Distinguishing Short and Long Memory Volatility Specifications

Abstract

Asset price volatility appears to be more persistent than can be captured by individual, short memory, autoregressive or moving average components. Fractional integration offers a very parsimonious and tempting formulation of this long memory property of volatility but other explanations such as structural models (aggregates of several autoregressive components) are possible. Given the ability of the latter to mimic the former, we investigate the extent to which it is possible to distinguish short from long memory volatility specifications. For a likelihood ratio test in the spectral domain, we investigate size and power characteristics by Monte Carlo simulation. Finally applying the same test to Sterling/Dollar returns, we draw conclusions about the minimum number of structural factors that must be present to mimic the long memory volatility properties that are empirically observed.