Bias-corrected cluster-robust standard errors for fixed effects PPML estimators of gravity panel models with autocorrelated disturbances

Abstract Panel gravity models with country-pair and time variation typically feature autocorrelated disturbances calling for country-pair clustered standard errors of the estimated structural parameters. Yet, Monte Carlo simulations reveal a pronounced downward bias of the country-pair clustered standard errors. With an estimated autocorrelation parameter of the disturbances at hand, it is straightforward to form a working variance that accounts for autocorrelation within country-pairs and to apply the Pustejovsky and Tipton (JBES 36:672–683, 2018)-bias correction. Monte Carlo simulations illustrate that this bias correction nearly eliminates the bias and yields correct coverage rates of the confidence intervals. Using the cluster-specific pairs bootstrap to form percentile confidence intervals also performs well and exhibits correct coverage rates.