Testing for strict stationarity in financial variables

This paper considers the stationarity properties of a variety of financial variables using statistical tests for strict stationarity. We find that there has been a gradual shift in unconditional variances for the variables examined during the 90's and 2000's and that this is the main cause of the widespread rejection of the strict stationarity null hypothesis. This is a powerful result which suggests that the consideration of conditional mean and, especially, conditional variance models which assume stationarity is problematic for the period under examination. This casts serious doubts on the usefulness of models that assume strict stationarity and model conditional second moments, such as GARCH and stochastic volatility models.