Cryptocurrency markets trade continuously across time zones, so capturing seasonality is essential before interpreting high-frequency volatility. Classic high-frequency finance research shows that a large part of intraday variation is deterministic and must be removed to avoid biased volatility estimates. Robert Engle at New York University Stern School of Business and Torben G. Andersen at Northwestern University emphasize preprocessing intraday patterns when applying ARCH-type and realized-volatility models, because residual dynamics then reflect genuine stochastic risk rather than predictable timing effects.
Estimating and removing deterministic intraday patterns
Start with a flexible estimate of the deterministic intraday component using nonparametric smoothing, Fourier series, or splines, and allow for day-of-week and hour-of-day interactions. Matteo Corsi at Bocconi University demonstrated how multi-horizon structures matter for volatility persistence, implying that seasonality removal should precede HAR and other long-memory models. Deseasonalization can be applied to raw returns or to high-frequency realized measures; the aim is consistent scaling so that subsequent models target stochastic volatility rather than calendar-driven variability.
Noise, jumps, and territorial cycles
High-frequency crypto data contains microstructure noise and frequent jumps from news or large trades. Ole E. Barndorff-Nielsen and Neil Shephard at University of Oxford developed robust realized-variance techniques that separate continuous variation from jumps, which is critical when seasonality interacts with event-driven spikes. Use pre-averaging or robust realized estimators to mitigate bid-ask bounce and exchange-specific quirks, then model the cleaned series with HAR, GARCH, or state-space stochastic volatility frameworks.
Seasonality in crypto carries cultural and territorial nuance: Asia–Pacific trading peaks, U.S. business hours, and weekend retail behavior create systematic weekly and hourly patterns that differ across exchanges and asset types. These patterns are not static; regulatory announcements, miner behavior, and liquidity shifts can change seasonal shapes, so re-estimate intraday components regularly.
Modeling practice should therefore follow a pipeline: estimate and remove deterministic seasonality, correct for microstructure noise and jumps using robust realized measures, and then fit stochastic-volatility or heterogeneous-autoregressive models that include exogenous seasonality controls when forecasting or computing risk metrics. This approach aligns with best-practice methods from Andersen, Bollerslev at Duke University, and Francis X. Diebold at the University of Pennsylvania for realized-volatility analysis and ensures seasonality is treated as a separate, estimable phenomenon rather than conflated with high-frequency risk. Adapting this pipeline to each asset and venue preserves interpretability and improves risk management in 24/7 crypto markets.