Comparing market instruments for forest conservation in Brazil using farm-level census data

Abstract

Almost 40 per cent of Brazil's native vegetation is located on over five million private properties. This study assesses the potential of agricultural land taxes and tradable forest certificates for conserving Brazil's fragmented native vegetation across commercial farms, using micro census data from 2006 and 2017. We explore the variability of optimal tax rates and market prices for forest certificates, revealing a supply-demand imbalance in the Amazon and high sensitivity of conservation outcomes to changes in farmland opportunity costs, especially in productive areas. Despite a more positively skewed distribution of opportunity costs by 2017, market outcomes remained unaffected. Notably, expanding the market to include the Amazon's agricultural frontier microregions could achieve 45 per cent of the conservation target. Our analysis underscores the interplay between market-based conservation mechanisms and regional agricultural economics, highlighting the need for tailored approaches to optimize conservation efforts.

1. Introduction

Conserving fragmented forests spread across millions of farms is challenging and requires costly monitoring and enforcement. In Brazil, around 40 per cent of the native vegetation is spread across roughly five million private properties. That equates to about 200 million hectares (ha) of forest, savanna, and wetlands, an area a little larger than the Midwest region of the United States. Despite improvements in monitoring technology, deforestation in the Amazon has increased from 7,536 square km in 2018 to 11,600 square km in 2022, which represents a 53 per cent increase in five years (INPE, 2023).¹ Due to these continuous losses, scientists and policymakers have been studying market solutions for conserving natural vegetation within private properties (Panayotou, 1994; Chomitz, 2004; Sparovek et al., 2012; Soares-Filho et al., 2014; Alix-Garcia et al., 2015; Soares-Filho et al., 2016; Alix-Garcia et al., 2018; Souza-Rodrigues, 2019). Tax schemes and tradable forest certificates are among the solutions that economists and policymakers have considered to incentivize landowners to allocate their conservation efforts based on the cost of restoring or maintaining native vegetation (Soares-Filho et al., 2016; Souza-Rodrigues, 2019). However, it remains unclear whether these market instruments can break Brazil's long cycle of deforestation.

This study explores the effectiveness of two market-based instruments – a tax on agricultural land and a market for forest certificates – in conserving Brazil's fragmented native vegetation across millions of farms. These instruments are designed to encourage landowners to weigh their conservation efforts against the opportunity cost of farmland (OC), defined as the foregone income from not farming the land, and the costs associated with reforesting and maintaining forested areas. For example, a rancher might reforest part of a pasture to avoid the agricultural land tax, while a soybean producer might pay the tax to continue farming. Conversely, in a forest certificate market, soybean farmers could purchase certificates from ranchers who, in turn, dedicate more land to conservation or reforestation.² Both methods leverage market mechanisms to find the most cost-effective conservation solutions, balancing agricultural development with environmental conservation.

Tradable Forest Certificates (Cotas de Reserva Ambiental in Spanish) are an example of a market mechanism for forest conservation in Brazil, established by the revised Brazilian Forest Code (Law 12.651/2012). The 2012 reform of the Forest Code aims to regulate land use on private properties to protect the country's natural vegetation and promote biodiversity and ecosystem services, such as freshwater conservation. It introduces requirements for farms to maintain a legal reserve and permanent preservation areas, with specific proportions of a farm to be left in their natural state. Farmers with highly productive and valuable land (high OC) can fulfill their conservation obligations by purchasing certificates from those with less economically viable land (low OC).³

The effectiveness and cost of using market instruments for forest conservation depend on the hard-to-measure variability in OC over space and time. The spatial variability in OC plays a crucial role as it determines the trading possibilities between farmers who own lands of different values. For example, if a market comprises only soybean farmers with similar land quality, there would be limited trading opportunities. Additionally, farmers are less likely to commit to conservation if they expect their land's profitability to increase due to technological advancements or market shifts.⁴ However, measuring OC variability at the farm level is challenging as it requires detailed land use and quality data across a vast area, such as different state-biome combinations in Brazil. Unfortunately, publicly accessible farm-level profitability data is unavailable in Brazil. Even municipality-level profitability information is scarce, and researchers often have to rely on broader regional estimates or industry sources to infer municipal profitability, which provides an important reference for policy analysis but tends to average out the spatial variability in OC and hinder the study of temporal variation in OC within disaggregated markets for forest conservation.⁵

The main contribution of our study is to assess how the spatial and temporal variability in the OC affects the potential of market instruments for the conservation of natural vegetation within Brazilian farms. We use confidential farm-level data from the Brazilian Agricultural Census (IBGE, 2006, 2017) for 1.1 million commercial farms to infer the farm-level OC from farmers’ land use choices. We then use our farm-level measures of the OC to estimate the market supply function of reforested land, which specifies the amount of land reforested (or afforested) at a given price, for 44 markets in Brazil.⁶ A supply function of reforested land for conservation is an abatement function similar to the supply function of carbon sequestration used to study markets for carbon sequestration (Lubowski et al., 2006, 2008) and to the deforestation cost curve used in the analysis of payments for environmental services (Börner et al., 2010). Finally, for each of these 44 state-biome markets in Brazil, we estimate the tax and the market prices of forest certificates to achieve the legal conservation requirements in two census years, 2007 and 2017. We then complete a series of sensitive analyses to assess the effectiveness of the two market instruments when the OC varies over time.

Our analysis indicates that market mechanisms for conserving forests could impact Brazil's three major biomes – the Amazon, Cerrado, and Atlantic Forest – differently. In the Amazon, a tax on agricultural land could advance conservation goals set by the Forest Code. However, its effectiveness, measured by the amount of reforested land, may vary with OC changes. We also assessed a novel market design for the Amazon's agricultural frontier, proposing an expanded forest certificates market. This approach could meet 45 per cent of the Amazon's conservation objectives at approximately $100 per ha. The expanded frontier market is not very sensitive to small changes in OC.

In the Cerrado biome, implementing either a tax on agricultural land or expanding the market for forest certificates can effectively meet the Forest Code (FC) conservation targets. The tax rates and market prices for forest certificates in the Cerrado are comparable. Notably, the São Paulo and Mato Grosso markets within the Cerrado face the highest costs and exhibit significant sensitivity to OC fluctuations. A substantial increase in OC, measured at 2 standard deviations, could diminish reforestation efforts by about 30 per cent under tax policies in these areas and escalate conservation costs by more than 150 per cent.

In the Atlantic Forest biome, like the Cerrado, the tax mechanism and forest certificate market can achieve the conservation targets across all state-biome markets, albeit with generally higher taxes and prices. The markets in the Atlantic Forest are also more responsive to OC shifts. A modest increase in OC by 0.5 standard deviations could potentially double the conservation costs for farmers, indicating a higher volatility in the Atlantic Forest markets than in the Cerrado. The results for the three biomes suggest that policymakers should consider combining market instruments with expanding permanent conservation areas.

Our study contributes to a growing literature that studies market instruments for environmental conservation. Our empirical analysis follows most closely the modeling of Lubowski et al. (2006, 2008) and Souza-Rodrigues (2019). In particular, Souza-Rodrigues (2019) assesses the potential of an agricultural land tax to conserve the Amazon rainforest using an aggregated version of the agricultural census dataset. Our tax estimates for the Amazon biome are consistent with the estimate by Souza-Rodrigues (2019); however, we find a large variation in the optimal tax within the Amazon biome, ranging from $18/ha in the Rondonia–Amazon market to $414/ha in the Tocantins–Amazon market. Several other authors have studied the potential for tradable allowances for forestland in the context of the revised FC (Sparovek et al., 2012; May et al., 2015; Silva et al., 2019). Our price estimates are consistent with other state-level simulations that use more disaggregated data (Chomitz, 2004 simulates the Minas Gerais markets using data at the municipality-farm size level) but are lower than estimates using more aggregated measures of the OC (Soares-Filho et al., 2016).

