Long Memory or Structural Change: Testing Method and Empirical Examination

In this paper, we focus on how to test for long-range dependence when the process may have a one-time mean change and how to estimate the change point when data may be long-range dependent. We first analyzed why traditional long-memory tests have serious size distortions when data have short memory with breaks. In order to overcome this problem, a local Whittle method is proposed. Simulation results confirm that our change-point estimator is well behaved even when data are long-range dependent, and that our test for long memory maintains proper size when a change is present. These results indicate that our method is practically useful and has a much wider applicability. In order to assess the empirical relevance of our procedure, we applied it to analyze monthly G7 inflation rates.