Abstract
In this article, we introduce \textsf{LASER} (Locally Adaptive SmoothingEstimator for Regression), a computationally efficient locally adaptivenonparametric regression method that performs variable bandwidth localpolynomial regression. We prove that it adapts (near-)optimally to the localH\"{o}lder exponent of the underlying regression function\texttt{simultaneously} at all points in its domain. Furthermore, we show thatthere is a single ideal choice of a global tuning parameter under which theabove mentioned local adaptivity holds. Despite the vast literature onnonparametric regression, instances of practicable methods with provableguarantees of such a strong notion of local adaptivity are rare. The proposedmethod achieves excellent performance across a broad range of numericalexperiments in comparison to popular alternative locally adaptive methods.