Econometric modeling of heterogeneous panel data using independent and dependent mixture distributions

This dissertation uses flexible finite mixture distributions to simultaneously make inferences about the group membership of heterogenous firms and households and their behavior. It first exhibits the relationship existing among several alternative mixture models and shows how their comparison can be used to learn about specific features of a given panel data set.;These models are then used to learn about a firm's financial states and about the interdependence between these states and a firm's investment decision using a panel data set on US manufacturing firms obtained from the COMPUSTAT database. Their comparison reveals that time dependence is a more important issue than the interdependence between the firm's financial status and its investment decision when it comes to the fit of the models to the data.;Similar models are also used to reexamine the effect of the 1997 healthcare reform enacted in Germany using a panel data set on the number of doctor visits derived from the German Socio-Economic Panel by Winkelmann (2004) for the period going from 1995 to 1999. Again, we find evidence in favor of time dependence. However, controlling for the presence of random effects in the mixture components leads to models which fit much better to the data.