Title: Boosted panel data approach for program evaluation
Authors: Zhentao Shi - CUHK (Hong Kong) [presenting]
Abstract: Policy evaluation is a central question in empirical economic studies, but economists mostly work with observational data in view of the limited opportunities to carry out controlled experiments. The lack of genuine control groups motivated to exploit the correlation between cross-sectional units in a panel data to construct the counterfactual. The choice of cross-sectional units, a key step in implementing such a method, has not been not addressed in the case of many potential controls. We propose to use the component-wise boosting for control-unit selection. We show that such a choice is asymptotically valid even if the number of potential controls grows, in the limit, faster than the time dimension. Both in theory and in practice, we open the possibility the above-mentioned method to be applied to empirical research in big data environment.