SEMIPARAMETRIC ESTIMATION OF A QUASI-SCORE

The method of maximum quasi-likelihood estimation gives sa-tisfactory results in a parametric regression model, where the link function r an d the variance function V are well specified. In semiparametric models , when the functions r and V are unknown, this method fails. Neverthel ess, it is possible to define the quasi-score function and its estimat ion, computed from kernel regression estimators of the functions r and V. We propose an estimator for the regression coefficients based on a one step Newton-Raphson iteration in a maximum quasi-likelihood optim ization starting from an initial root n-consistent estimate and using the estimated quasi score. We derive the asymptotic properties of this estimator and its semi parametric efficiency.