In this paper we construct a parsimonious causal model that addresses multiple issues researchers face when trying to use aggregate time-series shocks for policy evaluation: (a) potential unobserved aggregate confounders, (b) availability of various unit-level characteristics, (c) time and unit-level heterogeneity in treatment effects. We develop a new estimation algorithm that uses insights from treatment effects, panel, and time-series literature. We construct a variance estimator that is robust to arbitrary clustering pattern across geographical units. We achieve this by considering a finite population framework, where potential outcomes are treated as fixed, and all randomness comes from the exogenous shocks. Finally, we illustrate our approach using data from a study on the causal relationship between foreign aid and conflict conducted in Nunn and Qian [2014].
↧