Bitcoin Volatility, Jump Risk, and Realized Variance Forecasting

Summary

This study examines Bitcoin risk using high-frequency returns and realized variance, with particular attention to jumps and their role in forecasting. It finds that conventional jump measures can be biased by frequent, extended price moves. Threshold-based jump estimation is proposed to better distinguish discontinuous moves from continuous variation. The analysis also separates upside and downside realized variation to investigate whether jump direction matters.

Using full-sample HAR-style regressions and rolling forecasts, the study reports that lagged realized variance predicts future variance, while positive jumps are associated with lower future variance at some horizons. Out-of-sample results depend on forecast length: jump and signed-jump measures add value at longer horizons, but not for short-term forecasts. The authors also report that combining Bitcoin data from multiple exchanges can reduce exchange-specific jump risk. Findings are based on Bitcoin data from 2017–2020, so their relevance to other assets, periods, and market structures is uncertain.

Key ideas

• High-frequency realized variance captures intraday price variation and can be decomposed into continuous and jump components.
• Frequent extended moves can bias standard jump estimates, motivating threshold-based estimators.
• Lagged realized variance helps forecast future Bitcoin variance, while positive jumps can have a negative association with later variance.
• Jump measures improve some longer-horizon forecasts, but add little to short-horizon forecasts in the reported tests.
• Combining prices across exchanges may reduce exposure to exchange-specific jump risk.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.