Bias and consistency in three-way gravity models

We study the incidental parameter problem for “three-way” Poisson Pseudo-Maximum Likelihood (“PPML”) estimator recently recommended identifying effects of trade policies and in other panel data gravity settings. Despite number variety fixed involved, we confirm PPML is consistent T show it fact only among a wide range PML estimators that generally this context when fixed. At same time, asymptotic confidence intervals fixed-T panels are not correctly centered at true values, cluster-robust variance estimates used to construct standard errors biased as well. characterize each these biases analytically both numerically empirically they salient even real-data settings with large countries. also offer practical remedies can be obtain more reliable inferences time-varying variables, which make available via an accompanying Stata package called ppml_fe_bias.