Cleaning Trade Data: Outliers, Busts, and Fat Fingers
Raw trade feeds contain prints that never reflected a real market. A disciplined approach to detecting them without deleting genuine extremes.
Trade feeds occasionally contain prints far from the prevailing market: a mistyped order that swept a thin book, a trade later cancelled by the venue, or a message corrupted in transit. A single such print can set a daily high, spike realized volatility, or trigger a simulated stop-loss that never would have filled.
Detection approaches
- Deviation from a robust local reference, such as a rolling median of recent trades.
- Deviation from the concurrent quote midpoint on the same venue.
- Deviation from prices on other venues at the same moment.
- Venue notices of cancelled or busted trades.
The difficulty is that genuine extreme moves also look like outliers. A real flash crash that swept the book on one venue is exactly the kind of event risk research needs to retain. Cross-venue confirmation and the presence of follow-on trades at similar prices help separate the two.
This publication is provided for informational purposes only and does not constitute investment, legal, or tax advice, or an offer or solicitation to buy or sell any asset. Live figures are computed from third-party public market data and may be delayed, incomplete, or inaccurate.
