Power properties of linearity and long memory tests: A simulation study

Abstract

We show that specific long memory which is common in internet traffic data can hardly be distinguished from nonlinear time series model such as Markov switching by standard methods such as the GPH estimator for the memory parameter or linearity tests. We show by Monte Carlo that under certain conditions, the nonlinear data generating process can have misleading either stationary or non-stationary long memory properties. © 2011 Springer-Verlag.