Observation-driven Models for Realized Variances and Overnight Returns

Abstract

We present a new model to decompose total daily return volatility into a filtered (high-frequency based) open-to-close volatility and a time-varying scaling factor. We use score-driven dynamics based on fat-tailed distributions to limit the impact of incidental large observations. Applying our new model to 100 stocks of the S&P 500 during the period 2001-2014 and evaluating (in-sample and out-of-sample) in terms of Value-at-Risk and Expected Shortfall, we find our model outperforms alternatives like the HEAVY model that uses close-to-close returns and realized variances, and models treating close-to-open en open-to-close returns as separate processes. Results also indicate that the ratio between total and open-to-close volatility changes substantially through time, especially for financial stocks.