Entropy testing for nonlinear serial dependence in time series

open3siThis work was supported by the Italian Ministry of University and Research. Simone Giannerini acknowledges the support of the Institute for Mathematical Sciences of the National University of Singapore.We propose tests for nonlinear serial dependence in time series under the null hypothesis of general linear dependence, in contrast to the more widely studied null hypothesis of independence. The approach is based on combining an entropy dependence metric, which possesses many desirable properties and is used as a test statistic, with a suitable extension of surrogate data methods, a class of Monte Carlo distribution-free tests for nonlinearity, and a smoothed sieve bootstrap scheme. We show how, in the same way as the autocorrelation function is used for linear models, our tests can in principle be employed to detect the lags at which a significant nonlinear relationship is present. We prove the asymptotic validity of the proposed procedures and the corresponding inferences. The small-sample performance of the tests in terms of power and size is assessed through a simulation study. Applications to real datasets of different kinds are also presented.embargoed_20160630Simone Giannerini; Esfandiar Maasoumi; Estela Bee DagumSimone Giannerini; Esfandiar Maasoumi; Estela Bee Dagu