Iterative regularisation in nonparametric instrumental regression |
| |
Authors: | Jan Johannes Sébastien Van Bellegem Anne Vanhems |
| |
Institution: | 1. Institut de statistique, biostatistique et sciences actuarielles, Belgium;2. Université catholique de Louvain, CORE, Voie du Roman Pays 34, 1348 Louvain-la-Neuve, Belgium;3. Universite de Toulouse, Toulouse Business School & Toulouse School of Economics, France |
| |
Abstract: | This paper considers the nonparametric regression model with an additive error that is correlated with the explanatory variables. Motivated by empirical studies in epidemiology and economics, it also supposes that valid instrumental variables are observed. However, the estimation of a nonparametric regression function by instrumental variables is an ill-posed linear inverse problem with an unknown but estimable operator. We provide a new estimator of the regression function that is based on projection onto finite dimensional spaces and that includes an iterative regularisation method (the Landweber–Fridman method). The optimal number of iterations and the convergence of the mean square error of the resulting estimator are derived under both strong and weak source conditions. A Monte Carlo exercise shows the impact of some parameters on the estimator and concludes on the reasonable finite sample performance of the new estimator. |
| |
Keywords: | Nonparametric estimation Instrumental variable Ill-posed inverse problem Iterative method Estimation by projection |
本文献已被 ScienceDirect 等数据库收录! |