Research

INTRODUCTION

Deforestation and economic growth in the Brazilian Amazon have long been intertwined, with natural resource extraction often viewed as a primary driver of regional development (Angelsen and Kaimowitz 1999). The Amazon region, home to an ecosystem with the most incredible biodiversity in the world and a vital carbon sink, has been subjected to significant deforestation since the 1970s, primarily driven by agricultural expansion, illegal mining, and infrastructure development (Sant’Anna 2016). Although deforestation may bring short-term economic benefits, mainly through logging and ranching, these gains are often temporary and can lead to long-term economic stagnation and environmental degradation. The complexities surrounding this relationship between deforestation and economic development are critical for Brazil and for global economic and ecological sustainability (Beuchle et al. 2021).

The environmental Kuznets curve (EKC) has often been used to describe the relationship between economic growth and environmental degradation, suggesting that such degradation intensifies during the early stages of development but eventually declines once a certain income level is reached. Although this model may apply to some flow pollutants, it fails to adequately account for the dynamics of natural capital, particularly in the case of stock environmental assets like forests. Forest ecosystems embody intricate ecological feedbacks that fundamentally challenge the assumptions of the EKC (Choumert et al. 2013). This is particularly true in tropical regions, where deforestation can trigger cascading effects, such as altered rainfall regimes, accelerated soil erosion, and significant biodiversity loss. These processes can undermine long-term economic performance, contradicting the notion of a natural reversal in environmental harm as income rises. Supporting this view, Stern (2004) argues that although deforestation may initially boost gross domestic product (GDP), its long-term consequences are likely to threaten the sustainability of economic growth.

Considering these limitations, this paper proposes a transformed interpretation of the EKC, which reverses its conventional analytical focus. Rather than analyzing how economic growth affects environmental degradation, the paper examines the relationship between environmental depletion (specifically deforestation) and economic performance. This conceptual shift is significant because it repositions environmental conditions as core components of economic systems, recognizing that the degradation of natural capital may coincide with patterns of reduced economic output. This approach facilitates a deeper understanding of the environmental foundations of development, particularly in regions where natural resource depletion is closely tied to short-term economic activities.

Unlike the traditional EKC framework, which examines how economic growth affects environmental degradation, this paper reverses the causal direction by exploring how environmental depletion (specifically deforestation) affects economic performance. It investigates the relationship between deforestation and economic growth in the Brazilian Legal Amazon, proposing the transposed environmental Kuznets curve (TEKC) as an adaptation of the conventional EKC, which typically illustrates an inverted U-shaped relationship between economic growth and environmental harm. The TEKC focuses on how ecological degradation, specifically deforestation, is associated with economic output as forest depletion intensifies across municipalities. Rather than merely treating environmental degradation as a side effect of development, the TEKC suggests that exploiting natural resources, such as forests, may stimulate economic activity initially. However, continued degradation may hinder sustainable growth beyond a certain point. Although the TEKC is applied to cross-sectional data, without explicitly modeling changes over time, variation in deforestation levels across municipalities may also reflect underlying temporal processes shaped by factors such as agricultural expansion, territorial occupation policies, logging activity, and population growth (Arraes et al. 2012). These dynamics highlight the need to explore how different stages of land use transformation can shape the region’s economic development trajectory.

The research utilizes propensity score weighting (PSW) to address potential selection biases, providing a robust empirical framework for assessing the causal impacts of deforestation on regional GDP. To further refine the analysis, trimmed models were employed, focusing exclusively on municipalities that experienced some degree of deforestation and removing outliers to ensure the robustness of the findings. This trimming allowed the study to concentrate on relevant cases, eliminating extreme values that could skew the results. By integrating these trimmed models, the research offers a clearer view of the short-term economic benefits of deforestation and its long-term risks, providing deeper insights into how deforestation-driven activities may ultimately undermine regional development.

LITERATURE REVIEW

This literature review investigates the intricate relationship between deforestation and economic growth within the Brazilian Amazon, focusing on the environmental Kuznets curve (EKC) and its inherent limitations. It introduces the concept of the transposed environmental Kuznets curve (TEKC), which hypothesizes that environmental stock, such as forests, serves as a crucial input in the production process. By examining this model, the review aims to provide a more nuanced understanding of the economic repercussions of deforestation, emphasizing the structural risks associated with ecosystem depletion and the pattern in which higher levels of forest loss tend to coincide with reduced economic performance, even where initial gains are observed.

Environment and economic growth in the Brazilian Amazon

The Brazilian Legal Amazon covers approximately 5 million km², accounting for nearly 60% of the country’s total territory (Haddad et al. 2024). As the largest tropical forest on Earth, it hosts 10% of the world’s biodiversity and is crucial in climate change mitigation (Stabile et al. 2022). Despite its international importance, the Brazilian Amazon has lost approximately 20% of its original area since the 1970s because of deforestation, illegal mining, encroachment on public lands, forest fires, agricultural expansion, and large infrastructure projects that lack adequate safeguards (Beuchle et al. 2021, United Nations Environment Programme 2023). Simulation models, however, show that about 40% deforestation would be enough to degrade the Amazon hydrological cycle, leading to diminished rainfall and a lengthier dry season. Should this happen, the models predict a shift in the rainforest biomes to savanna vegetation (Lovejoy and Nobre 2018). Therefore, we may already be halfway to this tipping point.

The region has shown moderate economic expansion over the past two decades. According to Assunção et al. (2022), the Legal Amazon’s real GDP increased from approximately R$ 510 billion in 2010 to R$ 612 billion in 2019, representing a growth of about 20% over the period. Despite this expansion, the region contributed only around 8% of Brazil’s total GDP in 2019, reflecting its limited weight within the national economy. However, this growth was uneven and heavily concentrated in a few urban and industrialized areas. During the same timeframe, real GDP per capita rose from approximately R$ 16,500 to R$ 18,200, an increase of about 10% (Assunção et al. 2022). Nevertheless, the region’s GDP per capita remained substantially lower than the national average, reaching only about 60% of Brazil’s GDP per capita by 2019. Similarly, World Bank (2023) data highlight that although certain Amazonian states have undergone structural economic transformation, the region continues to lag behind the rest of Brazil regarding labor productivity and income. The combination of localized growth hubs and widespread socioeconomic vulnerability characterizes the complex economic landscape of the Legal Amazon today. These figures reveal that although aggregate regional GDP has increased, the benefits of this expansion have not been uniformly distributed, and many municipalities remain dependent on primary sectors with limited long-term growth prospects. According to Agência de Notícias - IBGE (Santos et al. 2021), only 19% of the municipalities in the Legal Amazon are part of metropolitan regions,^[1] yet they are responsible for 40% of the region’s total GDP. These economic figures are crucial for understanding how deforestation-driven activities intersect with uneven development trajectories at the municipal level.

Economic development in the Brazilian Amazon often relies heavily on natural resources, particularly agriculture, mining, logging, and other resource-intensive industries, contributing to forest depletion. As a result, deforestation is frequently viewed as an unavoidable and essential aspect of the region’s economic growth (Feltran-Barbieri et al. 2023). Theory suggests that environmental degradation occurs when the financial returns from converting forests to alternative uses are perceived to outweigh the benefits of conservation (Pearce 2002). New research, however, indicates that deforestation is not an inevitable consequence of economic development in the Amazon. The World Resources Institute (WRI) Brasil and New Climate Economy highlight that Brazil can halt deforestation while growing its economy across significant sectors like agriculture, clean energy, and the bioeconomy (Feltran-Barbieri et al. 2023). A deforestation-free model could add BRL (Brazilian Reais) 40 billion annually to the Brazilian Amazon’s GDP by 2050, creating sustainable jobs for local and Indigenous communities. This alternative path would be more cost-effective than the current trajectory, especially as the financial toll of climate change becomes more severe. Without significant changes, deforestation could reach 59 million hectares by 2050, leading to greenhouse gas emissions that far exceed global climate goals (Feltran-Barbieri et al. 2023).

