Avoiding biased versions of Wooldridge's simple solution to the initial conditions problem

Wooldridge (2005) provided a simple and elegant solution to the initial conditions problem for dynamic nonlinear unobserved-effects models. His original auxiliary model includes the time-varying explanatory variables at each period. Unfortunately, a popular constrained version that includes within-means of the explanatory variables can be severely biased. We show that there are several ways to avoid this problem.