Tests for the error distribution in nonparametric possibly heteroscedastic regression models

摘要

Consistent procedures are constructed for testing the goodness-of-fit of the error distribution in nonparametric regression models. The test starts with a kernel-type regression fit and proceeds with the construction of a test statistic in the form of an L_(2) distance between a parametric and a nonparametric estimates of the residual characteristic function. The asymptotic null distribution and the behavior of the test statistic under alternatives are investigated. A simulation study compares bootstrap versions of the proposed test to corresponding procedures utilizing the empirical distribution function.