Overnight jumps are large price moves that occur between a market's previous close and the next open and represent a distinct component of total volatility. Models that ignore these gaps risk underestimating tail risk and mispricing derivatives, especially for instruments sensitive to end-of-day exposures.
Model classes that capture overnight jumps
Jump-diffusion models that include an explicit overnight jump term perform well when jumps are driven by discrete news events. Research by Torben G. Andersen at Northwestern University and Tim Bollerslev at Duke University shows that separating continuous intraday volatility from discontinuous jump components improves forecast accuracy when high-frequency data are available. Nonparametric jump detection using bipower variation developed by Ole E. Barndorff-Nielsen and Neil Shephard at the University of Oxford isolates jumps from continuous variation and is widely used to flag overnight jumps before modeling.
Realized volatility frameworks augmented for jumps are particularly effective in practice. Matteo Corsi at Bocconi University introduced the HAR-RV model to capture persistent volatility components across horizons; however in its basic form HAR-RV can underreact to isolated overnight shocks. Combining realized measures with jump-robust estimators and including a separate close-to-open term yields more accurate forecasts for open-to-open risk. Robert F. Engle at New York University developed GARCH extensions that incorporate realized measures and jump terms, and these Realized GARCH specifications better reconcile intraday information with daily outcomes when overnight jumps are material.
Relevance, causes, and consequences
Overnight jumps often stem from scheduled macroeconomic releases, corporate announcements, geopolitical developments, or sudden liquidity evaporation during off-hours. Market structure and geography matter: markets with thin overnight liquidity or large timezone mismatches between economic news sources and trading hours tend to exhibit larger and more frequent overnight gaps. The consequences include biased Value-at-Risk assessments, hedging slippage for end-of-day portfolios, and mispriced options if implied volatility conventions do not reflect jump risk.
Practical modeling guidance is to combine nonparametric jump detection such as bipower variation with a two-part volatility specification that models open-to-close continuous volatility and close-to-open jump risk separately, and to estimate jump intensities rather than assume their absence. Empirical work by Andersen, Bollerslev, Engle, Corsi, Barndorff-Nielsen, and Shephard across Northwestern University, Duke University, New York University, Bocconi University, and the University of Oxford supports this hybrid approach as the most robust for capturing overnight jump risk.