Despite the potential for short-term economic gains, deforestation-driven growth in the Brazilian Amazon often follows a boom-and-bust trajectory. An influential study analyzing 286 municipalities along the Amazon deforestation frontier found that indicators of human development, such as income, literacy, and life expectancy, tend to improve in the early stages of forest clearing but decline as deforestation progresses. In municipalities where natural resources are rapidly depleted, the initial surge in development is typically not sustained, resulting in outcomes that are no better than in areas that remained largely forested (Rodrigues et al. 2009). This pattern reflects the exhaustion of ecosystem services and productive land, and growing pressures on local infrastructure and economies. These findings highlight the need for development strategies that do not rely on continued forest loss but promote lasting improvements in well-being through sustainable use of already-cleared areas and forest-based economic alternatives.

According to Tateishi et al. (2021), credit policies can reduce land conversion by influencing producers’ behavior, but fines have mainly proven ineffective because of weak enforcement mechanisms. The study also highlights that socioeconomic development tends to be associated with lower rates of forest conversion, whereas urbanization exerts the opposite effect by increasing pressure on land. Given the spatial heterogeneity of the Amazon, uniform policies often fail to produce consistent results across regions. Moreover, short-term fluctuations in agricultural production can significantly influence land use decisions, and economic incentives alone may be insufficient to halt deforestation. Institutional shortcomings, such as inadequate land regulation, fragile environmental governance, and the persistence of illegal land annexation, undermine conservation efforts. These gaps are frequently exploited by actors engaged in extensive cattle ranching, who use deforestation to establish territorial claims or expand production cheaply. Although agricultural and ranching expansion are key drivers of forest loss, the study suggests that sustainable forest management could become a viable alternative, particularly in contexts where the economic value of standing forests outweighs the returns from land conversion.

The interplay between productivity growth, economic incentives, and environmental degradation emphasizes the complexity of addressing deforestation in the Amazon. Although the commodity boom of the 2000s fueled economic expansion, public policies such as the Action Plan for the Prevention and Control of Deforestation in the Legal Amazon (PPCDAm) were initially effective in curbing forest loss. However, deforestation resurged as policy enforcement weakened and market pressures persisted (Pereira 2019). This pattern reveals that sustainable development in the region cannot rely on temporary measures or extractive growth alone. Instead, it demands long-term strategies to maintain ecological integrity while fostering inclusive economic opportunities.

A growing body of literature, including insights from Forest Transition Theory, emphasizes the need to transition from deforestation-based models to those that align economic growth with forest conservation. Long-term prosperity depends on structural changes in institutions, land use governance, and technological innovation that enable development to be decoupled from forest loss (Rudel et al. 2005, Lambin and Meyfroidt 2010, Meyfroidt and Lambin 2011). In the Amazonian context, this entails restoring degraded lands, intensifying production in already-cleared areas, and developing value chains compatible with standing forests, such as sustainable agriculture, agroforestry, ecotourism, and non-timber forest products (Barbier et al. 2010). The Paradoxo Amazônico framework reinforces this perspective by identifying degraded areas, remaining carbon stocks, and a young labor force as key assets for forest-based development (Santos et al. 2022). Policy instruments like Ecological-Economic Zoning (EEZ) and Payments for Environmental Services (PES) are crucial for aligning local incentives with conservation outcomes and guiding land use planning toward sustainability (Santos et al. 2022, World Bank 2023). Moreover, gains in productivity through innovation, technical assistance, and better resource management can reduce the pressure to expand agricultural frontiers, enabling economic dynamism without further forest loss. Collectively, these approaches outline a roadmap toward a low-carbon, inclusive development model for the Amazon.

The environmental Kuznets curve (EKC)

In one study about economic growth and environmental degradation, Grossman and Krueger (1991) suggested that although economic growth initially leads to increased emissions because of heightened activity levels, a turning point is reached when societies attain a certain income threshold. At this stage, increased wealth and concern for a sustainable environment drive the adoption of cleaner technologies and the enforcement of stricter environmental regulations, ultimately leading to reduced pollution levels. The relationship between income and pollution is often termed the environmental Kuznets curve (EKC) because of its inverted U-shaped pattern, which resembles Professor Kuznets’s income-inequality curve (Kuznets 1955, Harbaugh et al. 2002).

A significant body of literature employs CO₂ emissions as a critical variable in EKC models to assess environmental degradation. These studies frequently validate the EKC hypothesis, consistently demonstrating the inverted U-shaped relationship when pollutant emissions are incorporated into the analysis (Ulucak and Bilgili 2018, Wang et al. 2024). Some critics have challenged the EKC hypothesis, arguing that economic growth alone may not induce environmental improvement, particularly when resources are permanently degraded, like stock resources (Arrow et al. 1996). Additionally, the EKC hypothesis assumes that environmental damage does not significantly reduce economic activity to the extent that it halts the growth process and that any irreversibility is not severe enough to lower future income levels (Stern 2004).

Studies have also demonstrated that as technological advancements enhance resource use efficiency, overall consumption of that resource may increase rather than decrease because of lower costs and heightened demand; this phenomenon is known as the Jevons Paradox (Gunderson and Yun 2017, Goulart et al. 2023). In conclusion, the traditional EKC suggests that economic growth initially leads to environmental degradation, but improvements follow as income levels rise. However, this view is limited because it overlooks the crucial role that natural resources, such as forests, play in sustaining long-term economic growth (Arrow et al. 1996, Stern 2004). Although deforestation may boost GDP in the short term through activities like agriculture and logging (Barbier 2004, Balboni et al. 2023), it can also deplete natural capital, disrupt ecosystem services, and ultimately undermine economic sustainability (Blaney et al. 2016).

The EKC and the Forest Transition Theory (FTT)

Forest transition refers to the turning point at which a country shifts from net forest loss to net forest gain through natural regeneration or afforestation (Meyfroidt and Lambin 2011). However, forest gain does not necessarily imply ecological recovery, as secondary forests and plantations generally retain lower biodiversity and carbon storage levels than primary forests (Rudel et al. 2005). The EKC is frequently used to frame this dynamic, suggesting that deforestation tends to increase in the early phases of economic development and decline as income rises and societies allocate resources toward conservation. Nevertheless, empirical support for this hypothesis in the context of forest recovery remains ambiguous. Existing research has documented heterogeneous effects of macroeconomic variables, globalization, democratization, and institutional development on forest cover change, indicating that the EKC framework is not universally applicable (Lambin and Meyfroidt 2010, Meyfroidt and Lambin 2011). Furthermore, reestablishing forest ecosystems is often not a passive consequence of income growth but is typically associated with targeted environmental policies and deliberate shifts in land use (Lambin and Meyfroidt 2010).

Many empirical models grounded in the EKC also fail to incorporate critical elements of forest transition theory, such as forest recovery’s timing, duration, and permanence (Barbier et al. 2010). These omissions are significant, as they overlook the heterogeneity of land use trajectories across countries and the socioecological feedbacks that influence them. Rudel et al. (2005) emphasize that forest transitions can yield limited ecological benefits, particularly from plantation monocultures or regrowth on degraded land. They caution that portraying forest transitions as a natural byproduct of economic development may justify delaying necessary political interventions. Collectively, these critiques underscore the theoretical and empirical limitations of applying the EKC framework to forest dynamics and point to the need for more context-specific and policy-aware approaches to understanding forest transitions.

The empirical investigation of the EKC hypothesis typically models the relationship between economic growth and environmental degradation by including GDP and its square as regressors, thus capturing a non-linear relationship (Villanthenkodath et al. 2021, Azam et al. 2024). By transposing the axes of the EKC curve (placing environmental degradation on the horizontal axis and GDP on the vertical) it becomes possible to represent the economic consequences of ecological depletion. This downward-opening parabolic relationship illustrates how environmental degradation may initially accompany economic growth but eventually lead to diminishing returns. From an empirical standpoint, adopting a production function that incorporates the environment as a productive input offers a helpful framework for examining how ecological conditions are associated with economic outcomes across units with different levels of environmental stress.

