Sign Tests for Long-memory Time Series

Abstract

This paper proposes sign-based tests for simple and composite hypotheses on the long-memory parameter of a time series process. The tests allow for nonstationary hypothesis, such as unit root, as well as for stationary hypotheses, such as weak dependence or no integration. The proposed generalized Lagrange multiplier sign tests for simple hypotheses on the long-memory parameter are exact and locally optimal among those in their class. We also propose tests for composite hypotheses on the parameters of ARFIMA processes. The resulting tests statistics have a standard normal limiting distribution under the null hypothesis.