PERSISTENCE OF STOCK RETURN VOLATILITY IN CANADA

In this article, a generalized autoregressive conditional heteroscedastic (GARCH) model is applied to Canadian data covering the period 1976-1991. The time period is chosen to correspond to distinct monetary regimes, and to the more recent period surrounding the October 1987 crash. The excess price volatility observed over the entire period is reduced by appropriately taking into account three conditional mean and variance shifts corresponding to three distinct monetary regimes and interest rate processes. In spite of the above correction, the hypothesis of a unit root in the variance cannot be rejected for one sub-period. Subsequently a substantial reduction in the degree of persistence in volatility is achieved by introducing a dummy variable in the GARCH model to account for extraordinary price movements or ''outliers'. The outliers were investigated and found to correspond to the release of business and economic news, in most cases of U.S. origin. Conversely, no reduction in conditional variance is obtained when allowing for seasonal effects, such as the Monday and January effects, in the GARCH equation. The results suggest that, when modelling the conditional variance equation, it is important to account for structural changes in the economic environment and to include in the analysis other sources of non-stationarity. In doing so, the degree of persistence in market volatility is vastly reduced. The conclusion calls upon further study of the impact of extraordinary price movements and their link to economic information, as well as of other factors likely to affect volatility persistence, such as trading behaviour, in the search to unravel the dynamic behaviour of conditional variances.