No label defined (Q4356554)

Measuring dependence and testing independence are important problems in all of statistics. We look at aspects of these problems in the framework of time series analysis. For a given stationary time series \(\{X_t\}\) we are interested in testing independence between the \(X_t\)'s. Such tests are of interest in diagnostic fitting but also in other contexts. Correlation type tests are mostly used within the ARMA model class. However, it is known that such tests have poor power against many nonlinear alternatives. The approach in this paper is more general. We will be concerned with nonparametric tests based on estimated densities. Then the power is reduced compared to the correlation test in ARMA models, but the tests retain power in nonlinear situations. Both asymptotic properties and bootstrap arguments will be used in constructing the tests. The bootstrap seems crucial in moderate sample sizes. There are many open problems, some of which we plan to explore in subsequent publications. Problems currently under investigation are the use of empirical distribution functions instead of estimated density functions, functionals extending over several lags and tests of conditional independence involving more than one time series. Also, in this paper our examples are restricted to first order models, although it is fairly obvious how the analysis can be extended to higher order models. For the entire collection see [Zbl 0846.00010].