Several other studies have investigated the potential for incentive mechanisms such as payments for ecosystem services to reduce deforestation (Pattanayak et al., 2010; Busch et al., 2012; Mason and Plantinga, 2013; Alix-Garcia et al., 2015; Jayachandran et al., 2017). These studies assess incentives for avoiding deforestation in locations with multiple types of land tenure. Our analysis complements these studies by focusing on reforestation (or afforestation) in private agricultural land. Börner et al. (2010) assesses the economics and institutional conditions for implementing payments for ecosystem services in the Amazon. Börner et al. (2010) simulates a cost function for avoided deforestation. The concept is similar to our estimated supply function of reforested land. However, as we focus only on private agricultural land, our estimated function would be more closely related to the nonlinear part of the avoided deforestation function, which corresponds to the forest frontier.⁷ A more extensive literature investigates the effect of governmental interventions in reducing deforestation in the Amazon (Pfaff, 1999; Börner et al., 2010; Nepstad et al., 2014; Assunçao et al., 2015; Alix-Garcia et al., 2018; Miranda et al., 2019).⁸ Our study departs from previous analyses of forest conservation in Brazil by estimating an empirical model of the OC to study and contrast market instruments for forest conservation across all biomes in Brazil. This study provides the first empirical estimates for a supply function of reforested land in Brazil using farm-level census data.

The article is organized as follows: Section 2 introduces an optimal reforestation model guiding our analysis. Section 3 outlines the datasets used. Section 4 presents the empirical analysis, while section 5 discusses market simulation results. Section 6 summarizes policy implications. The online appendices provide detailed dataset descriptions (appendix A), explain the discrete choice model of land use (appendix B), report additional results (appendix C), and describe a robustness analysis (appendix D).

2. Optimal reforestation with variation in OC

Our study employs a simple model of optimal reforestation to demonstrate how the spatial and temporal variability in OC influences both the total area reforested and the overall conservation cost to farmers. The model, depicted in figure 1, determines the optimal level of reforestation by balancing the marginal benefit against the marginal cost of reforestation, aiming for the regulator's minimization of conservation costs (Stavins, 1996; Muller and Mendelsohn, 2009; Keohane and Olmstead, 2016). The marginal benefit for each market is set at the forest debt level mandated by the FC, reflecting the required reforestation area ($Area_0$). The government, through FC land use regulations, establishes the marginal benefit function, acknowledging that many forestland benefits are market-independent and challenging to quantify. Conversely, the marginal cost of reforestation, including OC and expenses on reforesting, monitoring, and protection, is market-quantifiable. Our analysis primarily focuses on the OC, assuming constant monitoring and protection costs.

Figure 1. Optimal reforestation with spatial and time variation in OC. (a) Tax on agricultural land, (b) Market of forest certificates.

Notes: Figure 1 illustrates a simple model of optimal reforestation with spatial and time variation in OC for two market instruments: a tax on agricultural land (panel a), and a market for tradable forest certificates (panel b). The curvature of the OC functions represents the spatial variation in the OC. The time variation is illustrated by the difference in the three functions plotted for time 0, time 1, and time 2: OC ₀, OC ₁, OC ₂, respectively. At time 0, the regulator can set a tax or a quantity for area reforested equal to Area ₀ and allow for trading among farmers (the market for forest certificates). If the OC increases at time 1, the OC function shifts to the left, and the area that is reforested under the tax and the market instruments will differ. Under the tax, the area reforested decreases to Area ₁. A tax policy, therefore does not ensure the provision of the targeted reforestation area. Under the market instrument, the area reforested does not change, but the market price for forest certificates increases to MktPrice ₁, and the overall conservation cost to farmers is higher.

Our analysis examines the mechanics of a tax on agricultural land and a market for forest certificates as mechanisms to achieve the reforestation target, represented by $Area_0$. In the tax scenario (figure 1, panel a), the optimal tax level is set such that reforestation up to $Area_0$ is economically advantageous for farmers due to the tax being higher than the OC for areas below $Area_0$. Beyond this point, the rising OC makes it more profitable to pay the tax and continue farming. Conversely, in the forest certificate market model (figure 1, panel b), the policymaker sets the total amount of reforestation $Area_0$ and the equilibrium market price $Mkt \ Price_0$ is determined by the market as farmers trade forest certificates. Both the tax and market mechanisms can achieve optimal reforestation at an identical conservation cost, provided that the OC is accurately assessed and remains consistent at $OC_0$. (See Weitzman, 1974 for an analysis of the comparative advantage of price instruments under OC uncertainty).

Suppose the OC changes over time due to technological advancements or market shifts. In that case, the extent of reforestation and the associated conservation costs will also change. The difference between a tax and a market-based approach becomes apparent in such a scenario. For instance, if the OC increases, as indicated by a shift of the OC function to the left ($OC_1$ in figure 1), the two mechanisms would be impacted differently. With a tax, while the tax rate stays constant, the reforested area would decrease from $Area_0$ to $Area_1$ (panel a). On the other hand, in a forest certificates market, the reforestation area remains the same, but the market price would increase from $Mkt \ Price_0$ to $Mkt \ Price_1$, leading to higher conservation costs for farmers (panel b).

Should the OC increase further to $C_2$, the market might not find an equilibrium price due to an insufficient supply of forest certificates to meet the conservation goal. However, even with $OC_2$, the tax mechanism would still promote the reforestation of $Area_3$ ha.

The OC function illustrated in figure 1 essentially acts as a supply function for reforested land, linking the quantity of land reforested to its corresponding price. This function mirrors the marginal cost of producing the supplied goods – in this case, reforested land. The shape of this supply curve reflects spatial variations in OC across different farms. A constant OC across all farms would result in a horizontal supply line. For instance, a market comprising large ranches with low OCs would feature a supply line below the optimal tax, whereas a market dominated by commercial sugarcane plantations, with higher OCs, would have its supply line above the optimal tax. The sensitivity of reforestation levels to tax rates is pronounced in scenarios with constant OCs: setting a tax below the constant OC would lead to no reforestation, while a tax above it would result in full reforestation. Additionally, in a market characterized by a constant OC, trading would offer no advantages. While the concept of a constant OC serves as an illustrative tool, it oversimplifies the reality given the substantial spatial variation in land quality, rendering the supply function for reforested land a nonlinear equation relative to the area reforested. In subsequent sections, we delve into how farm-level data is employed to empirically derive the supply function for reforested land across different state-biome markets in Brazil.

3. Data

3.1. Farm-level agricultural census data

The analysis primarily relies on the 2006 and 2017 agricultural census surveys conducted by the IBGE (2006, 2017). These surveys gather data from over five million farmers every ten years to provide information on farm and farmer characteristics, including crop-level land use choices, production output and technology, farm size, farm water access, farm production value, compliance with the FC, and share of commercial silviculture. Online appendix A provides a detailed description of each variable.

Our analysis focuses on medium and large commercial farms in Brazil that are more likely to participate in markets for forest certificates due to their formal land ownership and trading requirements. Small farmers are not obligated to reforest their land up to the legal reserve requirement after the latest revision of the FC. Therefore, we concentrate on medium and large commercial farms, as they specialize in one type of land use, making it easier to estimate the land use model. We define commercial farms based on their total production value and farm size, following the analysis of Alves et al. (2013).⁹

To identify our preferred sample of commercial farms, we considered “All Farms,” which includes all farms in Brazil with a production value above 2 MW or R$7,200 (2006) and farm size above 5 ha. This sample comprises 1,195,450 commercial farms in 2006 and 1,129,400 in 2017. We also tested our models and simulations using alternative farm classification samples. According to Alves et al. (2013) analysis, around 10 per cent of farmers in Brazil with production value above 10 MW generate approximately 86 per cent of agricultural production value. We thus created a sample of “Large Farms” for commercial farms with gross yearly revenue above 10 MW and farm size above 50 ha. This sample has approximately 230 thousand farms in the 2006 census and 242 thousand in the 2017 census. Our empirical specifications control for farm size and state-biome fixed effects and yield consistent results using alternative samples.¹⁰

We have merged the data from the agricultural census with information about the potential crop yield, the cost of transporting agricultural products, and the socioeconomic characteristics of the municipality. This helps us to determine the relative profitability of each land use and account for market access in our discrete choice model.

