Robust functional estimation in the multivariate partial linear model (Q2317880)

The author considers a multivariate regression model of the form \[ Y_i = a + X'_i\beta + f(U_i) + \xi_i \] where the first two components represent the linear part of the model, the third component captures the nonlinear part, and the \(\xi_i\)'s are error terms with null median. Each \(U_i\) is a \(q\)-dimensional vector with coordinates on an equispaced grid on \([0,1]\). An adaptive estimator for the functional part of such a model is defined, and its properties in terms of robustness are proved. In particular, the proposed estimator is robust to a wide choice of distributions of the linear part and of the random errors. Robustness is particularly useful in econometrics where the distribution of the \(X_i\)'s is often not known, and the distributions of the error terms are far from being normally distributed.