The transposed environmental Kuznets curve (TEKC)

An essential limitation of representing a production function with the environment as an input is that traditional economic measures account for man-made capital depreciation but overlook the depletion of natural capital, leading to incomplete assessments. By incorporating the environment into the production function, we can better understand the short- and long-term impacts of environmental changes, such as forest reversal, on economic production (World Bank 1992). The production-function approach (PFA) allows for measuring these impacts by examining the effects of ecological changes on production processes. This method involves two key steps: first, identifying the physical impacts of environmental changes on production activities, and second, quantifying the corresponding changes in output (Dosi 2001).

Stock resources (S) are assets whose current use directly impacts their future availability, such as fisheries, forests, or mineral reserves. In contrast, flow resources (F), like solar radiation or wind power, do not diminish with use and remain available for future production (Perman et al. 2003). The distinction between these two resource types is crucial in production function modeling. When stock resources are used in production, the consumption pattern over time must be considered, as overexploitation can reduce future benefits. In dynamic production models, stock resource depletion affects current and future market outcomes. In contrast, flow resources allow for a more static approach because their availability is constant regardless of use. Correctly modeling these variables in the production function is essential for accurate economic output and sustainability predictions (Barbier et al. 2023).

The following model, a modified version of López (1994), represents output as a function of labor, capital, and environmental inputs based on technological parameters, as follows:

(1)

Where y_i is the output of firm i; f_i (K_i,L_i;t) is the conventional production function that depends on the firm’s use of capital (K_i), labor (L_i), and a technological parameter t; S_i represents the stock of environmental resources used by firm i; F_i the flow of environmental resources used by firm i; and τ is the technology parameter specific to environmental inputs, reflecting how innovations or advancements in technology affect the firm’s ability to utilize stock and flow resources more efficiently. The function g_i is increasing and concave for inputs and incorporates the conventional and environmental production factors into the production process. This formulation implies that as firms increase their use of capital, labor, and environmental resources, output will rise, but at a diminishing rate. Furthermore, the function exhibits linear homogeneity in f_i, S_i, and F_i, indicating that output will double if a firm doubles its capital, labor, and environmental inputs. This reflects constant returns to scale across conventional and environmental resources within the production process (López 1994, Perman et al. 2003). Considering environmental stock as the input of interest in the present work, the diminishing rate of output growth to S_i curve is referred to hereafter as the transposed environmental Kuznets curve (TEKC).

Figure 1 illustrates a concave curve depicting the relationship between using environmental stock input (S) and economic product (Y). The marginal product of environmental stock input (MgS) is positive (MgS > 0) when the curve ascends and negative (MgS < 0) when the curve descends. This hypothesized relationship indicates that economic returns from environmental depletion may be positive in the short term but tend to become negative in the long run.

The TEKC is based on the production function approach (PFA), which treats environmental stock, such as forests, as an input in the production process (International Institute for Environment and Development 2003). This framework helps estimate how ecological functions contribute to both production and consumption, recognizing that the depletion of environmental resources impacts not only the creation of goods and services but also their use and availability for consumers. The TEKC further illustrates that resource exploitation may be associated with gains in economic output at lower levels of environmental depletion, but tends to result in diminishing returns as natural stocks are progressively exhausted.

Unlike flow resources, such as solar power, the exhaustion of stock resources like forests has lasting effects on future availability. The TEKC also emphasizes that whereas economic gains may be observed at lower levels of environmental exploitation, higher levels of degradation are associated with serious risks, including potential irreversibility, reduced productive capacity, and ecological destabilization. This degradation impacts soil quality and biodiversity, and exacerbates climate change, posing a severe threat to economic sustainability and ecological balance (Bennett 2017). Therefore, sustainable management practices are crucial, as the long-term hazards of deforestation outweigh any temporary benefits.

Methodological aspects

Environmental impact assessments often rely on non-randomized events because of the inability to randomly assign treatment and control groups, leading to biased comparisons if not carefully addressed (Ferraro 2009). Deforestation exemplifies a non-randomized phenomenon driven by specific human activities such as agricultural expansion, logging, and urban development. This deliberate selection process can introduce significant biases, necessitating methods like propensity score weighting to mitigate these biases.

Comparing results between two treatment groups is straightforward in randomized experiments because the groups are likely similar. However, in non-randomized experiments, such comparisons can be misleading because of systematic differences between the units exposed to different treatments (Rosenbaum and Rubin 1983). Addressing selection bias in non-experimental research is critical for obtaining valid causal inferences.

Propensity score matching (PSM) is a widely used statistical technique designed to mitigate selection bias by approximating a randomized controlled trial. This method involves estimating propensity scores, which represent the probability of receiving the treatment based on observed covariates. These scores then match treated subjects with non-treated counterparts with similar propensity scores, ensuring comparability of observed characteristics. By balancing the distribution of covariates between the groups, PSM reduces the bias arising from non-random assignment to treatment (Rosenbaum and Rubin 1984).

King and Nielsen (2019) critique propensity score matching (PSM), arguing that it frequently exacerbates issues such as imbalance, inefficiency, model dependence, researcher discretion, and statistical bias instead of mitigating them as intended. According to their analysis, PSM’s effort to approximate a wholly randomized experiment, rather than an entirely blocked one, contributes to these shortcomings. Despite these criticisms, the authors do not dismiss using propensity scores in other contexts. They highlight that regression adjustment, inverse probability weighting, and stratification can still be valuable for causal inference when applied correctly. In line with these alternatives, propensity score weighting (PSW) is a robust approach to propensity score analysis because it addresses many of the limitations of PSM. By transforming the sample for causal inference through the retention of all individuals and applying weights based on their propensity scores, PSW allows for a more balanced adjustment for confounding variables, yielding more reliable causal estimates (Narita et al. 2023).

PSW estimates treatment effects in observational studies by adjusting for confounding variables. The propensity score represents the likelihood of receiving treatment based on observed baseline characteristics, typically estimated by using a logistic regression model. The resulting propensity score is the predicted probability of treatment from the fitted model. By applying weights derived from these scores, PSW balances the distribution of observed covariates between treated and untreated groups, enabling a more accurate estimation of treatment effects (Austin 2011).

After estimating the propensity scores (e_x), weights (w) are assigned to treated and untreated individuals to facilitate a balanced comparison between the groups. Specifically, treated individuals are assigned a weight of w = 1/e_x, while untreated individuals receive a w = 1/(1 − e_x). This weighting process creates a “pseudo population” where the distribution of observed covariates is balanced across treated and untreated groups, thereby approximating the conditions of a randomized experiment. In estimating the average treatment effect (ATE), weights account for treatment and control individuals, ensuring a comprehensive population representation. In contrast, when estimating the average treatment effect on the treated, the focus is on the treated group, with treated individuals receiving a weight of 1, while control individuals are weighted as w = e_x/(1 − e_x). These weighting arrangements are critical for reducing bias in observational studies and improving the validity of causal inferences (Guo et al. 2020).

Hence, PSW involves three critical steps to ensure accurate causal inference. First, the propensity scores, which represent the probability of receiving treatment based on observed covariates, are estimated, typically using logistic regression. Second, weights are applied to balance the treated and untreated groups, creating a simulated population that mimics the structure of a randomized trial. Moreover, third, the treatment effect is calculated by using the weighted data to reduce confounding and provide a more accurate causal estimate (Kostouraki et al. 2024).

In addition to PSW, a trimmed regression will be employed in this study. This approach focuses exclusively on municipalities that experienced deforestation, eliminating outliers to create a more robust estimation. Given that all observations in this model are subjected to the treatment, selection bias should be minimal. Nevertheless, because of the cross-sectional nature of the data, heteroskedasticity tests will be conducted to ensure the reliability of the model’s estimates.

DATA AND EMPIRICAL STRATEGY

Database

The data in this research encompass information from 772 municipalities that comprise the Legal Amazon, a region officially designated by the Brazilian government for environmental, social, and economic analysis. The Legal Amazon includes nine states: Acre (22 municipalities), Amapá (16), Amazonas (62), Mato Grosso (141), Pará (144), Rondônia (52), Roraima (15), Tocantins (139), and part of Maranhão (181 municipalities, of which 21 were partially integrated) (Agência de Notícias - IBGE; Santos et al. 2021). Table 1 lists all variables used in the present work and the respective definitions.

