Approximation with ergodic processes and testability (Q6617613)

Stationary time series taking values in a Polish space are identified with measure-preserving transformations, allowing approximation results and methods from ergodic theory to be applied. The main results give uniform approximation over all finite time intervals by processes satisfying various properties related to mixing, and by codings of aperiodic processes. In the non-parametric setting considered here this shows that the power a test for any of those mixing-type properties cannot exceed its size, so is not testable (in contrast to the theory of tests for ergodicity in the Markovian setting). These results also contrast with consistent estimation schemes for more restricted classes of time series. The approximation results are also used to deduce Baire category results for mixing properties in this setting.