Forecasting Bitcoin volatility: A new insight from the threshold regression model

Asset returns, especially negative returns, represent the leverage effect and are found to be informative for forecasting financial market volatility. The purpose of this paper is to dig out more useful information in Bitcoin returns when we predict Bitcoin volatility. We use the threshold regression model to differentiate positive and negative returns. The threshold regression results suggest that not only a decrease in normal returns but also an increase in extremely positive returns would lead to an increase in future Bitcoin volatility. To further capture and address the leverage effect of large positive returns in the Bitcoin market, we propose a model switching method. We empirically demonstrate the significant predictive ability of large positive returns when forecasting Bitcoin volatility both in- and out-of-sample. And, our findings still hold after considering long-horizon forecasts, additional leverage effects, jump components, and a wide series of extensions and robustness tests. In an asset allocation exercise, the proposed volatility forecasting models, which highlight the predictive ability of large positive returns, can generate larger economic gains.