Simple Semiparametric Z-Estimation for Bundled Parameter Models

Image credit: Unsplash


We propose a new iterative estimation algorithm for use in semiparametric models where calculation of Z-estimators by conventional means is difficult or impossible. Unlike a Newton–Raphson approach, which makes use of the entire Hessian, this approach only uses curvature information associated with portions of the Hessian that are relatively easy to calculate. Consistency and asymptotic normality of estimators obtained from this algorithm are established under regularity conditions and an information dominance condition. Two specific examples, a quantile regression model with missing covariates and a GARCH-in-mean model with conditional mean of unknown functional form, demonstrate the applicability of the algorithm. This new approach can be interpreted as an extension of the maximization by parts estimation approach to semiparametric models.

Econometric Theory
David T. Frazier
DECRA Fellow and Associate Professor in Econometrics and Statistics

My research interests include simulation-based statistical theory and inference, approximate Bayesian analysis, forecasting and scoring rules.