A test of nonlinear dependencies

This paper proposes a test of iid distribution of data that allows to uncover nonlinear patterns in a time series. The proposed alternative (henceforth the K2K test) extends and generalizes the widely known BDS test. By its construction, it removes subjectivity in arbitrary choice of a proximity parameter. The Monte Carlo simulation is used to tabulate critical values of the alternative statistic. Previously published empirical studies are replicated as well as power tests executed in order to evaluate performance of the K2K test relative to the BDS test. The results are favorable for the K2K test.