Least squares estimation in the monotone single index model (Q2325372)

The authors study the semi-parametric single index model, where the conditional expectation of the outcome variable is a nonlinear transform of a linear combination of the explanatory variables. The authors additionally assume that this nonlinear transformation is monotone. The regression function can be estimated by the least square estimator and the authors show that this estimator is not unique, but converges in \(L_2\) to the true regression function with rate \(n^{(-1/3)}\) under some assumptions (mainly boundedness of the covariates and some conditions on their density). They also show that the estimator of the coefficients of the linear combination converges with the same rate. The \(n^{(-1/3)}\)-rate is illustrated in a simulation study. An alternative estimator based on splitting the sample into two parts is also discussed. The proof is based on results about the bracketing entropy of the class of regression functions.