ABSTRACT

Introduction There is a growing literature on the modelling of temporal dependencies in financial market volatility. To some extent the theoretical under-pinning for these dynamic dependencies has lagged behind. However, the so-called mixture-of-distributions hypothesis (MDH) does provide a rationale for the many empirical studies that have found evidence of a strong positive correlation between returns volatility and trading volume. According to MDH, returns volatility and trading volume are driven by the same latent news (information) arrival variable. The arrival of good news results in increased trading, as the market adjusts to a new equilibrium, and a price increase, while the arrival of bad news results in increased trading and a price fall. Consequently, returns volatility and trading volume should be positively and contemporaneously correlated. A problem in testing this implication of the MDH is the likelihood that the news arrival process has a long memory property. It follows then that both volatility and volume will have the long memory property. Bollerslev and Jubinski (1999) show that in the presence of this long memory property the contemporaneous correlation between volatility and volume is likely to be incorrectly rejected in cases where the test equation does not account for long memory (or persistence). The use of fractionally integrated GARCH (FIGARCH) offers a way to take account of long memory (and indeed to test for long memory) in testing for a contemporaneous correlation between volatility and volume (an implication of the MDH).