Deforestation is a matter of pressing global concern, yet surprisingly little is known about the relative efficacy of various policies designed to combat it. This article sets out a framework for measuring the cost effectiveness of alternative policies—both command-and-control and incentive-based—in the Brazilian Amazon. First, I estimate the demand for deforestation on private properties, exploiting regional variation in transportation costs as a means to recover farmers’ responses to permanent policies. Here, rescaling transportation costs using local yields allows me to express changes in farmers’ valuations in dollars per hectare. I then use the estimated demand to infer farmers’ willingness to deforest under different counterfactual policies, such as payments to avoid deforestation and taxes on land use, along with the corresponding potential farmers’ lost surpluses. The results indicate that payment programmes and land use taxes on agricultural land can be highly effective in preserving the rainforest and also be substantially less expensive than command-and-control policies (approximately 8 times less costly). A carbon tax equal to the social cost of carbon could virtually eliminate all agricultural land in the Amazon, given the low agricultural returns there.

Discrete choice data allow researchers to recover differences in utilities, but these differences may not suffice to identify policy-relevant counterfactuals of interest. In fact, in the case of dynamic discrete choice models, only a narrow set of counterfactuals are point-identified. In this paper, we explore how much one can learn about counterfactual outcomes of interest within this framework. We focus on the partial identification of counterfactuals, while allowing for (mild) model restrictions that can gradually shrink the identified set. We derive bounds for low-dimensional objects (such as average welfare) as arguments of optimization programmes, along with a uniformly valid inference procedure. Furthermore, we develop new and tractable computational tools and algorithms suitable for dealing with high-dimensional problems like this. Finally, we illustrate in Monte Carlos, as well as an empirical exercise of firms’ export decisions, the informativeness of the identified sets, and we assess the impact of (common) model restrictions on results.

This article sets out a data-driven approach for targeting environmental policies optimally in order to combat deforestation. We focus on the Amazon, the world’s most extensive rainforest, where Brazil’s federal government issued a “Priority List” of municipalities in 2008—a blacklist to be targeted with more intense environmental monitoring and enforcement. First, we estimate the causal impact of the Priority List on deforestation (along with other relevant treatment effects) using “changes-in-changes” due to Athey and Imbens (2006), finding that it reduced deforestation by 43% and cut emissions by almost 50 million tons of carbon. Second, we develop a novel framework for computing targeted optimal blacklists that draws on our treatment effect estimates, assigning municipalities to a counterfactual list that minimizes total deforestation subject to realistic resource constraints. We show that the ex post optimal list would result in carbon emissions over 10% lower than the actual list, amounting to savings of more than $ $ $1.2 billion (34% of the total value of the Priority List), with emissions over 23% lower on average than a randomly selected list. The approach we propose is relevant both for assessing targeted counterfactual policies to reduce deforestation and for quantifying the impacts of policy targeting more generally.

Satellite-based image classification facilitates low-cost measurement of the Earth’s surface composition. However, misclassified imagery can lead to misleading conclusions about transition processes. We propose a correction for transition rate estimates based on the econometric measurement error literature to extract the signal (truth) from its noisy measurement (satellite-based classifications). No ground-truth data are required in the implementation. Our proposed correction produces consistent estimates of transition rates, confirmed by longitudinal validation data, while transition rates without correction are severely biased. Using our approach, we show how eliminating deforestation in Brazil’s Atlantic forest region through 2040 could save $100 billion in CO2 emissions.