3.2. Potential yield data

The International Institute for Applied Systems Analysis (IIASA) and the Food and Agriculture Organization (FAO) have estimated the potential yield of 154 crops under three levels of land management and input use. They have differentiated between rain-fed and irrigated farming. The estimates are based on crop models that are applied to a global climate, soil, and terrain information dataset. The maximum potential and agronomically attainable yields have been calculated by IIASA/FAO.¹¹

Given the focus on commercial farms, we are using potential yield measurements for high input use and rain-fed farming. This is because less than 3 per cent of agricultural production in the sample uses irrigation. IIASA/FAO also computes actual yields and yield gaps for their Global Agroecological Zones (GAEZ) dataset. However, for this project, we are using only agro-climatic potential yield estimates that are averaged over 30 years (1961–1990) before the study period. It is important to note that these potential yield measurements are theoretical estimates from general production functions and do not reflect the actual production decisions of Brazilian farmers. For a detailed description of the GAEZ methodology, please refer to IIASA/FAO (2012).

GAEZ agro-climatic potential yields offer two benefits for estimating OC. Firstly, IIASA and FAO provide potential crop yield estimates for millions of small grid cells with a latitude and longitude of 0.5$^\circ$ by 0.5$^\circ$. These estimates match well with farm-level census data, enabling us to model farm-level variability in OC. A geographical information system combines IIASA/FAO's potential yield measures for soy and alternative crops such as sugarcane, rice, cotton, coffee, and corn with the census dataset at the census block level. There are approximately 70,000 rural census blocks in Brazil. Secondly, IIASA/FAO's potential yield variable is determined independently of Brazilian farmers’ choices, thereby reducing the potential for bias in OC estimation.

3.3. Complementary datasets

To account for the profitability differences of crops in different parts of Brazil, we used the variation in transportation cost. First, we estimated transportation costs using freight data from the System of Freight Information (SIFRECA) maintained by the Luiz Queiroz College of Agriculture at the University of São Paulo. Since most agricultural production in Brazil is transported by road, SIFRECA has data on the average transportation cost by road routes for the main agricultural products in Brazil. We estimated the transportation costs using a quadratic function of the distance traveled using the SIFRECA route/product dataset. The transportation cost dataset is a subset of the SIFRECA annual report and contains 1,048 routes: 625 for soy transportation and 423 for corn transportation. We then calculated the transportation cost of agricultural products, excluding cattle, for each rural census block in Brazil using the estimated transportation cost model. To estimate the distance to the market, we used the straight-line distance from each census block to the closest large city or port.¹²

Finally, we have complemented the dataset with additional information on municipality characteristics. This includes variables such as income per capita and population density, which helps us understand the varying market access and demand for agricultural products across Brazil (IBGE, 2006). We have also included the mean elevation and standard deviation of elevation to measure the suitability of the land for mechanized agriculture, as well as the additional demand for forestland due to the FC requirements for native vegetation on hilltops (IBGE, 2006). Table A1 in online appendix A provides a summary of the dataset for commercial farms. Panel A presents the land use area for each biome in Brazil for the census year 2006, while panel B reports the land use area for the census year 2017.

4. Empirical analysis

We have used a nested logit discrete choice model to study how farmers decide their land use (Bell et al., 2006; Lubowski et al., 2006, 2008; Souza-Rodrigues, 2019). The process of decision-making is illustrated in figure A5 in the online appendix. The first decision that a farmer has to make is whether to use a piece of land for farming or to leave it as a forest. If the farmer chooses to use the land for agriculture, they select a specific farming activity, such as grazing, cereals, or soy cultivation. This enables us to evaluate the profitability of different land uses. Our analysis found that grazing is the most common choice among Brazilian farmers, influencing their decision to convert forest land to agricultural use. Simulations prioritize grazing as the benchmark farm type and examine outcomes with different pasture conditions.^{13, 14} The OC is, by definition, the maximum profitability of a plot of land, which can be calculated using the inclusive value formula of the nested logit model.

We performed a simulation to implement a tax and a market for forest certificates in Brazil for every census year. The simulation involved three steps. Firstly, we estimated the OC for Brazil's 1.1 million commercial farms. Secondly, we estimated the land use choices of farmers, i.e., whether to engage in agricultural production or forestry, using the farm-level OC calculated in the first step. Finally, we simulated the empirical market supply function of reforested land for each of the 44 markets in Brazil and estimated the market price and the optimal level of reforestation. This section provides a detailed explanation of the three steps of our analysis and presents the empirical results.

4.1. Estimation of the opportunity cost

To begin our analysis, we need to estimate the OC by observing the land use choices made by 1.1 million commercial farms in Brazil. Specifically, we employ a conditional logit model for grazing and the six largest crops in Brazil, namely soybeans, corn, cotton, rice, sugarcane, and oranges. This model converts the variations in potential yield and transportation costs into an index of the OC of forestland.

We estimate the OC by estimating the relative variation in maximum net revenue (MNR) across crops and locations. To begin with, we'll explain how the interaction between maximum potential yield and transportation cost drives variation in the MNR for each type of agricultural land use. Let's assume the farmer's profit for commercializing crop $j$ is $\pi _j=(p_j-tc_j)y_j-wx$, where $y$ represents the farmer's production function and $x$ is the input quantity. Here, $tc$ captures the transportation cost per ha. We consider a quadratic production function, $y_j(x)=\alpha _j- ({1}/{\lambda _j}) (\beta _j-x)^2$, where $\alpha _j$ is the maximum potential yield that a farmer can achieve when producing crop $j$ at the optimal input level $\beta _j$, and the parameter $\lambda _j$ captures the sensitivity of production to variations from the optimal input level.¹⁵ Solving the farmer's profit maximization problem using the quadratic production function, we obtain the optimal profit function for crop $j$:

(1)\begin{equation} \pi_j^*= \alpha_j (p_j-tc_j ) - \beta_jw + \frac{\lambda_j}{2} \frac{w^2}{(p_j-tc_j )}. \end{equation}

The MNR for farmer i and crop j, $MNR_ij$, is the term $\alpha _j (p_j-tc_j )$ in the optimal profit function.¹⁶ The potential yield variable, $\alpha _j$, captures the variation in land quality. The interaction $\alpha _j p_j$ represents the maximum potential revenue, and the interaction $\alpha _j tc_j$ measures the revenue penalty due to transportation costs. For farmers close to ports or markets, the transportation cost is low regardless of land quality. Nevertheless, if the farmer is far from the market, the interaction between the potential yield and transportation cost varies significantly with differences in land quality. As transportation costs increase, the interaction term $\alpha _j tc_j$ plays an important role in determining the profitability of alternative crops in Brazil. We estimate the conditional logit model using the variation in MNR:

(2)\begin{equation} C_i = {\rm arg\,max}_j \left\{ \left\{ \delta_{0j}+\delta_1 MNR_{ij} + \delta_2 Z_m^{'} MNR_{ij} + \delta_{3j} Z_m + \delta_{4j} F_i + \epsilon_{ij} \right\}_{(j=1)}^J,0 \right\}, \end{equation}

where $C_i$ is the farmer's land use choice, $\delta _j$ is an alternative-level constant, $MNR_{ij}$ is the maximum net revenue of land use $j$ at farm $i$, and $Z_m$ is a vector of municipality characteristics and municipality population density. We include the interactions between the alternative-level variable, $MNR$, and municipality mean elevation and population density. $F_i$ is a vector of the farmer's characteristics, transportation cost between the census block and the closest large city or port, log of farm size, and municipality income per capita.¹⁷

Tables A3 and A4 in the online appendix contain the findings from the conditional logit model of Brazil's land use for 2006 and 2017. The reference land use primarily comprises grazing, as most of Brazil's agricultural land is used for pasture.¹⁸ The estimated coefficients show the impact of various factors on the selection of each crop compared to grazing. We find that MNR is a significant predictor of land use selection. Furthermore, the interactions between MNR and the municipality's average elevation and population density are also substantial. They indicate that the impact of MNR on land use selection tends to decrease slowly in areas with higher elevation or higher population density.¹⁹

