Fusing Order Book Feeds for More Realistic Backtests
Summary
The document explains why a single exchange depth stream may not capture every order-book change. It compares Binance Futures incremental Level 2 data with the more frequently updated book-ticker feed, then shows how to combine them into a consolidated feed for replay and backtesting. The workflow records best bid and ask prices and quantities, compares the separate and fused streams, and evaluates a market-making strategy against each data set.
The example reports that fusion can change simulated fills, positions, and equity, with divergences tied to strategy order placement. Fused data can improve update frequency and granularity, but it takes longer to process. The method also has important reconstruction limits: events are processed by local receipt time, older exchange-timestamped updates can be discarded, and differing snapshots can leave parts of the consolidated book inconsistent. Maintaining a separate book per feed would address that consistency issue. The results are illustrative rather than a universal accuracy guarantee; the right fidelity and runtime trade-off depends on the strategy and its modeling needs.
Key ideas
- Exchange depth feeds may aggregate updates, while a separate BBO feed can provide more frequent changes.
- Fusing depth and book-ticker data can improve the update frequency used in simulated order books.
- Differences in BBO data can change order fills, positions, and backtest equity.
- Local receipt ordering and exchange timestamps can produce stale or inconsistent book levels.
- More frequent fused data increases backtest processing time, so fidelity has a runtime cost.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.