The paper compares two approaches to the estimation of panel probit models: the Generalized Method of Moments (GMM) and the Simulated Maximum Likelihood (SML) technique. Both have in common that they circumvent multiple integrations of joint density functions without the need to impose restrictive variance-covariance specifications on the error term distribution. Particular attention is paid to a three-stage GMM estimator based on nonparametric estimation of optimal instruments. A Monte Carlo study reveals slight efficiency gains from SML when the underlying model is correctly specified. GMM turns out to be more robust than SML when heteroskedasticity over time is ignored as well as in the presence of multiplicative heteroskedasticity. An application to the product innovation activities of German manufacturing firms is presented.
panel probit model conditional moment restrictions optimal instruments k-nearest neighbor estimation GHK simulator heteroskedasticity