Do you know your biases? A Monte Carlo analysis of dynamic panel data estimators

We assess the performance of widely-used dynamic panel data estimators based on Monte Carlo simulations of a dynamic economic process. Knowing the true underlying coefficient of the autoregressive term, we show that most estimators exhibit a severe bias even in the absence of measurement errors, omitted variables, and endogeneity issues. We analyze how the bias changes with the sample size, the autoregressive coefficient, and the estimation options. Based on our insights, we recommend i) carefully choosing appropriate estimators given the underlying structure of the data and ii) scrutinizing the estimation results based on the insights of simulation studies.