What Can We (Machine) Learn about Welfare Dynamics from Cross-Sectional Data?

Lucchetti, Leonardo

World Bank, Washington, DC

关键词: POVERTY;    POVERTY TRANSITIONS;    LASSO;    MACHINE LEARNING;    WELFARE DYNAMICS;   

DOI  :  10.1596/1813-9450-8545
RP-ID  :  WPS8545

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

This paper implements a machine learningapproach to estimate intra-generational economic mobilityusing cross-sectional data. A Least Absolute Shrinkage andSelection Operator (Lasso) procedure is applied to explorepoverty dynamics and household-level welfare growth in theabsence of panel data sets that follow individuals overtime. The method is validated by sampling repeatedcross-sections of actual panel data from Peru. In general,the approach performs well at estimating intra-generationalpoverty transitions; most of the mobility estimates fallwithin the 95 percent confidence intervals of povertymobility from the actual panel data. The validation alsoconfirms that the Lasso regularization procedure performswell at estimating household-level welfare growth betweentwo years. Overall, the results are sufficiently encouragingto estimate economic mobility in settings where panel dataare not available or, if they are, to improve panel datawhen they suffer from serious non-random attrition problems.