Combining Order Book Depth and Trade Flow to Improve Mid-Price Estimates

Summary

This article develops adjusted mid-price estimates from high-frequency order book and transaction data. Using top-of-book bid and ask quantities, it starts with the standard midpoint and tests volume-weighted and nonlinear imbalance adjustments. It then examines imbalance at deeper book levels and adds trade-flow measures based on the rate, quantity, and volume of buy and sell transactions. Predictions are evaluated against subsequent transaction prices with summed squared error.

In the sample, adding deeper-level imbalance yields modest improvements, while transaction volume imbalance reduces error more noticeably; a combined measure gives the lowest reported error among the variants tested. The author describes the fitted weights as exploratory and partly chosen by trial and error. The data are limited to one trading pair, a subset of depth observations, and simplified timing that omits latency and live bid–ask updates. The resulting estimates can even fall outside the spread, and lower prediction error alone does not establish profitable execution or generalizability.

Key ideas

• The standard midpoint can be adjusted using the imbalance between bid and ask quantities.
• The article finds that the first levels of book depth carry more useful signal than deeper levels.
• Trade-flow volume imbalance improves the reported price prediction error more than book imbalance alone.
• Combining book and trade data produces the lowest reported error among the tested estimates.
• The results are exploratory and limited by one sample, simplified timing, and trial-and-error weights.

Tags

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