Does Google search index really help predicting stock market volatility? Evidence from a modified mixed data sampling model on volatility

Accurately predicting volatility plays an important role in many financial interactions and has already attracted considerable attention from both academics and practitioners. Much effort has been devoted to looking for key determinants of volatility and developing more powerful models. In this paper, we investigate the influences of high frequency event impact and low frequency macroeconomic fundamentals on volatility. To handle the mixed frequency data, we develop a modified GARCH-MIDAS model, multiple factors GARCH-MIDAS (MF-GARCH-MIDAS), to allow for a long-run component driven by multiple factors sampled at different frequencies. We present the technical details of the MF-GARCH-MIDAS model and apply it to predict monthly volatility of the US DJIA. In our empirical illustration, we consider weekly Google trends (GT) as a proxy for event impact and use three US macroeconomic variables: quarterly GDP, monthly PPI, and monthly IP. Our empirical findings support that event impact measured by GT is also a source of volatility besides macroeconomic fundamentals. The economic implications are significant in that both event impact and macroeconomic fundamentals really matter for volatility forecasting and the combination GT+PPI+GDP in M2 performs best in predicting volatility of DJIA. •We develop a novel multiple factors GARCH-MIDAS (MF-GARCH-MIDAS) model.•It allows for a long-run volatility driven by multiple factors sampled at different frequencies.•It is used to investigate the usefulness of Google trends in predicting stock market volatility.•The empirical results show favorable evidence about our proposed model.•We find that Google trends really matters for volatility forecasting.