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

Background and Context

Research Purpose

The study investigates how machine learning can help analyze modern market microstructure and predict market behavior in an era of high-frequency, computerized trading.

Data & Methodology

Analysis conducted on 87 liquid futures contracts globally using 5 years of tick data (2012-2017), employing random forest machine learning algorithms to evaluate predictive power of market microstructure measures.

Key Features Analyzed

Study examines traditional microstructure measures including Roll measure, Kyle's lambda, Amihud measure, VPIN (volume-synchronized probability of informed trading), and VIX.

Predictive Power of Microstructure Measures Across Features

  • Shows relative importance of different microstructure measures in predicting market behavior
  • Amihud measure and VPIN demonstrate highest predictive power
  • Traditional measures like Roll and Kyle's lambda show lower importance

Prediction Accuracy Across Different Market Variables

  • Shows prediction accuracy for different market variables using machine learning
  • Return variance shows highest predictability at 56.63%
  • Bid-ask spread shows lowest predictability at 45.39%

Impact of Cross-Asset Features on Prediction Accuracy

  • Compares prediction accuracy with and without cross-asset features
  • Including cross-asset features improves prediction accuracy for most variables
  • Greatest improvement seen in return variance predictions

Forecast Horizon Impact on Prediction Accuracy

  • Shows how prediction accuracy varies with different forecast horizons
  • Shorter forecast horizons (25-50 bars) generally show better accuracy
  • Accuracy decreases slightly with longer forecast horizons

Contribution and Implications

  • Demonstrates that traditional microstructure measures remain relevant in modern markets, but their relative importance has shifted
  • Shows that cross-asset relationships are crucial for market predictions, suggesting systemic influences in market behavior
  • Provides evidence that machine learning can improve market prediction accuracy beyond traditional statistical methods

Data Sources

  • Feature Importance Chart: Based on Table 2, Panel A showing MDI feature importance
  • Prediction Accuracy Chart: Based on Table 8 comparing dollar-volume and time bars
  • Cross-Asset Impact Chart: Based on Table 14 comparing prediction accuracy with cross-asset features
  • Forecast Horizon Chart: Based on Table 12 showing accuracy across different forecast horizons
  • Cross-Asset Features Chart: Based on analysis of results presented in Section 5 and Table 11