We use the inclusive value formula for the logit discrete choice model to estimate the OC for each commercial farm $i$:

(3)\begin{equation} \widehat{OC}_i = Log \sum_{j=1}^J Exp \left( \widehat{\delta}_{0j} + \widehat{\delta}_1 MNR_{ij} + \widehat{\delta}_2 Z_m^{'} MNR_{ij} + \widehat{\delta}_{3j} Z_m + \widehat{\delta}_{4j} W_i \right). \end{equation}

4.2. Modeling the choice of agriculture or forest

The second step in our empirical analysis models the farmer's land use choice of agriculture or forest using a logistic share equation,

(4)\begin{align} Log \frac{share_{agriculture,i}}{1- share_{agriculture,i}} = \theta_0 + d_{Bio \times UF} + \theta_1 \widehat{C}_i + \theta_2 TC_i + \theta_3 W_i + \theta_4 CO_i + \theta_5 F_i + \nu_i, \end{align}

where $\widehat {OC}_i$ is the estimated OC of forestland for farm $i$. The transportation cost from the census block to the nearest large city or port is represented by $TC$. $W$ is a vector of dummy variables that indicate the availability of water sources in the farm. These water access variables capture the effects of forestland requirements for protecting native vegetation along freshwater sources. We also use a set of biome/state fixed effects called $d_{Bio \times UF}$ to account for market-specific unobserved factors that affect agricultural productivity and deforestation, and the dummy variable $C$ at the municipality level to identify compliance with the forest reserve requirements of 1996. This compliance variable helps us control for conservation policy enforcement efforts.²⁰ The logistic share equation also has a vector $F$ of farm characteristics that capture demand for agricultural products, forestry products, and land suitability for mechanized farming. The vector $F$ includes farm size, municipality elevation, population density, and the share of commercial silviculture. By analyzing these factors, we can identify locations with high agricultural productivity and historical deforestation that may require monitoring and enforcement measures.

We find a negative nonlinear relationship between the percentage of forestland within a farm and the OC. Figure 2 shows the nonparametric relationship between the percentage of forestland and our estimate of the OC for census years 2006 and 2017.²¹ If the OC index goes up from 0.2 to 0.3, the percentage of forestland within a farm would decrease by 10 per cent. These results suggest that the percentage of forestland on low-quality land is susceptible to changes in the OC. On land with the highest OC, the percentage of forestland tends to increase above 0.20, likely due to targeted monitoring and enforcement efforts.

Figure 2. Share of forestland and the OC.

Notes: Figure 2 plots a nonparametric relationship between the share of farmland allocated to native vegetation and our estimate for the OC for census years 2006 and 2017. Figure 2 is a binscatter regression of the share of forestland within a farm and the OC using the dataset of 1.1 million commercial farms and all the explanatory variables included in equation (1). The binscatter plot breaks down the forestland share and the OC variable into 100 bins after partialling out the effect of the other explanatory variables, including the biome-state fixed effects. It is a graphical representation of a nonparametric regression.

Table 1 presents the results of the estimation of the land use model in equation (4). The dependent variable is the logit log transformation of the agricultural land share. The OC has a significant positive effect on the share of agricultural land across the specifications. A one-point increase in the OC index increases the ratio of agricultural land to forestland by 23 per cent. Municipalities in compliance with the FC and municipalities with a higher share of commercial silviculture have a higher share of forestland (lower share of agricultural land). All the indicator variables for water access have a significant negative effect on the share of agricultural land, suggesting that farmers comply with the requirement for forest buffers around water sources. Also, farms in regions with higher elevations tend to have more forestland, likely because of the limitations to mechanized farming and legal requirements for forest buffers close to hilltops. We find consistent results for the model across the specifications when we add the state, biome, and biome-state fixed effects.

Table 1. An empirical model of the log of agriculture land share within farms

Notes: Standard errors are bootstrapped with 700 iterations and clustered at the municipality level. The “All Farms” sample refers to commercial farms defined as farms with annual gross revenue above 2 MWs and farm size above 5 ha. The “Large Farms” sample refers to commercial farms defined as farms with annual gross revenue above 10 MWs and farm size above 50 ha. The minimum wage in 2006 was R$300 and the minimum wage in 2017 was R$937.

4.3. Simulation of the supply of forestland for conservation

As the final step of our analysis, we conduct a simulation to determine the supply function of forestland for each of the 44 markets in Brazil. The simulation involves incentivizing reforestation in the empirical land use model established in Step 2, calculating new land allocation between forest and agriculture for each commercial farm, and then aggregating the results to determine the amount of reforested land.

The reforested land supply equation is derived from the agricultural land share equation by constraining the coefficient of transportation cost to minus one to set the unit of measurement to Reals per ton:

(5)\begin{align} Log \frac{share_{agriculture,i}}{1- share_{agriculture,i}} = \widehat{\theta}_0 + d_{Bio \times UF} + \widehat{\theta}_1 \widehat{OC}_i - TC_i + \widehat{\theta}_3 W_i + \widehat{\theta}_4 CO_i + \widehat{\theta}_5 F_i + \tau . \end{align}

We can use equation (5) to simulate the area of land that can be converted into a forest by varying the payment to the farmer (denoted by $\tau$). The optimal tax or equilibrium market price is the $\tau$ value that incentivizes farmers to reforest an area equal to the forest debt in the market. Although $\tau$ is an annual expense, the optimal tax and market price are represented as a single payment to the farmer. To convert the simulated incentive from 2006 Reals per ton ($\tau$) into a single payment expressed in 2017 US$, we multiply it by a factor of 0.6, which accounts for the 2017 Real US$ exchange rate and the deflator from 2006 to 2017. We then divide the result by an interest rate of 5 per cent.

The monetary unit of $\tau$ must be the same as the tax or market price: Reals per ha. To correctly set the unit of measurement in equation (5), we use a two-step rescaling procedure (Souza-Rodrigues, 2019). In the first step, the coefficient of the transportation cost variable, $TC$, in equation (5) is set to minus one by estimating a constrained regression model. The unit of measurement in this model is Reals per ton. In the second step of the rescaling procedure, we normalize the parameter $\tau$. The normalized $\overline {\tau }$ is the ratio ${\tau }/{\overline {yield}}$, where the unit of $\tau$ is Reals per ha, the unit of $\overline {yield}$ is tons per ha, and the unit of the normalized incentive $\overline {\tau }$ is Reals per ton. We use the average yield of cereals in each municipality as the rescaling factor, $\overline {yield}$.²²

5. Simulation results

We have described each of the 44 biome-state markets in Brazil based on their forest debt and OC. We have then determined the optimal tax and equilibrium market price for each market and conducted a sensitivity analysis to compare the cost-sensitivity of the two market instruments. Finally, we have assessed how changes in OC over the two agricultural census periods would affect the effectiveness of each policy instrument. Throughout our analysis, we have modeled the tax and market price as a one-time payment measured in 2017 US$ per ha.

5.1. Market characteristics

The Brazilian state-biome markets have distinct characteristics. In markets where the forest debt exceeds 100,000 ha, table 2 shows the market characteristics. The FC aims to restore 15.5 million ha of native vegetation, which is also known as the forest debt (Soares-Filho et al., 2014, 2016). The forest debt is unevenly distributed throughout Brazil, with Mato Grosso, São Paulo, Paraná, Maranhão, and Pará having the highest environmental liability, accounting for almost 70 per cent of the total. Additionally, 44 per cent of the forest debt is in the Amazon biome frontier. The discrepancies in the forest debt are due to agricultural expansion in Brazil and the variance in conservation requirements in the FC. Amazon farmers are subject to an 80 per cent conservation requirement, while farmers in the Atlantic Forest biome are required to conserve only 20 per cent. Figure 3 displays the forest debt, the standard deviation (SD) of the OC, and the marginal cost across the 44 markets in Brazil.

