A Nonparametric Goodness-of-fit Test for a Class of Parametric Autoregressive Models
Abstract
We derive a nonparametric test for discriminating between generalized autoregressive models. This test is based on a suitably normalized sum of residuals. The null distribution and the power of the test under both fixed and a sequence of local alternatives are studied under mild stationarity and $\alpha$--mixing conditions. This procedure can be applied to testing linear models against nonlinear models or certain nonlinear models against others. Numerical simulations show that the proposed test is powerful against most of the alternatives considered.