Empirical strategy

Propensity score weighting

The analysis of deforestation and economic growth in the Brazilian Amazon was initially conducted by using the entire sample. The variable RDEF served as the basis for constructing the binary variable DEF, which takes the value of zero when the percentage of recent deforestation is less than a unit and one otherwise. The ANOVA results support the decision to use one as the cutoff point for RDEF, demonstrating a statistically significant difference between the two groups.^[2] Deforestation can be considered a non-random intervention; therefore, this initial research phase will be approached as an observational study. Choosing PSW in observational studies preserves a larger sample size than methods like matching, which may exclude observations lacking suitable matches. King and Nielsen (2019) discuss the problem of data pruning in the standard PSM method.

Consider an observational study involving N units, where each unit i(i = 1,2,··· ,N) is assigned a binary treatment indicator Z_i; for the treatment group Z_i = 1, whereas for the control group Z_i = 0. Additionally, each unit is associated with p covariates X_i = (X_1i,···,X_pi). The potential outcomes for each unit, Y_i(1) and Y_i(0), represent the outcomes under treatment and control conditions, respectively. However, only the outcome corresponding to the treatment is observed, while the other remains counterfactual. The observed outcome is thus defined as Y_i = Z_iY_i(1) + (1 − Z_i)Y_i(0) (Zhou et al. 2022). Furthermore, the ATE conditional on X measures the average difference in outcomes between units that received the treatment and those that did not, according to the formula:

(2)

The first step to estimating the PWS is to set up a logistic regression to generate propensity scores, which reflect the probability of an individual receiving the treatment based on their covariates. The logistic regression model, which estimates the probability of receiving the treatment based on the covariates X_i, is expressed as:

(3)

Where Z_i = DEF_i is the binary treatment indicator (0 for control, 1 for treated); X_i = (METRO_i×lGDP_i, lGDP_i, lPOP_i, NET_i, VEG_i, HOT_i, HOMI_i) represents the vector of covariates for unit i, lGDP_i and lPOP_i stands for the natural log of variables GDP_i and POP_i, respectively, and β is a vector of coefficients.

Once the propensity scores e(X_i) = P(Z_i = 1|X_i) are estimated from the logistic regression model, the weights are calculated. For the ATE, the weights are:

(4)

(5)

The weighted outcome model

Using the calculated weights, we perform a weighted regression to estimate the causal effect of the treatment as detailed below:

(6)

Where Y_i = lGDP_i is the response variable in the weighted regression; Z_i = (RDEF_i, RDEF²_i, CDEF_i) is a vector of intervention variables; X_i = (lPOP_i, EDU_i, MOB_i, VEG_i, AGR_i, WASTE_i) is a vector of covariates; γ and δ are vectors of coefficients of the intervention variables and covariates, respectively, and ε_i is a random error.

Two weighted regression models were estimated to assess deforestation’s short- and long-run effects on economic growth. The first model included RDEF_i and CDEF_i as intervention variables, along with all covariates in X_i. The second model included RDEF_i and RDEF²_i, and all covariates in X_i except for VEG_i. The inclusion of RDEF_i and its quadratic term RDEF²_i aims to capture the potential nonlinear relationship between deforestation and economic outcomes. The exclusion of the variable VEG_i in the second model was due to model fitting reasons.

The Breusch-Pagan test for heteroskedasticity used the bptest() function from the lmtest package in R (Zeileis and Hothorn 2002). This test assesses whether the variance of residuals is constant across observations. To address potential heteroskedasticity, a robust variance-covariance matrix was calculated by using the vcovHC() function from the sandwich package (Zeileis 2004), specifying the heteroskedasticity-consistent estimator type “HC1.” The estimation procedure used the coeftest() function from the lmtest package to obtain robust standard errors and coefficient estimates, providing a heteroskedasticity-robust model summary.

The trimmed outcome model

An outcome model was estimated by using a Brazilian Amazon municipalities subsample, which excluded observations with zero recent deforestation (RDEF_i = 0) and outliers above the upper bound of 1.5 times the interquartile range (IQR). This trimming ensures that the analysis focuses only on municipalities with active deforestation, making the sample more relevant for studying the impact of deforestation interventions on economic outcomes. Additionally, removing outliers reduces the influence of extreme values, allowing for a more robust estimation of the relationship between deforestation and economic growth, particularly in the TEKC context. Three regression models are estimated by using the trimmed sample to analyze the relationship between deforestation interventions and economic growth. The general form of the outcome model for each of the three models is:

(7)

Where Y_i = GDP_i represent the response variable in all trimmed outcome models; Z_i^’j is the vector of intervention variables for model (where j = 1,2,3); X_i^’j is the vector of covariates for model j; γ^j and δ^j are the vectors of coefficients corresponding to the intervention variables and covariates for model j, respectively, and ε_i^j is the random error for model j. The vectors of intervention and covariates are, respectively:

a. Model (1) :

(8)

(9)

b. Model (2) :

(10)

(11)

c. Model (3) :

(12)

(13)

The research conducts a Breusch-Pagan test for heteroskedasticity using the bptest() function from the lmtest R package to assess whether the variance of the residuals remains constant across observations. If heteroskedasticity is detected, the analysis will calculate a robust variance-covariance matrix using the vcovHC() function from the sandwich package, applying the ‛HC1’ heteroskedasticity-consistent estimator. Finally, the coeftest() function from the lmtest package will provide robust standard errors and coefficient estimates, ensuring a heteroskedasticity-robust summary of the regression models.

RESULTS

In this section, we present the variables’ descriptive statistics and the results of the empirical models investigating the relationship between deforestation and economic growth. The propensity score model estimated weights for the weighted regression in the outcome model, effectively addressing potential selection bias. This weighted regression evaluates the impact of deforestation while controlling for key covariates such as population, education, and agricultural activity. Additionally, outlier detection and trimming methods were employed to refine the dataset, leading to the estimation of trimmed models that further explore the robustness of our findings. Breusch-Pagan tests addressed potential heteroskedasticity, and robust standard error corrections were applied to ensure reliable inference. Table 2 presents the descriptive statistics for the main variables included in the analysis.

Recent deforestation (RDEF) presents a low mean (2.15%) but exhibits high skewness and kurtosis, indicating concentration in a few municipalities. Cumulative deforestation (CDEF) averages 38%, reflecting widespread historical land cover loss. GDP and municipal population (POP) show extreme skewness and kurtosis, capturing a few urban centers’ economic and demographic concentration. Infrastructure indicators, such as broadband internet access (NET) and mobile phone density (MOB), are more symmetrically distributed, whereas waste collection services (WASTE) display moderate heterogeneity. Environmental variables, particularly hotspots (HOT) and per capita CO₂ emissions (CO2), are highly concentrated, underscoring localized ecological pressure. Educational attainment (EDU) also varies significantly across municipalities.

Propensity score weighting

The following results present the logistic model estimates used to calculate propensity scores for the weighted regression analysis. The model estimates the probability of recent deforestation occurrence (DEF) on the basis of several covariates, including metropolitan status, economic activity, infrastructure, and environmental factors. The sample consists of N = 772 observations, with 470 cases where DEF = 0 and 302 cases where DEF = 1. Table 3 presents the logistic model’s estimated coefficients and corresponding statistical measures.

The logistic regression analysis results offer essential insights into the factors influencing recent deforestation (DEF) in municipalities of the Brazilian Legal Amazon. All estimated coefficients are statistically significant at conventional levels, with significance levels ranging from 0.000 (***) to 0.01 (*). The METRO coefficient suggests that municipalities in metropolitan regions are more likely to experience deforestation, consistent with earlier studies by Geist and Lambin (2002).

