糖心TV

Title: Identification Using Internal Instruments in Large N and T Panel Models (joint with Andrei Zeleneev).

This is preliminary work so we do not yet have a complete paper, here is an abstract .

Abstract - In order to identify causal effects using panel data, researchers may exploit the presence of external instruments in the form of aggregate shocks which vary over time but not between individuals, e.g., policy shocks or cost shocks. By construction, these shocks are uncorrelated with any idiosyncratic variation in confounding factors, and they are uncorrelated with aggregate confounding factors if they satisfy an exclusion restriction. In this work we show that with large N and T panel data, causal effects can be identified and estimated when such exogenous shocks exist, even if they are not directly observed. Our identification approach can be summarized as follows. First, we propose that one form a vector of candidate (i.e., possibly endogenous) instruments by extracting period-specific factors from the matrix of outcomes and treatments. Second, we show that with large and panel data, given a vector of candidate instruments, one can identify causal effects so long as there exists some (unknown) linear combination of these candidate instruments that is exogenous. We propose an estimator whose form is motivated by our identification results and we provide some simulation evidence of its efficacy.