2008年-世界发展银行全球_Estimating_the_Effects_of_Aggregate_Agricultural_Growth_on_the_Distribution_of_Expenditures_24页_279kb

Summary of the World Development Report 2008 Background Paper: Estimating the Effects of Aggregate Agricultural Growth on the Distribution of Expenditures

Core Content

This paper explores how changes in the sectoral composition of income—specifically, agricultural and non-agricultural income growth—affect the distribution of expenditures across households in various countries. The study uses household-level survey data from multiple countries over several decades to estimate the impact of income growth on expenditure distribution.

Main Questions

• How do changes in the sectoral composition of income influence the distribution of expenditures?
• What is the effect of agricultural income growth on different deciles of the expenditure distribution?
• How do these effects compare to those of non-agricultural income growth?

Key Findings

• Agricultural income growth disproportionately benefits poorer households, particularly the poorest decile, with a significant positive impact on their expenditures.
• Non-agricultural income growth has a more substantial overall effect on total income and expenditures, but it has a negative effect on the poorest decile.
• The elasticity of expenditure growth for agricultural income is higher for lower deciles, while for non-agricultural income, it increases with higher wealth.
• The third decile shows an elasticity of almost exactly one for non-agricultural income growth, indicating proportional response.
• The distributional effects of agricultural growth are more pronounced than those of non-agricultural growth, even though the latter contributes more to total income.

Methodology

The paper employs a two-stage instrumental variables (IV) estimator to address potential endogeneity in income growth variables. The first stage uses average growth rates of neighboring countries' agricultural and non-agricultural income as instruments for own country's income growth. The second stage estimates the effect of predicted income growth on expenditure growth across deciles.

Estimation Equation (1)

$$
\Delta \log c_{t}^{(\ell ,q)} = \alpha^{(\ell , q)} + \eta_{t} + \eta_{t}^{\ell} + \beta_{q}^{1}\Delta \log y_{\ell t}^{1} + \beta_{q}^{2}\Delta \log y_{\ell t}^{2} + \epsilon_{t}^{(\ell ,q)}.
$$

• $\alpha^{(\ell,q)}$: Country-quantile fixed effects.
• $\beta^1$, $\beta^2$: Elasticities of expenditure growth with respect to agricultural and non-agricultural income growth.
• $\eta_t$, $\eta_t^\ell$: Common global shocks and year effects.

Data and Panel Structure

• The dataset includes household-level expenditure data from over 40 countries.
• A country-decile panel is constructed, with data available for varying numbers of years.
• The panel is unbalanced, requiring at least three years of data to estimate the fixed effects.
• Deciles are used to represent the distribution of expenditures.

Instrumental Variables Strategy

• Neighboring countries' agricultural income growth is used as an instrument for own country's agricultural income growth.
• This is based on the assumption that unobserved shocks affecting income and expenditures are likely to be country-specific, while productivity shocks (e.g., weather) may be shared across borders.

Estimation Results

• First stage regression (Table 2):
  • Agricultural income growth has lower elasticity estimates (0.135 to 0.211) compared to non-agricultural (0.541 to 0.584).
  • The fit of non-agricultural income regressions is better, with higher $R^2$ values.
• Second stage regression (Table 4):
  • The poorest decile (10%) experiences a 1.65% increase in expenditure growth with a 1% increase in agricultural income.
  • The poorest decile (10%) experiences a 0.7% decrease in expenditure growth with a 1% increase in non-agricultural income.
  • Elasticities for non-agricultural income increase with wealth, while those for agricultural income decrease.
  • Standard errors are robust to heteroskedasticity and arbitrary correlation across deciles.

Alternative Estimation Approaches

• Ordinary Least Squares (OLS) is also used (Table 5), but a Hausman test rejects the null hypothesis that OLS estimates are equivalent to IV estimates, suggesting that endogeneity is a significant concern.
• The results from OLS are similar to IV estimates, but with lower significance for agricultural income growth, reinforcing the need for instrumental variables.

Implications

• Agricultural growth is more effective in reducing poverty than non-agricultural growth, as it has a stronger impact on the poorest households.
• Sectoral composition of income plays a crucial role in shaping expenditure distribution.
• The validity of instruments and error correction strategies are important for accurate estimation and inference.
• The trade-off between estimator consistency and precision is a challenge in panel data analysis, especially with small, unbalanced panels.

Conclusion

The paper highlights the asymmetric effects of agricultural and non-agricultural income growth on expenditure distribution. It underscores the importance of instrumental variables in addressing endogeneity and provides evidence that agricultural growth is more beneficial for poorer households, thus having a more equalizing effect on the overall distribution of expenditures.