Additionally, the coefficient for the natural logarithm of GDP is positive, implying that municipalities with higher GDP are more prone to deforestation, reflecting the potential role of economic activity in driving environmental degradation in the region. Xu et al. (2023) identified a similar relationship between deforestation in the Amazon rainforest and economic development, highlighting population growth as a significant driver of forest loss. Consistent with this, the coefficient for the natural logarithm of population (POP) in the current analysis is positive, suggesting that larger populations will impact an increase in the likelihood of deforestation. These findings corroborate the results of Geist and Lambin (2002), who also recognized population pressure as a contributing factor to tropical deforestation.

The negative coefficient for the variable NET indicates an inverse relationship of this variable with the dependent variable, suggesting that municipalities with higher broadband internet density are less likely to experience deforestation. This result underscores the crucial role of information access in mitigating deforestation by raising awareness and facilitating the dissemination of sustainable land use practices. Information plays a critical role in shaping deforestation dynamics, as it influences decision-making through exposure to market trends, new technologies, and environmental impacts (Angelsen and Kaimowitz 1999). The negative coefficient for VEG suggests that municipalities with higher historical rates of vegetation suppression are less likely to experience recent deforestation. In contrast, the coefficient for HOT demonstrates a positive association between the number of hotspots in a municipality and the likelihood of deforestation. These results highlight the role of past land use patterns and fire activity in influencing current deforestation dynamics. Higher vegetation suppression appears to reduce the availability of land for further deforestation, whereas increased hotspots heighten deforestation risks because of fire-related disturbances.

The positive coefficient for HOMI suggests that higher homicide rates are associated with an increased likelihood of deforestation, a finding corroborated by Magalhães de Oliveira and Varella Miranda (2024). Additionally, the negative parameter for the interaction term between METRO and the natural logarithm of GDP implies that the positive effect of GDP on deforestation is less intense in metropolitan regions. The attenuation could be due to the nature of economic activities in urban municipalities, which are more likely to be concentrated in sectors such as services and industry, as opposed to agriculture and mining sectors that traditionally drive deforestation in rural areas. Consequently, the urban structure of metropolitan regions may reduce the direct pressure of GDP growth on deforestation.

Table 3 also presents the Pseudo R² and chi-squared statistics for the PSW model. The Pseudo R² is a goodness-of-fit measure in logistic regression, similar to the R² in ordinary least squares (OLS) regression. However, it reflects the improvement in model fit with added predictors rather than the proportion of variance explained. It calculates as 1 - (Residual Deviance / Null Deviance), where the null deviance represents a model without predictors, and the residual deviance reflects the deviance of the fitted model. A higher Pseudo R² indicates a better model fit, though interpreting it requires caution, as it does not directly correspond to the R² in OLS regression. Additionally, the chi-squared test evaluates the overall significance of the model by comparing the null deviance to the residual deviance (Hosmer et al. 2013). In this study, the model yields a Pseudo R² of 0.206, suggesting a reasonable fit (Louviere et al. 2000), and the chi-squared statistic significantly rejects the null hypothesis, indicating that including predictors substantially improves model fit (Hosmer et al. 2013).

The weighted outcome model

In this study, we employ a weighted regression model to estimate the impact of deforestation and various control variables on GDP. The weights used in the regression are derived from a previously estimated propensity score logit model, ensuring that the estimation accounts for potential imbalances between treated and control units. The first model assesses the effects of recent deforestation (RDEF), cumulative deforestation (CDEF), and other control variables, including population size (POP), education (EDU), mobility (MOB), agricultural activity (AGR), and waste management (WASTE), on the logarithm of GDP. Furthermore, we conduct the BreuschPagan test to address heteroskedasticity and calculate robust standard errors using a heteroskedasticity-consistent covariance matrix (HC1).

In the second model, we introduce a quadratic term for recent deforestation (RDEF²) to capture the potential nonlinear effects of deforestation on GDP. Similar to the first model, this approach incorporates robust standard errors to ensure reliable inference, controlling for the same variables as in the first model. Both models comprehensively assess how deforestation and control variables such as population and education influence GDP while accounting for heteroskedasticity and the proper weighting of observations based on the propensity score model. Table 4 presents the estimated coefficients, standard errors, diagnostic test results, and goodness-of-fit statistics for intervention models (1) and (2).

The results of Table 4 provide essential insights into the drivers of GDP growth across municipalities. All estimated coefficients are statistically significant at conventional levels, with significance levels ranging from 0.000 (***) to 0.01 (*). In model (1), the intercept reflects an upward baseline for GDP growth, while the Recent Deforestation (RDEF) coefficient suggests a positive impact on GDP. This result likely reflects the short-term economic gains from land clearing, driven by agricultural expansion and resource extraction activities. In contrast, cumulative deforestation (CDEF) presents a negative coefficient, indicating that municipalities with higher levels of accumulated deforestation experience a reduction in GDP growth. This combination of a positive effect from recent deforestation and a negative impact from cumulative deforestation aligns with the TEKC hypothesis. Initially, deforestation stimulates economic growth, but as it accumulates, the adverse environmental effects, such as resource depletion and loss of ecosystem services, begin to outweigh the economic benefits, leading to a decline in GDP growth.

Other variables in model (1) also play critical roles in GDP growth. Population size (lPOP) has a strong positive effect, showing that municipalities with larger populations tend to experience higher GDP growth, likely due to a larger workforce and more significant economic activity. Educational attainment (EDU) is positively associated with GDP growth, reflecting the importance of a skilled labor force in driving economic performance. Mobile phone density (MOB) also positively influences GDP growth, highlighting the role of communication infrastructure in supporting economic development. Also, the suppression rate of primary and secondary vegetation (VEG) positively and significantly impacts GDP growth. This effect could be due to economic activities such as logging, which contribute to economic output without involving complete land clearing. The variable AGR, representing whether agriculture was the highest gross value-added sector in the municipality, positively impacts GDP, underscoring agriculture’s pivotal role in driving economic growth within regional municipalities. Lastly, waste collection services (WASTE) contribute positively to GDP growth, reflecting the broader economic benefits of improved public services and infrastructure. The control variables in model (1), particularly those related to population, education, and infrastructure, had estimated coefficients consistent with the predictions of the Solow-Swan economic growth model (Swan 1956).

In model (2), recent deforestation (RDEF) continues to exhibit a positive and statistically significant impact on GDP growth, consistent with the findings in model (1). However, introducing the squared term for recent deforestation (RDEF²) yields a negative and statistically significant coefficient. This result suggests a non-linear relationship wherein lower levels of deforestation are associated with higher GDP growth, but beyond a certain threshold, additional deforestation begins to exert a negative influence on economic performance. This pattern forms a downward-facing parabola, confirming the TEKC hypothesis. In model (2), the control variables remain consistent with those in model (1). In summary, the results of both models confirm the TEKC hypothesis, where lower levels of deforestation contribute positively to GDP growth but become detrimental beyond a certain point. This relationship is evident in the contrasting effects of recent and cumulative deforestation in model (1) and the quadratic term for recent deforestation in model (2).

The research applied the Breusch-Pagan (BP) test to both models, with the degrees of freedom shown in parentheses in Table 4 corresponding to the number of regressors. The test rejected the null hypothesis of homoscedasticity, indicating heteroskedasticity in the residuals. To resolve this issue, we estimated heteroskedasticity-robust standard errors using the HC1 estimator proposed by White (1980). This adjustment corrects for heteroskedasticity, ensuring more reliable inference by providing robust estimates of the standard errors and enhancing the validity of the statistical conclusions. Regarding model fit, the adjusted R² values were 0.814 for model (1) and 0.810 for model (2), indicating that both models explain a substantial proportion of the variance in GDP growth, with model (1) slightly outperforming model (2) in terms of explanatory power.