Table 2. Simulation of a tax on agricultural land and markets of forest certificates in Brazil - census year 2006

Figure 3. Market characteristics. (a) Panel A. Market characteristics, Forest debt (million ha), OC Index std. dev., Marginal cost ($/1,000 ha), (b) Panel B. Sensitivity of reforestation to higher OC, Optimal tax ($/ha), Reduction in reforestation with OC $+$ 0.5SD, Reduction in reforestation with OC $+$ 2SDs, (c) Panel C. Sensitivity of market price of forestland certificates to higher OC, Market price ($/ha), Market price ($/ha) with OC $+$ 0.5SD, Market price ($/ha) with OC + 2SDs.

Notes: The geographical unit in each map is a state/biome market. There are 44 state/biome markets in Brazil. Panel A maps the forest debt (the amount of illegal deforestation that must be restored), the SD in the OC, and the marginal cost for each market (the slope of the supply function at the optimal reforestation level). Panel B maps the optimal tax rate in 2017 US$ per ha for the baseline OC and the reduction in the amount of land reforested when the OC is 0.5 and 2 SDs higher than the baseline. Panel C maps the equilibrium market price of forestland certificates for the baseline OC and the equilibrium market prices when the OC is 0.5 and 2 SDs higher than the baseline. The “No Market” areas do not have a sufficient supply of forest certificates to achieve the target.

The variability of the OC differs among markets. Within-market OC variation is measured by the SD of the OC in each market. The greater the variation in OC within a market, the greater the opportunity for trading between farmers with high and low OC. Most Cerrado markets and three Atlantic Forest states (Minas Gerais, Rio de Janeiro, and São Paulo) exhibit significant within-market OC variation. However, this variation is not correlated with forest debt. Only the Mato Grosso market in the Cerrado has both significant OC variation and high forest debt. The Amazon markets have high forest debt and low OC variation, while the market with the highest forest debt in the Atlantic Forest (Parana) has relatively low OC variation. For instance, the OC SD in the São Paulo–Cerrado and the São Paulo Atlantic Forest markets is similar. However, the forest debt in the São Paulo Atlantic Forest market is almost twice that in the São Paulo Cerrado market. This mismatch between conservation goals and OC variation increases conservation expenses in states with high conservation targets.²³

The marginal cost measures the financial incentive required to add 1,000 ha of forestland to the conservation target. It is calculated as the slope of the forestland supply function at the optimal level of reforestation. The marginal cost calculation is based on the reescaled forestland supply function in equation (5) where we set the unit of prices to Reals per ha. The marginal cost is high in markets with a small supply of forestland and low in markets with an ample supply of forestland.

For example, in the Rio de Janeiro–Atlantic forest market, a tax increase of $390 per ha is required to add 1 million ha of forest conservation, while in the state of Rondonia, a tax increase of $10 per ha would add 1 million ha to the conservation target. The marginal cost is also small in some Cerrado markets at the agricultural frontier (Mato Gross do Sul, Tocantins, Goias, and Minas Gerais). In these states, an increase in the optimal tax ranging from $20–50 per ha would add another 1 million ha of native vegetation to the conservation target.²⁴

5.2. Optimal tax and equilibrium market price

We begin by determining the baseline optimal tax and market price needed to achieve the conservation goals of the FC. This assumes that policymakers and farmers are aware of the OC and that it remains constant over time. Theoretically, if all parties know the OC, the optimal tax and the equilibrium market price should be the same. However, in practice, the tax and the market price may differ since each instrument impacts a different amount of agricultural land.

If we consider a tax policy, all agricultural land will be subject to taxation as it is considered available for reforestation. On the other hand, only the agricultural land above the legal reserve requirement for a market for forest certificates can be supplied for reforestation. As a result, the optimal tax will typically be lower than the market price.

Figure 4 shows the simulation of the forest certificate market and the optimal tax for four state-biome markets: Mato Grosso–Amazon; Mato Grosso–Cerrado; São Paulo–Atlantic Forest; and São Paulo–Cerrado. The gray solid lines in figure 4 represent the simulated supply function of forested land for the census year 2006 (as shown in equation (5)). The vertical dashed line indicates the forest debt and the optimal tax and market price are determined at the intersection of the forest debt and the supply of forestland.

Figure 4. Market simulations. (a) Panel A. Mato Grosso–Amazon, Market of forest certificates, Tax on agricultural land. (b) Panel B. Mato Grosso–Cerrado, Market of forest certificates, Tax on agricultural land. (c) Panel C. São Paulo–Atlantic Forest, Market of forest certificates, Tax on agricultural land. (d) Panel D. São Paulo–Cerrado, Market of forest certificates, Tax on agricultural land.

Notes: This figure displays the simulated supply function of forested land for four markets in Brazil. The solid gray lines indicate the simulated supply function of forested land for 2006. The dotted gray lines represent the supply function with a 0.5 SD shock, while the dashed lines indicate the supply function with a 2 SD shock for 2006. The long dashed line represents the supply function of forested land for 2017.

In the Mato Grosso–Amazon market, there is a large gap between the forest debt and the supply of forestland, indicating that the market is not feasible due to insufficient supply.²⁵ An optimal tax of approximately $306 per ha would induce the reforestation of the 4 million ha of forest debt in the Amazon–Mato Grosso market. In the Mato Grosso–Cerrado market, the equilibrium market price is $150/ha, and the optimal tax is lower at $90/ha, as all agricultural land in Mato Grosso–Cerrado is available for reforestation under the tax policy.

Panels C and D of figure 4 show the São Paulo–Atlantic Forest and the São Paulo–Cerrado markets respectively. The optimal market price is slightly higher in the São Paulo–Cerrado market at $174/ha compared to the São Paulo–Atlantic Forest market at $138/ha. The market agricultural tax is $102/ha in the São Paulo–Cerrado market and $114/ha in the São Paulo–Atlantic Forest market. Both market instruments would achieve the target of reforestation in São Paulo.

Table 2 provides information on the optimal tax and the equilibrium market prices for all markets in the Amazon, Cerrado, and Atlantic Forest biomes with more than 100,000 ha forest debt. The optimal tax rate ranges from $102–414 per ha in the Amazon frontier states of Tocantins, Maranhão, e Mato Grosso, and in the state of São Paulo. The optimal tax rate in several Cerrado and Atlantic Forest states is much lower, ranging from $6–78 per ha. Three Cerrado markets have the lowest taxes, and a tax of $6 per ha would achieve the reforestation target under the FC in the Tocantins–Cerrado, Minas Gerais–Cerrado, and Goias–Cerrado markets. The marginal OC in these markets is minimal, indicating a flat supply function at the optimal price.

The Amazon region has three markets with the highest taxes: Tocantins–Amazon, Maranhão–Amazon, and Mato Grosso–Amazon. These markets have a large forest debt and a slight variability in the OC. According to a similar study by Souza-Rodrigues (Souza-Rodrigues, 2019), a tax of $42.5/ha would help achieve the reforestation target in the Amazon biome. However, the conservation cost varies significantly when estimated at the state-biome level. Our findings suggest an optimal tax of $18/ha in the Rondonia–Amazon and Para–Amazon markets but a much higher tax of $306/ha in the Mato Grosso–Amazon market. Combining state-biome markets reduces the optimal tax and concentrates the reforestation effort in locations with relatively low OC. Therefore, our estimates suggest that an Amazon biome tax would induce more reforestation in Rondonia and Para than in Mato Grosso.

We find significant variation in the equilibrium market price for forestland certificates in Brazil. Our estimates for market prices are reported in table 2. The lowest market price, which is $18/ha, is observed in the Cerrado markets in the states of Tocantins, Minas Gerais, and Goias. In contrast, the highest price, which is $174/ha, is observed in the São Paulo–Cerrado market. All Cerrado and Atlantic Forest biomes markets are feasible, with prices ranging from $18–174 per ha. However, we have found that several markets in the Amazon biome are not feasible under this specific state-biome market design, wherein the supply of forestland is cerrado, and Atlantic Forest biome markets are restricted to agricultural land above the legal reserve requirement. In particular, the Para–Amazon, Tocantins–Amazon, Maranhão–Amazon, and Mato Grosso–Amazon markets do not have sufficient land for reforestation to achieve the 80 per cent conservation target in the Amazon biome.

