Abstract:
Unlike equity returns, many fixed-income return and volatility measures appear to display considerable long memory. Connolly and G½ner (working paper, 1999) show this holds particularly strongly for shorter-maturity Treasury securities in the U.S. They show that fixed-income return and volatility series for other G-7 countries also display long memory. In a number of recent papers, Granger has argued that long memory in returns may only reflect infrequent structural breaks. He finds the case for long memory in volatility much stronger. Parke (Review of Economics and Statistics, 2000) develops a model that generates long memory through a type of structural shift. In this paper, we show that the extent of long memory depends crucially on whether gross or excess returns are under consideration and we provide a simple demonstration of why this distinction is so important. We also carefully explore the impact of structural instability on tests for long memory using a version of the supLM test developed by Andrews. Briefly, we find that evidence of long memory in both gross and excess returns is very considerably weakened after accounting for structural shifts. By contrast, we find that long memory in excess return volatility appears to exist even after controlling for structural shifts. Finally, our estimates of the long memory parameter for excess return volatility series shift from the nonstationary to the stationary range once we account for structural shifts. As a part of the analysis, we also compare the performance of different estimators of long memory volatility model parameters. We conclude the paper with some discussion of the implications of long memory in excess return volatility for asset pricing tests and risk management.