Structural Change and Time Dependence in Models of Stock Returns

In this paper, we provide evidence that the time-series properties of stock returns includeboth structural change and time dependence in the conditional variance. The absence of astructural change component tends to overstate the persistence parameter in a time-dependentmodel specification. The reason is that time-dependent model specifications do notdistinguish between positive residuals that increase volatility from those that represent aresolution of uncertainty. A sequential mixture of normal distributions model of structuralchange is employed to estimate discrete change points in the time-series of volatility ineither direction. Although the model of structural change appears to be the more descriptiveprocess in a mutually exclusive comparison, a joint model of time dependence andstructural change is most likely