Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When Does Agroforestry Income Reduce Deforestation? Evidence from a Natural Experiment in Madagascar

Camille DeSisto, Ranaivo Rasolofoson, Michelle Foley, Harsh Parikh

Abstract

Tropical deforestation and rural poverty are deeply intertwined, yet isolating the causal effect of income on forest loss remains challenging. We use the 2015 global vanilla price boom, triggered by food-industry shifts toward natural flavoring, as an exogenous income shock affecting Madagascar's primary vanilla-producing region. Using a matching-augmented synthetic control design, we estimate that income gains reduced annual deforestation by 1.7 percentage points in 2017, equivalent to approximately 701 hectares of avoided forest loss. Under a monotonicity assumption linking the price boom to farmers' income, the sign of this reduced-form effect is informative about the causal direction of income on deforestation. However, effects were strongly heterogeneous: higher incomes reduced deforestation in drier, more accessible municipalities but increased clearing in wetter, low-elevation areas with high agricultural potential. These divergent patterns suggest that income simultaneously relaxes subsistence pressures driving forest dependence and raises the opportunity cost of conservation where agricultural returns are high. Our findings indicate that commodity-based agroforestry can align poverty alleviation with forest conservation under conditions of low agricultural opportunity cost. Still, policies must anticipate contexts where rising incomes amplify deforestation in agriculturally suitable land. The strategic targeting of livelihood interventions based on local agricultural potential may help reconcile development and conservation objectives in tropical forest frontiers.

When Does Agroforestry Income Reduce Deforestation? Evidence from a Natural Experiment in Madagascar

Abstract

Tropical deforestation and rural poverty are deeply intertwined, yet isolating the causal effect of income on forest loss remains challenging. We use the 2015 global vanilla price boom, triggered by food-industry shifts toward natural flavoring, as an exogenous income shock affecting Madagascar's primary vanilla-producing region. Using a matching-augmented synthetic control design, we estimate that income gains reduced annual deforestation by 1.7 percentage points in 2017, equivalent to approximately 701 hectares of avoided forest loss. Under a monotonicity assumption linking the price boom to farmers' income, the sign of this reduced-form effect is informative about the causal direction of income on deforestation. However, effects were strongly heterogeneous: higher incomes reduced deforestation in drier, more accessible municipalities but increased clearing in wetter, low-elevation areas with high agricultural potential. These divergent patterns suggest that income simultaneously relaxes subsistence pressures driving forest dependence and raises the opportunity cost of conservation where agricultural returns are high. Our findings indicate that commodity-based agroforestry can align poverty alleviation with forest conservation under conditions of low agricultural opportunity cost. Still, policies must anticipate contexts where rising incomes amplify deforestation in agriculturally suitable land. The strategic targeting of livelihood interventions based on local agricultural potential may help reconcile development and conservation objectives in tropical forest frontiers.
Paper Structure (41 sections, 2 theorems, 16 equations, 11 figures)

This paper contains 41 sections, 2 theorems, 16 equations, 11 figures.

Table of Contents

  1. Introduction
  2. Results
  3. Discussion
  4. Causal Evidence
  5. Exploring Heterogeneity
  6. Limitations
  7. Conclusions
  8. Methods
  9. Study design
  10. Units, exposure, and period
  11. Outcome and covariates
  12. Statistical Analysis
  13. Identification strategy
  14. Constructing a comparable donor pool
  15. Estimation: augmented synthetic control
  16. ...and 26 more sections

Key Result

Theorem 1

Under (A1)--(A5), for any $t\ge t^\star$, the ATT is identified as where $Y_{i,t}^{(0)} := \Lambda_t^\top\Phi_i + \eta_{i,t}$ is the counterfactual outcome that treated unit $i$ would have experienced absent the shock, constructed from the pre-shock fit.

Figures (11)

  • Figure 1: Vanilla cultivation in the agroecological matrix of rural northeast Madagascar. A farmer hand-pollinates vanilla orchids grown on tutor trees, reflecting the labor-intensive nature of the crop. Deforestation in the rural tropics is often a dangerous, last-resort activity driven by economic necessity. If higher incomes relax this pressure, wealth should reduce deforestation. However, the same income gains can incentivize forest clearing to expand agricultural production, making the net effect ambiguous. We use the 2015 global vanilla price boom---an exogenous shock to regional income in Madagascar's SAVA region---to isolate this relationship. On average, the income shock reduced deforestation. However, effects were strongly heterogeneous: income decreased deforestation in drier, more accessible municipalities and increased it in wetter, low-elevation areas with high agricultural potential.
  • Figure 2: Statistical matching results: (a) map of municipalities across Madagascar, vanilla-farming (treated) municipalities and matched controls; (b) covariate balance before and after matching; and (c) mean deforestation in vanilla-producing municipalities (green) and matched controls (pink), alongside vanilla price (dashed black). Shaded ribbons denote 95% confidence intervals. Note that population density, percent forest cover, and forest area represent mean values for 2013 and 2014. "Prec. PC1" and "Prec. PC2" are principal components for precipitation (Fig. \ref{['fig:pca']}); "Roads" is the density of roads; "Protected" is the proportion of land area covered by a Madagascar National Park protected area.
  • Figure 3: Augmented synthetic controls: (a) treated municipalities pooled; (b) treated municipalities modeled separately. In (a), points show estimates with 80%, 90%, and 95% confidence ribbons (light to dark). In (b), green lines are municipality-level effects; the solid black line is the cross-municipality mean. Vertical dashed line marks 2015 (treatment onset). The $y$-axis in (b) is truncated to $[-5.5,\,3.5]$ pp for clarity.
  • Figure 4: Heterogeneity in estimated effects of the price shock: (a) decision tree; (b) map of municipality-level effects; (c) precipitation (PC1); (d) elevation; and (e) road density. PC1 loads negatively on precipitation variables (higher values $\Rightarrow$ drier). The decision tree in panel (a) shows partitions based on municipality characteristics, where each internal node shows a splitting rule, each branch shows a decision path, and each terminal node represents a group of municipalities with shared characteristics. Nodes also denote the number of municipalities in each partition and the mean deforestation rate across those municipalities.
  • Figure S1: PCA results of historical precipitation across all municipalities: a) biplot of the first two principal components and b) a heatmap showing the loadings of the first two principal components.
  • ...and 6 more figures

Theorems & Definitions (4)

  • Theorem 1: Identification of the reduced-form ATT
  • proof
  • Lemma 1: Sign identification for the income effect
  • proof