The impact of overnight returns on realized volatility

Abstract

We obtain intraday data on three stock indices listed on the Taiwan Stock Exchange (TWSE), and then analyse the data by incorporating an overnight returns indicator into the 'Heterogeneous Auto-Regressive' (HAR) model of realized volatility. Our overall aim is to enhance the forecasting of future volatility. Our findings demonstrate that the modified model significantly improves the forecasting performance of future realized volatility, with our results also being found to continue to hold for both in sample and out of sample forecasts.

Volatility forecasting: The role of lunch-break returns, overnight returns, trading volume and leverage effects

“…Thus, the HAR-RV model can capture the characteristics of the volatility well and is used widely in the literature on financial volatility. Moreover, it has already been extended by including jump components (Anderson & Vahid, 2007), leverage effects (Corsi & Renò, 2012), and absolute overnight returns (Tseng, Lai, & Lin, 2012). Proceeding in this way, we focus on volatility forecasting for the Chinese stock market by including lunchbreak returns, overnight returns and trading volumes, in addition to the negative daily returns, using the specification…”

Forecasting Volatility with Time-Varying Coefficient Regressions

“…Trading hours for the US stock market is the time when the Chinese stock market is closed. Overnight returns have been used as a proxy of overnight information flow to enhance forecast accuracy of volatility modes in recent literature [12,28,29]. Tseng et al [28] argued that the impact of overnight returns on future volatility is also asymmetric.…”

The information content of open interest for the realized range-based volatility: evidence from Chinese futures market

“…Thus, this paper uses an efficient range-based estimator to describe the dynamic volatility in the HAR mode. We adopt RTV as the regressor for prediction of realized range-based volatility, similar to Tseng et al (2012) and Todorova and Souček (2014), with the following specification of the HAR-RRV-RTV model:…”