The trimmed outcome model

The trimmed regression models focus exclusively on municipalities that experienced deforestation, reducing the sample size to 610 observations. To improve accuracy and minimize the influence of outliers, we removed the top 25% of observations with the highest GDP, which resulted in a final count of 546 observations. This trimming process ensures that the analysis centers on the core relationship between deforestation and economic variables without skewing results because of extreme cases. Unlike the weighted regression models, the trimmed models do not apply propensity weights, as deforestation uniformly affects all selected municipalities, suggesting a homogeneous treatment effect. In observational studies, matching balances covariates between treated and untreated groups to reduce confounding. However, because all observations in this case received the treatment (deforestation), no control group was available for comparison, making matching procedures unnecessary (Brazauskas and Logan, 2016). Table 5 presents the estimated coefficients, standard errors,^[3] diagnostic test results, and goodness-of-fit statistics for trimmed models (1), (2) and (3), offering insights into the effects of deforestation on critical economic variables across the selected municipalities.

Across all three models, coefficients are statistically significant at conventional levels, with p-values ranging from 0.000 (***) to 0.01 (*). The recent deforestation (RDEF) estimates and cumulative deforestation (CDEF) exhibit a consistent pattern, supporting the inverted U-shaped relationship between deforestation and GDP. As predicted by the TEKC hypothesis, deforestation is positively associated with GDP at lower levels, but this association becomes negative beyond a certain threshold. In model (1), the positive coefficient for RDEF suggests that municipalities with lower recent deforestation levels tend to have higher GDP. However, the negative coefficient for RDEF² indicates that this association weakens and reverses at higher levels of deforestation. In model (2), the coefficients for CDEF and CDEF² reveal a similar non-linear relationship, in which moderate forest loss corresponds to higher economic performance. However, extensive cumulative deforestation is linked to economic decline. Model (3) further reinforces this pattern, showing that the combination of recent and cumulative deforestation intensifies the downturn in GDP at higher deforestation levels. These findings illustrate that although forest exploitation may coincide with economic gains in some municipalities, escalating deforestation is structurally associated with diminishing economic performance.

The control variables included in all three models provide additional insights into the factors influencing GDP growth. Population (POP), education (EDU), and infrastructure-related variables (AGR, MOB, WASTE) consistently exhibit positive and statistically significant coefficients, suggesting their crucial contributions to economic development. These findings align with the predictions of the Solow-Swan economic growth model (Swan 1956), which emphasizes the importance of capital accumulation, technological progress, and human capital in driving economic growth. Although the weighted regression models do not include CO2, model (1) reveals a positive and statistically significant effect on GDP when CO2 is added, suggesting a possible link between carbon emissions and economic growth, at least for low levels of deforestation.

Figure 2 illustrates the relationship between deforestation and GDP in municipalities of the Brazilian Legal Amazon, where recent and cumulative deforestation has occurred. The model predicts GDP as a function of Recent Deforestation (RDEF) and its squared term (RDEF²), as well as with Cumulative Deforestation (CDEF) and its squared term (CDEF²), while keeping all other control variables constant at their mean values. This methodology enables the model to isolate the effect of deforestation on GDP by controlling for the influence of different factors. The resulting curves depict how GDP responds to varying levels of deforestation, highlighting a non-linear relationship in which lower levels of deforestation stimulate economic growth, but higher levels of deforestation results in an eventual economic decline. These findings align with the hypothesized TEKC, demonstrating that the economic benefits of deforestation observed at lower levels are not sustainable as deforestation intensifies.

The estimated TEKC for recent deforestation (RDEF) reaches its tipping point at 12% deforestation and a GDP of R$ 486 million. This result indicates that municipalities with a recent deforestation rate greater than 12% are likely to experience a negative relationship between forest clearing and gross domestic product (GDP). In contrast, the tipping point for CDEF is 33% for cumulative deforestation and R$ 455 million in GDP. RDEF is the percentage of the total accumulated deforestation between 2020 and 2022, thus covering the last two years of total accumulated deforestation (CDEF).

From the estimated parameters in Table 5, we observe that the impact of recent deforestation on GDP is approximately 3.4 times stronger than that of cumulative deforestation; the coefficient for RDEF is 16.7, while the coefficient for CDEF is 2.66. Additionally, the absolute value of the RDEF² coefficient is about 18 times that of the CDEF² coefficient. This outcome suggests that factors such as advancements in production technology may have intensified the impact of recent deforestation on GDP. In contrast, the historical deforestation rate has shortened until reaching the tipping point, beyond which further forest clearing will lead to a decline in the municipality’s economic output.

These findings resonate with key mechanisms proposed by Forest Transition Theory (FTT). As observed in the regression results, economic gains from deforestation, particularly those tied to agricultural expansion and logging, are more prominent at lower levels of forest loss but tend to diminish as deforestation intensifies. This aligns with the notion that initial forest clearing is often associated with frontier expansion, where natural capital is abundant and access costs are low. However, as deforestation reaches higher levels, the marginal returns from extractive land uses tend to decline because of reduced soil fertility, biodiversity, and ecosystem services essential to productive capacity (Rudel et al. 2005, Lambin and Meyfroidt 2010). In this context, the observed tipping points in GDP may reflect the onset of ecological constraints that limit the efficiency of traditional resource-based activities, pressuring land use and economic structures to adjust. FTT posits that such transitions toward forest recovery or diversification do not occur automatically with rising incomes, but require enabling institutional, demographic, and policy conditions. In municipalities where these conditions are absent, deforestation may continue without a shift toward forest-compatible development, contributing to economic performance stagnation at higher forest loss levels. The TEKC framework proposed in this study complements FTT by empirically identifying inflection points beyond which extractive development models are no longer economically advantageous.

Although the study identifies a declining pattern in municipal GDP levels across municipalities associated with higher degrees of deforestation, this finding does not necessarily contradict the broader trend of regional economic growth observed in the Legal Amazon. Differences in the scale of analysis and the composition of economic activities offer possible explanations for these distinct dynamics. At the regional level, GDP expansion has been largely driven by a few urban and industrialized centers, where growth is concentrated in services, industry, and commerce sectors less directly linked to land use change (Santos et al. 2022, World Bank 2023). Many of these urban areas are located in municipalities where significant deforestation occurred in earlier decades, and economies have diversified. In contrast, at the municipal level, particularly in rural areas reliant on agriculture, ranching, and extractive activities, economic gains associated with deforestation may exhibit a diminishing returns pattern. Initial forest clearing may offer a temporary stimulus to local economic output; however, it is possible that, over time, factors such as the depletion of natural resources, soil degradation, and the reduction of ecosystem services contribute to limiting productivity and income potential.

CONCLUSIONS

This study investigates the relationship between deforestation and economic growth in Brazilian Legal Amazon municipalities. It employs propensity score weighting and linear regression approaches, while robust regression techniques are used to control for heteroskedasticity and outliers. The results provide evidence supporting the proposed TEKC hypothesis, demonstrating the non-linear relationship between deforestation and GDP in municipalities analyzed in this study.

Through weighted regression models, we identified a relationship between deforestation and GDP that follows a downward-facing parabola. At lower levels, deforestation increases alongside GDP growth up to a certain threshold, beyond which further deforestation leads to a decline in gross domestic product. This pattern was observed for recent deforestation (RDEF) and cumulative deforestation (CDEF), underscoring the non-linear dynamics between deforestation and economic growth in the region. Also, the trimmed regression models identified specific tipping points, indicating when the municipalities’ economic growth halts and declines as deforestation continues. These turning points highlight the critical thresholds for recent and cumulative deforestation, after which its adverse impacts on municipal GDP outweigh the economic benefits of deforestation.

The analysis suggests that the turning point in the deforestation-GDP relationship differs depending on the percentage of land cleared. For recent deforestation (RDEF), the predicted peak GDP has reached a lower rate of forest loss and corresponds to a higher level of economic output. In contrast, for cumulative deforestation (CDEF), the peak occurs at a higher percentage of forest clearing but is associated with a lower predicted GDP. This contrast suggests that municipalities with a larger share of historically cleared land may face diminishing economic returns from further deforestation. Although the cross-sectional nature of the data does not allow for inferences about causality, the results are consistent with the possibility that accumulated deforestation reduces the potential for economic benefit.