We have estimated the equilibrium market prices using farm-level data. Our results show that these prices are generally lower than the previously estimated prices using aggregated data (Soares-Filho et al., 2016) because the process of aggregating data averages out the variability in OC.²⁶ Our estimates are more similar to those from Chomitz (2004). They used agricultural census data for 1995/1996 at the municipality-farm size class level to simulate market prices for the two markets in Minas Gerais.²⁷ Chomitz (2004) estimates a market price of $31/ha and$35/ha for the Minas Gerais–Cerrado and Minas Gerais–Atlantic Forest markets, respectively, which is similar to our estimates of $18/ha and $54/ha.

Table 2 provides information on the forest debt to agricultural land ratio, which, when taken together with the OC index statistics and marginal cost, can help to explain the differences in observed market outcomes. For instance, the Amazon–Tocantins market has a significantly higher optimal tax compared to other markets in the Amazon biome. This is due to the market's high ratio of forest debt to agricultural land, a small SD in the OC index, and a relatively low number of farms. Because of the smaller supply of agricultural land for reforestation in Tocantins, the optimal tax is located at the nonlinear section of the supply function, resulting in a substantial marginal cost and optimal tax. By contrast, the forest debt to agricultural land ratio for the Amazon–Mato Grosso market is 18 per cent lower, and the OC index SD is twice as large.²⁸

5.3. Sensitivity analysis of market instruments to changes in OC

We estimated the optimal tax and equilibrium market prices based on a constant OC in the previous section. However, this assumption might not always hold in reality. Therefore, we investigate how changes in the OC affect the market instruments. We achieve this by adding an OC shock to our land use model. This shock could represent the introduction of a new agricultural technology that increases land productivity or a shift in the demand for beef or grain due to changes in diets or disruptions to trade.

To add an OC shock, we first rewrite the land use model equation (5) using the exponential function: $share_{agriculture, i} = ({e^{\widehat {\theta }_0 + \widehat {\theta }_1 \widehat {OC}_i + \widehat {\theta }_2 X_i}}/{1+e^{\widehat {\theta }_0 + \widehat {\theta }_1 \widehat {OC}_i + \widehat {\theta }_2 X_i}})$. To simplify, we combine all the covariates into the vector $X$. We then calculate a new agricultural land share including the OC shock, $\widetilde {share}_{agriculture, i}$, by adding a shock to the OC proportional to the SD of the OC, $k\sigma ^2$:

(6)\begin{equation} \widetilde{share}_{agriculture, i} = \frac{e^{\widehat{\theta}_0 + \widehat{\theta}_1 \left(\widehat{OC}_i + k\sigma^2\right) + \widehat{\theta}_2 X_i}}{1+e^{\widehat{\theta}_0 + \widehat{\theta}_1 \left(\widehat{OC}_i + k\sigma^2\right) + \widehat{\theta}_2 X_i}}. \end{equation}

If there is a positive shock in the OC of $k\sigma ^2$, then the share of agricultural land would be affected by a change of $\Delta Share = \widetilde {share}_{agriculture, i} - share_{agriculture, i}$. To recalculate the newly reforested area after accounting for the OC shock, we use the formula $Area \ with \ OC \ shock (\tau ) = (1 + \Delta Share) \times \ Area(\tau )$. We use a multiplicative functional form to allow for a larger impact of the OC shock on the most productive land. This is because we assume that the most profitable land is more sensitive to changes in commodity prices and production technologies. Finally, we simulate the markets with OC shocks representing half the OC SD (0.5 SD) and twice the size of the OC SD (2 SD).

Figure 4 indicates how the supply function of forestland changes with the OC shocks for the Mato Grosso and São Paulo markets. The dashed lines represent the supply function of reforested land when the OC is either 0.5 SD (dotted line) or 2 SDs (short dashed line) higher or lower than the baseline OC. The long dashed curves are the supply functions of reforested land for the census year 2017. It is worth noting that the impact of an OC shock on the tax and the market for forest certificates is not the same. In the following subsections, we discuss the effect of an OC shock in each market instrument separately.

Sensitivity analysis for a tax on agricultural land

A tax on agricultural land is a market signal that induces low OC farmers into reforestation, thereby reducing total conservation costs. Although a tax on agricultural land is a cost-effective conservation policy, when the OC changes, it may not achieve its environmental objectives (Weitzman, 1974; Adar and Griffin, 1976; Mendelsohn, 1986; Stavins, 1996).²⁹ We thus estimate the reduction in reforestation under a tax policy with two OC shocks. First, we increase the OC by 0.5 SD and re-simulate all 44 markets. We then simulate a worst-case scenario with a two SD increase in the OC.

Panel B of figure 3 shows the optimal tax rate for the baseline OC and our estimations for reducing reforestation with a higher OC. Table 2 presents our results for OC shocks of 0.5 SD and 2 SDs. We found that an increase in OC affects the total land reforested in all markets in Brazil. For instance, let's consider the Amazon–Mato Grosso market. The optimal tax rate for targeted reforestation of 3.9 million ha of Amazon forest is $306/ha. Therefore, farmers with an OC lower than $306/ha would prefer to reforest rather than pay the tax. Conversely, those with an OC higher than $306/ha would choose to pay the tax and use their land for agricultural production instead. However, if the OC increases, fewer farmers would opt for reforestation, and the total area reforested would be less than the targeted 3.9 million ha. Our study shows that a 0.5 SD increase in the OC leads to a 9 per cent reduction in reforestation in the Mato Grosso market. A 2 SD increase in the OC results in a 36 per cent reduction in reforestation, meaning that farmers would only reforest 2.5 million ha of the targeted 3.9 million ha.

The extent to which deforestation is reduced varies across different markets due to the nonlinearity of the supply function and differences in the forest debt. In the Amazon biome, the markets in Maranhão, Mato Grosso, and Tocantins would see a reduction in reforestation by 9, 7, and 3 per cent, respectively, for a 0.5 SD OC shock. For a 2 SD OC shock, the reduction in reforestation would be even higher, namely, 36, 33, and 21 per cent, respectively. In the Cerrado, the reduction in reforestation would be between 5 and 10 per cent for a 0.5 SD shock, and between 14 and 34 per cent for a 2 SD shock. The markets in the Atlantic Forest biome are slightly more sensitive than those in the Cerrado. A 0.5 SD OC shock in the Atlantic Forest would reduce reforestation by 7–11 per cent, while a 2 SD OC shock would reduce reforestation by 25–36 per cent. It is worth noting that the most sensitive markets tend to have the lowest marginal cost, and in these markets, the optimal tax is located in the more elastic part of the supply function. Therefore, even small changes in the OC could lead to significant changes in the amount of reforestation.

The tax policy's sensitivity to increases in OC is higher in larger markets because the supply function of forestland tends to flatten as the geographic scope of the market increases. In addition, we conducted simulations of the OC shocks with larger market boundaries defined at the biome and country level.³⁰

Among state-biome markets, the Mato Grosso–Amazon market is the most sensitive to an increase in the OC. A 2 SD increase in the OC would reduce the optimal reforestation in most Amazon and Cerrado biomes markets by close to 30 per cent. The larger markets at the biome and country level are more sensitive to changes in the OC because the optimal reforestation level is located in the more elastic part of the forestland supply function. In a country-level market, a 0.5 SD increase in the OC reduces the total amount of land reforested by 23 per cent or 3.7 million ha. The most sensitive biome market is the Amazon, with a reduction in reforestation of 1.7 million ha for a 0.5 SD increase in the OC.

