[Paper Review] Explainable Patterns in Cryptocurrency Microstructure

Bartosz Bieganowski, Robert Ślepaczuk|arXiv (Cornell University)|Jan 31, 2026

Financial Markets and Investment Strategies0 citations

TL;DR

The paper shows that a unified set of top-of-book and trade-derived features predict short-horizon crypto returns across assets with vastly different capitalizations, using SHAP to reveal consistent, microstructure-aligned dependence patterns and robustness under a major flash crash.

ABSTRACT

We document stable cross-asset patterns in cryptocurrency limit-order-book microstructure: the same engineered order book and trade features exhibit remarkably similar predictive importance and SHAP dependence shapes across assets spanning an order of magnitude in market capitalization (BTC, LTC, ETC, ENJ, ROSE). The data covers Binance Futures perpetual contract order books and trades on 1-second frequency starting from January 1st, 2022 up to October 12th, 2025. Using a unified CatBoost modeling pipeline with a direction-aware GMADL objective and time-series cross validation, we show that feature rankings and partial effects are stable across assets despite heterogeneous liquidity and volatility. We connect these SHAP structures to microstructure theory (order flow imbalance, spread, and adverse selection) and validate tradability via a conservative top-of-book taker backtest as well as fixed depth maker backtest. Our primary novelty is a robustness analysis of a major flash crash, where the divergent performance of our taker and maker strategies empirically validates classic microstructure theories of adverse selection and highlights the systemic risks of algorithmic trading. Our results suggest a portable microstructure representation of short-horizon returns and motivate universal feature libraries for crypto markets.

Motivation & Objective

• Motivate the search for universal microstructure features that predict short-horizon crypto returns across assets with different liquidity and capitalization.
• Develop a portable, interpretable modeling pipeline using a unified feature library and CatBoost with a direction-aware GMADL objective.
• Use SHAP to diagnose cross-asset consistency in feature importance and dependence shapes, linking to microstructure theory.
• Assess economic significance through taker and maker backtests and stress-test robustness during a flash crash.

Proposed method

• Engineer a unified feature library from top-of-book and trade-flow metrics (spreads, VWAP deviations, order flow imbalance).
• Train gradient-boosted trees (CatBoost) with a direction-aware GMADL objective and forward-looking 1-second horizon, with time-series cross-validation.
• Explain model predictions with SHAP to obtain global feature rankings and local dependence plots.
• Validate cross-asset stability by comparing SHAP patterns across BTC, LTC, ETC, ENJ, ROSE and through robustness checks (R2-optimized model).
• Backtest trading signals under taker and maker execution assumptions, including a conservative cash-marking scheme to assess economics and risk.

Experimental results

Research questions

• RQ1 Do the same engineered microstructure features exhibit similar predictive importance across crypto assets spanning large to small market capitalizations?
• RQ2 Are the SHAP dependence shapes of key features (order flow imbalance, spreads, VWAP-to-mid deviations) consistent across assets?
• RQ3 Do the predictive patterns translate into tradable signals under conservative execution rules, and how do they behave during a major flash crash?
• RQ4 How does tick size influence the strength of imbalance effects and the microprice mechanism across assets?

Key findings

• Feature importance and SHAP dependence shapes for order flow imbalance, spreads, and VWAP deviations are stable across assets with different liquidity.
• Dependence patterns are broadly monotone for imbalance with concavity at extremes; wider spreads reduce predictability; VWAP-to-mid deviations show short-horizon asymmetry consistent with microstructure reversion.
• Taker execution yields economically meaningful returns and information ratios on several assets, with significant results for ETC, ENJ, and ROSE; maker execution shows more mixed performance.
• A major flash crash stresses the framework: taker strategies exploit abrupt moves while maker strategies suffer from adverse selection, validating classic microstructure theories.