Using Indirect Inference to Solve the Initial Conditions Problem

Authors:

Abstract
In the econometric analysis of labor market transitions, the data generating process is often specified as a continuous-time semi-Markovian process with a finite state space. With typically short panel data, analysts have long been concerned with the initial conditions problem: a complication associated with the very first spells observed in the data which are often left-hand censored or interrupted. It has been well documented that consistent estimation, based on the specification of the full likelihood function, requires explicit functional form for the density function of the left-hand censored spells, conditional on the unobserved heterogeneity. Such a requirement cannot be met, except in very special cases. In this paper we propose an estimation procedure using indirect inference. This procedure maximizes a pseudo likelihood function and uses simulations of the structural model to eliminate the discrepancy of the pseudo maximum likelihood estimator. The indirect inference estimator is shown to be consistent and asymptotic normal. The variance and covariance matrix of the estimator can also be estimated. We describe in detail the pseudo maximum likelihood estimation, the auxiliary parameter estimates and the simulation algorithm. We also implement the procedure using a Dutch data set on labor market histories.

This publication has 23 references indexed in Scilit: