A Varieties of Spurious Long Memory Process
Charfeddine Lanouar

Abstract
This paper introduces a large varieties of models which can generate a spurious long memory behavior. By using Monte Carlo simulations have investigated the behavior of the GPH, the Exact Local Whittle and the Wavelet methods when the data generating process is weakly dependent with changes in mean. The results show that these long memory estimation methods fail to accept the null hypothesis of short memory. A strong size distortion have been detected. The results stress the need for a test statistic which can discriminate the true long memory behavior from the spurious behavior. In the empirical applications we use stock returns of seven French companies with the CAC 40 stock market index. We show that the observed long memory behavior is a true behavior and that presence of breaks in these time series do not have a major effect on the long memory parameter.

Full Text: PDF