THE INITIAL CONDITIONS PROBLEM IN WORK

A random effect specification controls for heterogeneity and provides consistent estimates of unemployment and employment hazard parameters from an uninitialized work history data set. The initial conditions problem, arising because the first spell in the sample period has no known starting date, is solved by a Bayesian technique given by Chamberlain, in which the random effect distribution is conditioned on forward recurrence information. Several models are estimated which indicate the biases involved in ignoring heterogeneity and forward recurrence information, as well as the bias resulting from an assumption of independence of random effects across states.