Please rotate your device to landscape mode to view the charts.

Microstructure in the Machine Age

Journal: Review of Financial Studies

Date: 20210701

Author: Easley, David; Lopez de Prado, Marcos; O'Hara, Maureen; Zhang, Zhibai

Abstract:
Understanding modern market microstructure phenomena requires large amounts of data and advanced mathematical tools. We demonstrate how machine learning can be applied to microstructural research. We find that microstructure measures continue to provide insights into the price process in current complex markets. Some microstructure features with high explanatory power exhibit low predictive power, while others with less explanatory power have more predictive power. We find that some microstructure-based measures are useful for out-of-sample prediction of various market statistics, leading to questions about market efficiency. We also show how microstructure measures can have important cross-asset effects. Our results are derived using 87 liquid futures contracts across all asset classes.

Link: Google Scholar

Key Findings

Microstructure Features Still Matter

Market microstructure measures continue to provide valuable insights into price processes in modern complex markets, with some measures showing strong predictive power.

Cross-Asset Effects

Microstructure measures in one asset have significant predictive power for other assets, particularly from financial futures like Treasury bonds and currencies.

Machine Learning Value

Random forest models achieve prediction accuracy between 51-57% across multiple market variables, indicating potential market inefficiencies.

Prediction Accuracy Across Variables

  • Return variance shows highest predictability at 57.43%
  • All variables show accuracy above 50% random benchmark
  • Based on data from Table 14 using cross-asset features

Feature Importance Analysis

  • Own-asset features remain among most important predictors
  • Treasury and Currency features show strong cross-asset effects
  • Based on feature importance rankings from Table 11

Prediction Time Horizons

  • Accuracy remains robust across different time horizons
  • 50-bar horizon shows peak prediction accuracy
  • Based on Tables 12 and 14 forecast horizon comparisons

Contribution and Implications

  • Demonstrates continued relevance of market microstructure in modern electronic markets
  • Identifies important cross-asset relationships not captured by traditional single-asset models
  • Shows machine learning can effectively predict market dynamics with accuracy levels economically meaningful for trading strategies

Data Sources

  • Accuracy Chart: Constructed using data from Table 14 showing random forest prediction accuracy with cross-asset features
  • Feature Importance Chart: Based on Top 10 feature rankings from Table 11
  • Time Horizon Chart: Combines data from Tables 12 and 14 comparing prediction accuracy across different forecast windows