A1215
Title: Pseudo-indirect inference for binary dynamic models
Authors: Fausto Galli - University of Salerno (Italy) [presenting]
Antonio Cosma - University of Bergamo (Italy)
Abstract: The aim is to propose a pseudo-indirect inference method for estimating binary dynamic logit models with unobserved individual heterogeneity using short panel data. Traditional fixed-effects estimators in nonlinear panel models suffer from the incidental parameters problem under large-N asymptotics. The approach circumvents this by simulating data from a pseudo-true model that assumes independence between individual effects and covariates, an assumption not imposed on the true model. This enables consistent estimation of structural parameters without requiring knowledge of the joint distribution of unobserved heterogeneity and regressors. The method minimizes the distance between observed and simulated auxiliary estimates derived from either conditional or unconditional likelihood criteria. Although the simulation model is misspecified, it is shown that the resulting estimator remains root-N consistent. Monte Carlo experiments suggest the practical validity of the approach across various data-generating processes, confirming robust convergence properties. This framework extends indirect inference to a class of models where existing estimation techniques are either inconsistent or require strong assumptions, offering a flexible and computationally feasible alternative for dynamic discrete panel data analysis.
