Efficient Estimation in Convex Single Index Models
share
We consider estimation and inference in a single index regression model with an unknown convex link function. We propose two estimators for the unknown link function: (1) a Lipschitz constrained least squares estimator and (2) a shape-constrained smoothing spline estimator. Moreover, both of these procedures lead to estimators for the unknown finite dimensional parameter. We develop methods to compute both the Lipschitz constrained least squares estimator (LLSE) and the penalized least squares estimator (PLSE) of the parametric and the nonparametric components given independent and identically distributed (i.i.d.) data. We prove the consistency and find the rates of convergence for both the LLSE and the PLSE. For both the LLSE and the PLSE, we establish n^-1/2-rate of convergence and semiparametric efficiency of the parametric component under mild assumptions. Moreover, both the LLSE and the PLSE readily yield asymptotic confidence sets for the finite dimensional parameter. We develop the R package "simest" to compute the proposed estimators. Our proposed algorithm works even when n is modest and d is large (e.g., n = 500, and d = 100).
READ FULL TEXT
