This paper considers the estimation problem in dynamic games with finite actions. we derive the equation system that characterizes the markovian equilibria. the equilibrium equation system enables us to characterize conditions for identification. we consider a class of asymptotic least squares estimators defined by the equilibrium conditions. this class provides a unified framework for a number of well-known estimators including those by Hotz and Miller (1993) and by Aguirregabiria and Mira (2002). We show that these estimators differ in the weight they assign to individual equilibrium conditions. We derive the efficient weight matrix. A Monte Carlo study illustrates the small sample performance and computational feasibility of alternative estimators.

Recursive procedures which are based on iterating on the best response mapping have difficulties converging to all equilibria in multi-player games. We illustrate these difficulties by revisiting the asymptotic properties of the iterative nested pseudo maximum likelihood method for estimating dynamic games introduced by Aguirregabiria and Mira (2007). An example shows that the iterative method may not be consistent.