SIMULATION-BASED ESTIMATION OF MODELS WITH LAGGED LATENT-VARIABLES

We extend here our earlier work (Laroque-Salanie, 1989) and propose a dynamic simulated pseudomaximum likelihood method to deal with a very general class of dynamic non-linear models, including models with lagged latent variables. We test this method on Monte Carlo-generated data for a canonical disequilibrium model. It appears to provide very satisfactory estimates at little computational cost. However, accurate estimation of the standard errors of the estimates may require some care in nondifferentiable models.