Panel Regression with Unobserved Classes

We propose a panel regression model with a predetermined and fixed number of classes, where each class is defined by its parameters, but any reference as to which group any observation belongs to is absent. The classes or groups are rationalized by a willingness to attribute some of the observed heterogeneity on a higher level than the individual. The estimation procedures have a distinct Bayesian flavor, relying on the Gibbs sampler for parameter estimation, a method proven effective in situations with missing or latent variables.