Speaker: Professor HSIAO Cheng (University of Southern California)
Presentation: Panel Parametric, Semi-parametric and Nonparametric Construction of Counterfactuals
Date: 16 April, 2019 (TUE)
Venue: C/F, Library Complex, Hong Kong Shue Yan University
We consider panel parametric, semiparametric and nonparametric methods of constructing counterfactuals. We show through extensive simulations that no method is able to dominate other methods in all circumstances, since the true data‐generating process is typically unknown. We therefore also suggest a model‐averaging method as a robust method to generate counterfactuals.
The advantage of the parametric model approach is that it allows efficient estimation of the unknown parameters; hence, it is possible to identify the impact of each covariate on the outcome. The disadvantage is that, if the model is misspecified, then the resulting inference could be misleading. The advantage of the panel nonparametric approach is that there is no need to consider the conditional mean or the error distribution. The disadvantage is that there is no structural interpretation of the casual effects, only the measurement of the treatment effects. The semiparametric approach is somewhere in between. It assumes that the conditional mean of the observed covariates is specified correctly, but it lets the data control the impact of unobserved factors. Each method has its advantages and disadvantages. Our simulation results show that, if the observed data are stationary, the panel semi-parametric method appears capable of generating counterfactuals close to the (true) data generating process in a wide array of situations. If the data are nonstationary, then the panel nonparametric method appears to dominate the parametric and semiparametric approaches. However, no method appears capable of dominating all other methods under all different data generating processes and different sample configurations of cross-sectional dimension N and pre-treatment time dimension T0: Since the true data generating process is usually unknown and the statistical findings could be very different for different situations, we have also suggested a model averaging method as a robust method for generating counterfactuals.
Wiley Online Library: https://onlinelibrary.wiley.com/doi/abs/10.1002/jae.2702