FOR 1735/2: Efficient Nonparametric Regression When the Support Is Bounded (TP 03)

If in nonparametric regression the support of the error distribution has a sharp boundary, then the regression function and functionals thereof can be estimated with a higher rate of convergence than in regular models. We will first examine the geometry of such irregular statistical experiments and then develop efficient statistical procedures that adapt both to the smoothness of the regression function and to the degree of irregularity of the error distribution. Moreover, goodness-of-fit tests for the model assumptions will be constructed and analysed.