A bootstrap test for the comparison of nonlinear time series

We study the drift of stationary diffusion processes in a time series analysis of the autoregression function. A marked empirical process measures the difference between the nonparametric regression functions of two time series. We bootstrap the distribution of a Kolmogorov-Smirnov-type test statistic for two hypotheses: Equality of regression functions and shifted regression functions. Neither markovian behavior nor Brownian motion error of the processes are assumed. A detailed simulation study finds the size of the new test near the nominal level and a good power for a variety of parametric models. The two-sample result serves to test for mean reversion of the diffusion drift in several examples. The interest rates Euribor, Libor as well as T-Bond yields do not show that stylized feature often modelled for interest rates.