Relaxing the Strict Exogeneity Assumption in a Dynamic Random Probit Model

This paper is relaxing the strict exogeneity assumption in a dynamic random probit model to allow for the possibility of feedback effects. We take an MLE approach and specify a marginal distribution for the not strictly exogenous variable in question. Using a log-likelihood function similar to Wooldridge (2000) we propose two estimation strategies depending on what the object of interest is. We show that the parameters can be estimated using either quadrature or simulated maximum likelihood if all we are interested in is the parameters of the model. Subsequently average partial effects can be estimated. However, if we are more interested in knowing the average partial effects and less interested in the parameter estimates themselves, then it is useful to considering the problem as a method of moment problem rather than a MLE. This will allow an easy estimation of the average partial effect and in particular the variance of the APE such that statistical inference is possible. The insight is applied to a large Danish register data set on employment transitions to address the question of true state dependence in unemployment transitions. Moreover, we rise the important question, that a major part of the results in the state dependence literature could be invalid due to ignoring violations of the strict exogeneity assumption.