These findings emphasize the urgent need for a shift in the economic model toward sustainable practices that foster long-term growth without exhausting natural resources. The results provide critical insights for policymakers, emphasizing the importance of transitioning from deforestation-driven economic activities to sustainable alternatives. Implementing policies prioritizing forest conservation while encouraging sustainable economic activities is essential for long-term regional prosperity. Deforestation in the Brazilian Amazon depletes natural capital, progressively undermining critical ecosystem services such as rainfall regulation, soil fertility, and climate stability. As these services erode, the viability of economic activities diminishes, often resulting in stagnation or long-term decline after an initial resource-driven boom.

Scientific projections suggest that approximately 40% deforestation could trigger the collapse of the Amazon’s hydrological cycle, leading to savannization. This ecological shift would cause irreversible damage to agricultural productivity, biodiversity, and climate resilience, severely threatening regional and national economies. The environmental Kuznets curve (EKC) posits that environmental degradation naturally declines as income rises, but this assumption does not hold in the case of tropical forests. Forest recovery is not spontaneous; it depends on deliberate public policies and active land use management. Ignoring these dynamics overestimates the resilience of forest-based economies and underestimates the risks of continued ecological degradation. A transition toward forest-compatible development, centered on sustainable agriculture, the bioeconomy, ecotourism, and forest restoration, can generate long-term economic growth. Increased productivity, improved land governance, and targeted incentives are essential to decouple economic expansion from environmental degradation and enable a more resilient development path.

__________

^[1] According to the Federal Constitution of Brazil (Brasil 1988), metropolitan regions are groups of nearby municipalities created by state law that comprise a single, integrated urban area.
^[2] Detailed statistical results of the ANOVA analysis can be found in Appendix 2.
^[3] All coefficient estimates and standard errors in the table are reported in thousands of units.

LITERATURE CITED

Angelsen, A., and D. Kaimowitz. 1999. Rethinking the causes of deforestation: lessons from economic models. World Bank Research Observer 14(1):73-98. https://doi.org/10.1093/wbro/14.1.73

Arraes, R. A., F. Z. Mariano, and A. G. Simonassi. 2012. Causas do desmatamento no Brasil e seu ordenamento no contexto mundial. Revista de Economia e Sociologia Rural 50(1):119-140. https://doi.org/10.1590/S0103-20032012000100007

Arrow, K., B. Bolin, R. Costanza, P. Dasgupta, C. Folke, C. S. Holling, B.-O. Jansson, S. Levin, K. Maler, C. Perrings, and D. Pimentel. 1996. Economic growth, carrying capacity, and the environment. Ecological Applications 5(1):13-15. https://doi.org/10.2307/2269539

Assunção, J., P. Barreto, and A. Veríssimo. 2022. O Paradoxo Amazônico. https://amazonia2030.org.br/wp-content/uploads/2023/05/ParadoxoAmazonia_AMZ2030.pdf

Austin, P. C. 2011. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behavioral Research 46(3):399-424. https://doi.org/10.1080/00273171.2011.568786

Azam, M., Z. U. Rehman, H. Khan, and I. Haouas. 2024. Empirical testing of the environmental Kuznets curve: evidence from 182 countries of the world. Environment, Development and Sustainability. https://doi.org/10.1007/s10668-024-04890-1

Balboni, C., A. Berman, R. Burgess, and B. A. Olken. 2023. The economics of tropical deforestation. Annual Review of Economics 15:723-754. https://doi.org/10.1146/annurev-economics-090622-024705

Barbier, E. B. 2004. Explaining agricultural land expansion and deforestation in developing countries. American Journal of Agricultural Economics 86(5):1347-1353. https://doi.org/10.1111/j.0002-9092.2004.00688.x

Barbier, E. B., J. C. Burgess, and A. Grainger. 2010. The forest transition: towards a more comprehensive theoretical framework. Land Use Policy 27(2):98-107. https://doi.org/10.1016/j.landusepol.2009.02.001

Barbier, E. B., A. C. E. Mensah, and M. Wilson. 2023. Valuing the environment as input, ecosystem services, and developing countries. Environmental and Resource Economics 84(3):677-694. https://doi.org/10.1007/s10640-021-00570-0

Bennett, L. 2017. Deforestation and climate change. The Climate Institute, Washington, D.C., USA. https://greenwhs.weebly.com/uploads/1/7/8/5/17857089/deforestation_and_climate_change.pdf

Beuchle, R., F. Achard, C. Bourgoin, C. Vancutsem, H. D. Eva, and M. Follador. 2021. Deforestation and forest degradation in the Amazon: status and trends up to year 2020. EUR 30727 EN. Publications Office of the European Union, Luxembourg. https://doi.org/10.2760/61682

Blaney, R., J. Thorley, T. Salvaterra, and L. Miles. 2016. The economic costs of deforestation: constructing an economic account of the value of lost forest ecosystem services, with Argentina case study. Prepared on behalf of the UN-REDD Programme. UNEP World Conservation Monitoring Centre, Cambridge, UK. https://www.un-redd.org/sites/default/files/2021-10/Economics%20of%20Deforestation%20160927.pdf

Brasil. 1988. Constituição da República Federativa do Brasil. Brasília: Senado Federal. https://www.planalto.gov.br/ccivil_03/constituicao/constituicao.htm

Brazauskas, R., and B. R. Logan. 2016. Observational studies: matching or regression? Biology of Blood and Marrow Transplantation 22(3):557-563. https://doi.org/10.1016/j.bbmt.2015.12.005

Choumert, J., P. Combes Motel, and K. H. Dakpo. 2013. Is the environmental Kuznets curve for deforestation a threatened theory? A meta-analysis of the literature. Ecological Economics 90:19-28. https://doi.org/10.1016/j.ecolecon.2013.02.016

Dosi, C. 2001. Environmental values, valuation methods, and natural disaster damage assessment. CEPAL - SERIE Medio ambiente y desarrollo nº37. United Nations Publication, Environment and Human Settlements Division, Santiago, Chile. https://digitallibrary.un.org/record/445620?ln=en&v=pdf

Feltran-Barbieri, R., B. Felin, and A. Simpkins. 2023. Ending deforestation in the Amazon can grow Brazil’s GDP — but that’s not the only reason to do it. World Resources Institute. https://www.wri.org/insights/zero-amazon-deforestation-can-grow-brazil-gdp

Ferraro, P. J. 2009. Counterfactual thinking and impact evaluation in environmental policy. New Directions for Evaluation 2009(122):75-84. https://doi.org/10.1002/ev.297

Geist, H. J., and E. F. Lambin. 2002. Proximate causes and underlying driving forces of tropical deforestation. BioScience 52(2):143-150. https://doi.org/10.1641/0006-3568(2002)052[0143:PCAUDF]2.0.CO;2

Goulart, F. F., M. J. Chappell, F. Mertens, and B. Soares-Filho. 2023. Sparing or expanding? The effects of agricultural yields on farm expansion and deforestation in the tropics. Biodiversity Conservation 32:1089-1104. https://doi.org/10.1007/s10531-022-02540-4

Grossman, G. M., and A. B. Krueger. 1991. Environmental impacts of a North American Free Trade Agreement. NBER Working Paper No. 3914. National Bureau of Economic Research. https://www.nber.org/papers/w3914

Gunderson, R., and S.-J. Yun. 2017. South Korean green growth and the Jevons paradox: an assessment with democratic and degrowth policy recommendations. Journal of Cleaner Production 144:239-247. https://doi.org/10.1016/j.jclepro.2017.01.006

Guo, S., M. Fraser, and Q. Chen. 2020. Propensity score analysis: recent debate and discussion. Journal of the Society for Social Work and Research 11(3):463-482. https://doi.org/10.1086/711393

Haddad, E. A., I. F. Araújo, R. Feltran-Barbieri, F. S. Perobelli, A. Rocha, K. S. Sass, and C. A. Nobre. 2024. Economic drivers of deforestation in the Brazilian Legal Amazon. Nature Sustainability 7:1141-1148. https://doi.org/10.1038/s41893-024-01387-7

