Main Content

Machine Learning for Statistical Arbitrage: Introduction

Heatmap of trading matrix

Machine learning techniques for processing large amounts of data are broadly applicable in computational finance. The series of examples introduced in this topic provides a general workflow, illustrating how capabilities in MATLAB® apply to a specific problem in financial engineering. The workflow is problem-oriented, exploratory, and guided by the data and the resulting analysis. The overall approach, however, is useful for constructing applications in many areas.

The workflow consists of these actions:

  • Formulate a simple approach to algorithmic trading, through an analysis of market microstructure, with the goal of identifying real-time arbitrage opportunities.

  • Use a large sample of exchange data to track order dynamics of a single security on a single day, selectively processing the data to develop relevant statistical measures.

  • Create a model of intraday dynamics conditioned on a selection of hyperparameters introduced during feature engineering and development.

  • Evaluate hyperparameter tunings using a supervising objective that computes cash returned on a model-based trading strategy.

  • Optimize the trading strategy using different machine learning algorithms.

  • Suggest modifications for further development.

The workflow is separated into three examples:

  1. Machine Learning for Statistical Arbitrage I: Data Management and Visualization

  2. Machine Learning for Statistical Arbitrage II: Feature Engineering and Model Development

  3. Machine Learning for Statistical Arbitrage III: Training, Tuning, and Prediction

For more information about general workflows for machine learning, see: