Bayesian estimation of panel count data models: dynamics, latent heterogeneity, serial error correlation and nonparametric structures.

In this chapter we discuss how Bayesian techniques can be used to estimate the Poisson model with exponential conditional mean, a well-known panel count data model. In particular, we focus on the implementation of Markov Chain Monte Carlo methods to various specifications of this model that allow for dynamics, latent heterogeneity and/or serial error correlation. The latent heterogeneity distribution is assigned a nonparametric structure, which is based on the Dirichlet process prior. The initial conditions problem also is addressed. For each resulting model specification, we provide the associated inferential algorithm for conducting posterior simulation. Finally, relevant computer codes are posted as an online supplement.