Harbaugh, W. T., A. Levinson, and D. M. Wilson. 2002. Reexamining the empirical evidence for an environmental Kuznets curve. Review of Economics and Statistics 84(3):541-551. http://www.jstor.org/stable/3211570

Hosmer, D. W., S. Lemeshow, and R. X. Sturdivant. 2013. Applied logistic regression. Third edition. Wiley, Hoboken, New Jersey, USA. https://doi.org/10.1002/9781118548387

International Institute for Environment and Development. 2003. Valuing forests: a review of methods and applications in developing countries. Environmental Economics Programme. https://www.iied.org/8116iied

King, G., and R. Nielsen. 2019. Why propensity scores should not be used for matching. Political Analysis 27(4):435-454. https://doi.org/10.1017/pan.2019.11

Kostouraki, A., D. Hajage, B. Rachet, E. J. Williamson, G. Chauvet, A. Belot, and C. Leyrat. 2024. On variance estimation of the inverse probability-of-treatment weighting estimator: a tutorial for different types of propensity score weights. Statistics in Medicine 43(13):2672-2694. https://doi.org/10.1002/sim.10078

Kuznets, S. 1955. Economic growth and income inequality. American Economic Review 45(1):1-28. http://www.jstor.org/stable/1811581?origin=JSTOR-pdf

Lambin, E. F., and P. Meyfroidt. 2010. Land use transitions: socio-ecological feedback versus socio-economic change. Land Use Policy 27(2):108-118. https://doi.org/10.1016/j.landusepol.2009.09.003

López, R. 1994. The environment as a factor of production: the effects of economic growth and trade liberalization. Journal of Environmental Economics and Management 27(2):163-184. https://doi.org/10.1006/jeem.1994.1032

Louviere, J. J., A. D. Hensher, and D. J. Swait. 2000. Combining sources of preference data. Pages 227-251 in Stated choice methods: analysis and application. Cambridge University Press, Cambridge, UK. https://doi.org/10.1017/CBO9780511753831.008

Lovejoy, T. E., and C. Nobre. 2018. Amazon tipping point. Science Advances 4(2). https://doi.org/10.1126/sciadv.aat2340

Magalhães de Oliveira, G., and B. Varella Miranda. 2024. Environmental enforcement, property rights, and violence: evidence from the Brazilian Amazon. Journal of Institutional Economics 20:e27. https://doi.org/10.1017/S1744137424000122

Meyfroidt, P., and E. F. Lambin. 2011. Global forest transition: prospects for an end to deforestation. Annual Review of Environment and Resources 36:343-371. https://doi.org/10.1146/annurev-environ-090710-143732

Narita, K., J. D. Tena, and C. Detotto. 2023. Causal inference with observational data: a tutorial on propensity score analysis. The Leadership Quarterly 34(3):101678. https://doi.org/10.1016/j.leaqua.2023.101678

Pearce, D. W. 2002. The economic value of forest ecosystems. Ecosystem Health 7(4):284-296. https://doi.org/10.1046/j.1526-0992.2001.01037.x

Pereira, R. M. 2019. Commodity boom and environmental policy: what lies behind the Amazon deforestation. Institute for Applied Economic Research (Ipea). IPEA, Brasília. https://repositorio.ipea.gov.br/handle/11058/9590

Perman, R., Y. Ma, J. McGilvray, and M. Common. 2003. Natural resource and environmental economics. Third edition. Pearson Education, Harlow.

Rodrigues, A. S. L., R. M. Ewers, L. Parry, C. Souza Jr., A. Veríssimo, and A. Balmford. 2009. Boom-and-bust development patterns across the Amazon deforestation frontier. Science 324(5933):1435-1437. https://doi.org/10.1126/science.1174002

Rosenbaum, P. R., and D. B. Rubin. 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70(1):41-55. https://doi.org/10.1093/biomet/70.1.41

Rosenbaum, P. R., and D. B. Rubin. 1984. Reducing bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association 79(387):516-524. https://doi.org/10.2307/2288398

Rudel, T. K., O. T. Coomes, E. Moran, F. Achard, A. Angelsen, J. Xu, and E. Lambin. 2005. Forest transitions: towards a global understanding of land use change. Global Environmental Change 15(1):23-31. https://doi.org/10.1016/j.gloenvcha.2004.11.001

Sant’Anna, A. A. 2016. Land inequality and deforestation in the Brazilian Amazon. Environment and Development Economics 22(1):1-25. https://doi.org/10.1017/S1355770X1600022X

Santos, D., R. Salomão, and A. Veríssimo. 2021. Fatos da Amazônia 2021. http://doi.org/10.59346/report.amazonia2030.202103.ed3

Santos, M. L. dos, D. Santos, and A. Veríssimo. 2022. Fatos da Amazônia: Socioeconomia. https://doi.org/10.59346/report.amazonia2030.202301.ed53

Swan, T. W. 1956. Economic growth and capital accumulation. Economic Record 32:334-361. https://doi.org/10.1111/j.1475-4932.1956.tb00434.x

Stabile, M. C. C., A. S. Garcia, C. S. C. Salomão, G. Bush, A. L. Guimarães, and P. Moutinho. 2022. Slowing deforestation in the Brazilian Amazon: avoiding legal deforestation by compensating farmers and ranchers. Frontiers in Forests and Global Change 4:635638. https://doi.org/10.3389/ffgc.2021.635638

Stern, D. I. 2004. The rise and fall of the environmental Kuznets curve. World Development 32(8):1419-1439. https://doi.org/10.1016/j.worlddev.2004.03.004

Tateishi, H. R., C. Bragagnolo, and A. N. de Almeida. 2021. Forest, agriculture and land conversion: environmental efficiency in Brazilian Amazon rainforest. Forest Policy and Economics 133:102615. https://doi.org/10.1016/j.forpol.2021.102615

Ulucak, R., and F. Bilgili. 2018. A reinvestigation of EKC model by ecological footprint measurement for high, middle, and low income countries. Journal of Cleaner Production 188:144-157. https://doi.org/10.1016/j.jclepro.2018.03.191

United Nations Environment Programme. 2023. Last chance to save the Amazon. https://www.unep.org/news-and-stories/statements/last-chance-save-amazon

Villanthenkodath, M. A., M. Gupta, S. Saini, and M. Sahoo. 2021. Impact of economic structure on the environmental Kuznets curve (EKC) hypothesis in India. Journal of Economic Structures 10:28. https://doi.org/10.1186/s40008-021-00259-z

Wang, Q., X. Wang, R. Li, and X. Jiang. 2024. Reinvestigating the environmental Kuznets curve (EKC) of carbon emissions and ecological footprint in 147 countries: a matter of trade protectionism. Humanities and Social Sciences Communications 11:160. https://doi.org/10.1057/s41599-024-02639-9

White, H. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48(4):817-838. https://doi.org/10.2307/1912934

World Bank. 1992. World Development Report 1992: development and the environment. https://doi.org/10.1596/978-0-1952-0876-4

World Bank. 2023. A balancing act for Brazil’s Amazonian states: an economic memorandum (International Development in Focus). Washington, D.C. https://www.worldbank.org/en/country/brazil/publication/brazil-a-balancing-act-for-amazonian-states-report

Xu, J., Q. Zeng, and Z. Zhang. 2023. The relationship between Amazon rainforest deforestation and economic development. Highlights in Business, Economics and Management 5:273-278. https://doi.org/10.54097/hbem.v5i.5085

Zeileis, A. 2004. Econometric computing with HC and HAC covariance matrix estimators. Journal of Statistical Software 11(10):1-17. https://doi.org/10.18637/jss.v011.i10

Zeileis, A., and T. Hothorn. 2002. Diagnostic checking in regression relationships. R News 2(3):7-10. https://journal.r-project.org/articles/RN-2002-018/RN-2002-018.pdf

Zhou, T., G. Tong, F. Li, L. E. Thomas, and F. Li. 2022. PSweight: an R package for propensity score weighting analysis. R Journal 14(1):282-300. https://doi.org/10.32614/RJ-2022-011