Job durations with worker and firm specific effects : MCMC estimation with longitudinal employer-employee data

We study job durations using a multivariate hazard model allowing for worker-specific and firm-specific unobserved determinants. The latter are captured by unobserved heterogeneity terms or random effects, one at the firm level and another at the worker level. This enables us to decompose the variation in job durations into the relative contribution of the worker and the firm. We also allow the unobserved terms to be correlated. For the empirical analysis we use a Portuguese longitudinal matched employer-employee data set. The model is estimated with a Bayesian Markov Chain Monte Carlo (MCMC) estimation method. The results imply that firm characteristics explain around 30% of the variation in log job durations. In addition, we find a positive correlation between unobserved worker and firm characteristics. -- Job transitions ; assortative matching ; Gibbs sampling ; frailties ; dynamic models ; matched employer-employee data