Analysis of the predictive ability of information accumulated over nights, weekends and holidays

Recent empirical evidence suggests that the weekend and holiday calendar effects are much stronger and statistically significant in volatility as opposed to expected returns. This paper seeks an explanation for this empirical finding by undertaking a comprehensive investigation of the predictive ability of information accumulated over nights, weekends and holidays for a series of global indices. We study this form of seasonal heteroscedasticity by employing a generalized stochastic volatility model, in which the conditional daily volatility is measured in calendar time from open-to-close of the market, and depends on lagged close-to-open returns. We conduct a series of empirical tests and conclude that the information accumulated over weekends and especially holidays is a predictor of subsequent daily volatility. The SV parameters are estimated by implementing a Bayesian MCMC algorithm, which is adjusted for sampling the seasonal volatility level effects. We compute in-sample and out-of-sample density forecasts for assessing the adequacy of the conditional distribution. We also use Bayes factors as a likelihood-based framework for evaluating the SV specifications. Bayes factors account for both estimation and model risk. We conclude by computing volatility forecasts relevant for risk management