The kernel approach to goodness-of-fit tests in functional regression
In this contribution, we address the problem of testing the goodness-of-fit (GoF) of a parametric regression model involving functional covariates. In particular, we explore a robust kernel-based approach and show that the theoretical developments introduced in recent proposals for the Euclidean case can be extended to more general settings in which parameters and covariates are infinite-dimensional. An easily computable test statistic is derived and its asymptotic behaviour is discussed. Moreover, a consistent multiplier bootstrap scheme is proposed for the calibration of the test and a succinct simulation study is conducted to illustrate its finite-sample performance.
Keywords: Functional data; Kernels; Goodness-of-fit; Bootstrap