Modelling Income Processes with lots of heterogeneity

All empirical models of earnings processes in the literature assume a good deal of homogeneity. For example, all authors assume either that everyone has a unit root process or that everyone has a stationary process. In contrast to this we model earnings processes allowing for lots of heterogeneity between agents. To do this we have to formulate a series of increasingly complex processes which make maximum likelihood or GMM procedures very onerous. To avoid this we use a simulated minimum distance (SMD) estimation procedure. this is the first time that such an estimator has been applied to dynamic panel data models. We fit our models to a variety of statistics including most of those considered by previous investigators (for example, trends in the cross-section variance and transition probabilities from low income states). The principal sample we use is of a group of Danish male workers followed for 16 years. The sample we draw is very homogeneous in terms of observables such as education, age, experience, marital status and all have full year, full time employment during the period considered. Despite this observable homogeneity we find much greater latent heterogeneity than previous investigators. Applying our methods to a more heterogenous sample drawn from the PSID we find a completely different but still very heterogenous process is needed. This suggests that not only do processes vary a lot within groups they also vary between different samples so that detailed modelling is required in each instance. We show that allowance for heterogeneity makes substantial differences to inferences of interest. For example, we find that workers appear to trade off mean for variance in their choice of earnings process. Such a conclusion would be ruled out by a model that did not allow for correlated heterogeneity.