‐5.99%  0.0964  4.05%
   2011  1.0576  0.0000  1.0005  0.3914  1.0066  0.1544  0.9239  0.1019 1.0114  1.14%  0.7096  2.63%
   2012  1.0268  0.0204  1.0002  0.5813  0.9967  0.5029  1.0540  0.1976 0.8553  ‐14.47%  0.0185  2.15%
over 15 RM  2008  1.3725  0.0000  1.0107  0.0000  0.9565  0.0000  NA  NA  1.1595  15.95%  0.4517  8.83%
   2009  1.2345  0.0000  1.0040  0.0000  0.9591  0.0009  NA  NA  1.2643  26.43%  0.0014  5.95%
   2010  1.1505  0.0000  1.0035  0.0000  1.0988  0.0000  NA  NA  0.9380  ‐6.20%  0.1217  4.05%
   2011  1.1185  0.0000  1.0037  0.0000  1.0080  0.0884  0.9799  0.6978 0.9644  ‐3.56%  0.3066  2.63%
   2012  1.1166  0.0000  1.0029  0.0000  1.0371  0.0000  1.0038  0.9321 1.1277  12.77%  0.1649  2.15%

20