Heteroskedasticity‐Robust Standard Errors for Dynamic Panel Data Models with Fixed Effects

For linear panel data models with fixed effects, cluster‐robust covariance estimation does not use variability over time. The extant heteroskedasticity‐robust methods available under strict exogeneity do not generalize to dynamic models. We propose novel robust covariance estimators under a strong version of serial uncorrelatedness, where serial uncorrelatedness is required to identify dynamic panel models. Asymptotics are established, and simulations verify theoretical findings. The estimator can apply to the popular dynamic IV‐GMM estimators and be a sharper alternative for cluster‐robust covariance estimators in panel data models with limited cross‐sectional information.