We consider dynamic games that can have state variables that are partially observed, serially correlated, endogenous, and heterogeneous. We propose a Bayesian method that uses a particle filter to compute an unbiased estimate of the likelihood within a Metropolis chain. Unbiasedness guarantees...