Sensitivity analysis for a market of forest certificates

In a forest certificate market, the amount of reforestation is predetermined by the policymakers. Unlike a tax on agricultural land, an OC increase does not affect the environmental target. However, it can significantly raise market prices and, as a result, increase the conservation cost for farmers. The extent of the price increase depends on the OC function's curvature and the market's forest debt size. We conducted simulations for the 44 biome-state markets of forest certificates in Brazil, with the baseline OC and increases in the OC of 0.5 and 2 SDs. Following the revised FC, we only allow trading for land that exceeds the required forest reserve on each farm. We also exclude the trading of existing forest land since it has a lower risk of deforestation and, therefore, offers no additional environmental benefit.

We have estimated the market price of forest certificates for the baseline OC and increases in the OC of 0.5 and 2 SDs. Our findings are shown in figure 3, panel C. Additionally, we have reported the market price, conservation cost, and percentage increase in the OC with a 0.5 and 2 SD OC shock in table 2. We observe that only two state-biome markets, Rondonia–Amazon and Acre–Amazon, are feasible when the OC increases by 0.5 SD in the Amazon biome. In all the other Amazon markets, the supply of reforested land is insufficient to meet the targeted forestland reserve requirement of 80 per cent of private land. Graphically, the supply function of forestland does not cross the forest debt threshold.

In the agricultural producer states of Sao Paulo, Mato Grosso, and Paraná, the market price increases from $102–174 per ha under the baseline OC. If the OC increases by 0.5 SD, the market prices in these three states increase by 80–111 per cent. The highest market price with a 0.5 SD shock is in the São Paulo Cerrado, which is $186/ha. Motivating farmers from soybean production to reforestation in a productive agricultural market is more expensive.

In most other markets in the Cerrado and some markets of the Atlantic Forest, the price changes only slightly with a slight increase in the OC. This is due to the low OC in these markets, resulting from a large amount of marginal land and low conservation targets. For instance, the Goias–Cerrado and the Minas Gerais–Cerrado markets experience a marginal increase in the market prices with a 0.5 SD shock, with the baseline market price being $18/ha in both markets. These markets are resilient to small increases in the OC.

In a worst-case scenario with a 2 SD increase in the OC, only one market within the Amazon biome would be feasible. However, all markets within the Atlantic Forest and Cerrado biomes would remain feasible with a 2SD OC shock. It is important to note that despite being more robust, even in these markets, the cost of conservation to farmers increases significantly with a 2SD OC shock. In most markets, the cost of conservation more than doubles with such a shock. In states where the agricultural sector is highly developed, such as São Paulo and Paraná, the market price of conservation almost triples with a 2SD increase in the OC. In contrast, the price of forestland credits in most states within the Cerrado and some states within the Atlantic Forest remains relatively low, ranging from $18–54 per ha, even with a 2 SD increase in the OC. These states still have a plentiful supply of land for reforestation at low cost, and their markets are robust to increases in the OC.

In the online appendix, figure A6 shows how market prices are affected when the OC changes in state-biome markets, biome, and country-level markets.³¹ In most Cerrado and Atlantic Forest markets, the optimal price would double with a 2 SD shock in OC. However, some markets in the Atlantic Forest are more robust, at least to small changes in OC. For instance, the Minas Gerais–Atlantic Forest market is the most resilient state-biome market. Its high OC SD indicates various trading opportunities. More diverse markets tend to be more robust to changes in OC. Biome-level markets are more robust than state-biome markets. The Cerrado market is the most resilient biome market, comparable to the robustness of a country-level market.³²

5.4. Changes in OC and markets from 2006 to 2017

In this section, we assess how changes in market characteristics between 2006 and 2017 affected the outcomes of Brazil's market instruments for conservation. We estimate our empirical models for the OC and the supply function of forestland using the 2017 agricultural census survey completed by IBGE (2017). We find that our estimates for the discrete choice model of land use are consistent across census years (online appendix, tables A3 and A4).

We observed two changes in the distribution of OC from the 2006 to the 2017 census year. Firstly, the OC increased significantly for larger commercial farms (figure A4 in the online appendix). However, this change in OC was not uniform across all farms. Although the OC in 2017 tended to be lower than in 2006 for farms with low OC, it was significantly higher for farms with high OC. This result is consistent with our simulated shocks in OC, which assumed that the larger commercial farms are more sensitive to changes in OC. The significant soybean expansion in Brazil between 2006 and 2017 was the primary driver for this skewed distributional change.³³

Table A2 in online appendix A reports the land use choices of farmers in 2006 and 2017 by farm size and annual revenue. The national economic industry classification of the farm defines the land use choice. We found that the number of farms that have soy production as their leading economic activity increased by 32 per cent for the sample with all farms, 64 per cent for the subsample with large farms, and 84 per cent for the subsample of farms with more than 500 ha and ten MWs in annual revenue.

Secondly, the SD of OC increased across classes of farms. The box plot graph in figure A4 (online appendix) shows how the distribution of OC spread out in 2017. The higher OC variance in 2017 was driven by a significant increase in OC for the farms with the highest OC. The distribution of OC became more positively skewed in all subsamples of farms, even in the subsample of small commercial farms (0–50 ha).

Figure A3 in the online appendix depicts the changes in market characteristics from 2006 to 2017. Panel A shows the average forestland share change for each Brazilian microregion.³⁴ Panel B shows the average OC change in each microregion, while panel C shows the SD of the OC change in each microregion. The unit of measure in each map is an SD of the variable in 2016 in the respective microregion. The changes in these factors varied spatially across the country. The agricultural frontier, the transition region between the Cerrado and the Amazon biomes, saw the most significant decrease in forestland share, ranging from 0.2 to 1 SD.³⁵

The average OC levels increased significantly in areas known as the agricultural frontier and in states with large agricultural sectors like Paraná and Rio Grande do Sul, with Mato Grosso experiencing the most substantial rise. Specifically, certain microregions in Mato Grosso saw OC levels surge by over two SDs within a decade. This finding indicates that the previously considered worst-case scenario of a 2 SD increase in our sensitivity analysis might be more probable than initially thought. Nonetheless, numerous microregions across Brazil observed minimal changes in OC levels.

From 2006 to 2017, the most notable change in market characteristics was the increase in the SD of OC, especially in developed agricultural microregions of Mato Grosso, Mato Grosso do Sul, and Paraná. This variability underscores Brazil's uneven agricultural development and the high value placed on land suitable for soy cultivation compared to pasture land.

These market shifts could impact the efficacy of instruments aimed at forest conservation. The decline in forestland share presumably raised the forest debt, leading to a likely increase in both optimal tax rates and market prices for forest conservation efforts. Particularly in regions nearing the agricultural frontier, a significant reduction in forestland coupled with a spike in OC implies higher conservation costs and potentially insufficient conservation outcomes without adjustments in tax rates or forest debt. However, the increased variability in OC within microregions could create more opportunities for arbitrage among farmers, potentially benefiting market-based conservation instruments. Yet, the actual impact hinges on how the OC distribution changes, with significant increases among the most profitable farmers not necessarily creating new arbitrage opportunities.

Our analysis reveals that the OC changes from 2006 to 2017 did not significantly influence market instruments, primarily affecting farms with already high OC levels. This outcome underscores the importance of both the magnitude of OC change and its distribution for market instruments’ effectiveness. Notably, farmers with the highest OC in 2006, likely participants in a forest certificate market, saw their OC rise by 2017 but were already market participants. Conversely, farmers with lower OC, typically certificate sellers, saw little to no change in their OC levels, keeping the market dynamics between 2006 and 2017 relatively stable. Despite some variability, market prices and characteristics for forest certificates remained consistent across years, with few exceptions in the Amazon biome.³⁶

5.5. An agricultural frontier market

Market simulations are useful for testing and guiding market designs before implementation. In this section, we evaluate a different market design by simulating a market for the agricultural frontier in Brazil. This design is a potential solution for the supply constraints we identified for most of the state-biome markets in the Amazon. However, expanding the geographical scope of a market has its trade-offs. Increasing the market size leads to increased supply, reduced market prices, and conservation costs. On the other hand, a larger market can lead to an unequal distribution of natural vegetation that may not be environmentally beneficial. To overcome this trade-off, we propose an agricultural frontier market within the Amazon biome, which may offer a reasonable compromise.

