Sentiment, Herding and Volatility Forecasting: Evidence from GARCH-MIDAS Approach

Abstract

Investor sentiment and herding, as two important indicators of investors’ beliefs, both have great impacts on stock market volatility. Based on the daily transaction data in the Chinese stock market, we measure investor sentiment and herding on monthly basis from 2005 to 2020. Then, two indicators are, respectively, introduced to GARCH-MIDAS model for volatility forecasting. Moreover, their forecasting abilities are compared with five GARCH-type models in four crisis periods and two up markets through the model confidence set (MCS) test. We find that herding can better explain the long-term component of volatility than investor sentiment in GARCH-MIDAS model. What is more, both investor sentiment and herding possess excellent out-of-sample forecasting performances. Finally, the forecasting ability of herding is overwhelming compared with investor sentiment in crisis periods, while investor sentiment presents better prediction accuracy than herding in up markets. Our research shows that the forecasting abilities of investor sentiment and herding on volatility may be related to the market states.