We use farm-level census data to simulate the effect of the frontier market design and at a lower level of aggregation, the microregion level. We simulate how expanding the geographical scope of the market will affect the distribution of reforestation across the microregions in the frontier market.³⁷

We have simulated two agricultural markets known as Frontier 1 and Frontier 2. Figure 5, panel B illustrates these markets. The Frontier 1 market consists of 22 microregions located at the Amazon agricultural frontier and is closely associated with the ”Arc of Deforestation” area in the Amazon. The Frontier 2 market is an expansion of Frontier 1. It encompasses all 22 microregions of Frontier 1 plus 28 additional microregions located further into the Amazon biome.³⁸

Figure 5. Agricultural frontier market simulation. (a) Panel A. Simulations of Frontier market 1, Agricultural tax, Market of forest certificates. (b) Panel B. Allocation of reforestation with tax policy (relative to forest debt), Frontier 1 market, Frontier 2 market, (c) Panel C. Allocation of reforestation with market policy (absolute reforestation), Frontier 1 market, Frontier 2 market.

Notes: This figure simulates the agricultural frontier market and compares the impact of tax and market policies on reforestation allocation. Panel A depicts the supply function of forested land for both market and tax policies in Frontier 1. Panel B displays the allocation of reforestation, measured as a fraction of the forest debt by microregion, for a tax policy applied to Frontier 1 and Frontier 2 markets. Finally, panel C shows the allocation of reforestation, measured in millions of ha, for a market policy applied to the Frontier 1 and Frontier 2 markets.

In figure 5, panel A, we have presented the simulated Frontier 1 market under two policies, namely the agricultural tax policy and the market of forest certificates policy. Under the market of forest certificates policy, the supply function is restricted to agrarian land above FC reserve requirements. The simulated Frontier 2 market has a similar shape. The total forest debt in the Frontier 1 and Frontier 2 markets is 5.14 and 6.60 million ha of Amazon forest, respectively.

We assessed the potential of an agricultural tax for the Frontier 1 and Frontier 2 markets. Our findings showed that an optimal tax of $119/ha would encourage the reforestation of land equivalent to all forest debt in the Frontier 1 market. The optimal tax for the Frontier 2 market is $94/ha, which is expected as the tax decreases with the market size. The reduction in the optimal tax from Frontier 1 to Frontier 2 is small because the amount of agricultural land available for reforestation decreases as we move farther into the Amazon biome. The optimal taxes in the Frontier markets are lower than those for the Mato Grosso–Amazon e Tocantins–Amazon markets, which stand at $306 and 414 per ha, respectively.

We also found that the Frontier 1 and Frontier 2 markets are similarly sensitive to increases in the OC.³⁹

Expanding the geographical scope of the market can lead to a spatial concentration in reforestation. Figure 5, panel B shows the allocation of reforestation under a tax policy. The spatial unit is a microregion. The unit of measure in the maps in figure 5, panel B is the increase in reforestation relative to the forest debt in each microregion. It is a relative measure of the reforestation effort. For example, a -0.5 value means only 50 per cent of the forest debt in the microregion was reforested. The reforestation effort is unequally distributed across the market. In the Frontier 1 market, most of the reforestation effort happens in the states of Rondonia and Para. The reforestation effort in the more productive state of Mato Grosso will be lower than 50 per cent of the forest debt. We find a similar pattern of spatial allocation of reforestation efforts in the Frontier 2 market. However, reforestation is more extensively reallocated out of the Mato Grosso market and into the Rondonia, Para, and Acre microregions. In the Frontier 2 market, the reforestation in Mato Grosso will be lower than 25 per cent of the state forest debt.

Expanding the Frontier market leads to a slight reduction in the optimal tax (21 per cent) but a significant reallocation of reforestation efforts because of the considerable spatial variation in the OC within the Amazon.

Figure 5, panel A displays the forest certificate market simulation for Frontier 1. Although the geographical scope has expanded, the supply and demand for Frontier 1 and Frontier 2 markets are still imbalanced. The limited availability of land for reforestation, beyond the legal reserve requirements, makes it challenging to achieve the conservation target of 80 per cent in the Amazon. The forest debt line, which represents the forest restoration required to meet the conservation target, is located to the right of the horizontal scale in the graph.

We analyzed the potential of these forest markets to partially achieve the conservation target. The results show that the restoration of 2 million ha in Frontier 1 (40 per cent of the forest debt) could be achieved with an equilibrium market price of $302/ha. Similarly, 3 million ha (45 per cent of the forest debt) could be restored in the larger Frontier 2 market with an equilibrium market price of $306/ha. However, it is unlikely that a market policy in the Amazon biome can restore more than 50 per cent of the forest debt in the biome.

We also found that both Frontier 1 and Frontier 2 markets are not very sensitive to small changes in OC. A 0.5 SD increase in the OC would increase the equilibrium market price by 8 per cent in both markets. However, a 2 SD shock in the OC would lead to a price increase of 89 per cent in the Frontier 1 market. The Frontier 2 market with a conservation target of 3 million ha becomes unfeasible with a 2 SD shock in the OC.⁴⁰

Panel C of figure 5 shows the spatial distribution of reforestation when a market policy is applied. Restoring the entire forest debt is not possible, so we show the absolute value of reforestation in the figure. The unit of measure used is thousands of ha of reforested land. The spatial distribution of reforestation is similar under both tax and market policies. However, reforestation has shifted from the productive Mato Gross microregions to microregions with lower OC. Nevertheless, some reforestation will still happen in the Mato Grosso state because the OC varies significantly within the state. Panel C indicates that several microregions in the Mato Grosso state will reforest an area over 50,000 ha.⁴¹.

6. Conclusion

Our study evaluates the impact of implementing an agricultural land tax and tradable forest certificates to conserve Brazil's native vegetation across different farms. We have found significant differences in conservation costs and benefits among various regions, implying that a single policy may not be effective for all biomes in Brazil. Based on our simulations, we have identified four insights that can help refine market-based conservation strategies.

Our simulation results show the trade-offs between tax and market policies for the Amazon biome. A tax policy would achieve the conservation target only in the unlikely scenario of a constant OC in the Amazon biome. A market for forest certificates could partially achieve about 45 per cent of the conservation targets in the FC. Both policies will lead to an unequal spatial distribution of the reforestation effort. These results suggest that a combination of conservation policies may be necessary to achieve the conservation targets in the Amazon. For example, regulators could implement a market of forest certificates in the agriculture frontier and expand permanent conservation areas in the microregions with the highest OC to ensure an acceptable level of forest conservation and local ecosystem services.

In the Cerrado biome, conservation targets can be achieved through a tax or a market policy in all markets except Mato Grosso and São Paulo. Optimal prices for both policies are relatively low and are not very sensitive to small changes in OC. The choice of policy may depend on the flexibility of the conservation target. If the marginal damage caused by further deforestation in a market is too high, a market policy will ensure compliance with the reforestation target. If the conservation target allows for flexibility, a tax would reduce the impact of changes in the OC on farmers. The simulation model could also be expanded to assess the combined markets within the Cerrado biome.

In markets such as the Atlantic Forest, Mato Grosso and São Paulo Cerrado, the optimal tax and market prices are relatively higher and more sensitive to changes in OC. As a result, these markets tend to be more volatile. Locations with high OC tend to experience an increase in OC, such as the Atlantic Forest Markets. In such cases, policymakers may consider a hybrid policy that extends conservation zones and implements a market-based policy to reduce price volatility in forest markets.

Based on our sensitivity analysis, it appears that conservation instruments that rely on market mechanisms might need adjustments to account for changes in both economic and ecological characteristics over time. This could mean that tax rates need to be modified or reforestation targets reevaluated periodically.⁴² Our simulation model can be expanded to assess the potential impact of different distributional shocks or significant investments in transportation infrastructure before policy implementation. These areas could be promising avenues for future research.