Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

End-to-End Policy Learning of a Statistical Arbitrage Autoencoder Architecture

Fabian Krause, Jan-Peter Calliess

TL;DR

This work generalises statistical arbitrage by replacing traditional PCA-based factor models with Autoencoder-derived residuals and embedding them in an end-to-end neural trading policy. It systematically compares Autoencoder residual generation against Fama-French and PCA baselines and introduces an end-to-end AE-StatArb policy that optimizes both the residual representation and risk-adjusted returns. Across US equity data (2000–2022), Autoencoder approaches achieve competitive pre-cost performance, with the learned policy showing superior risk-adjusted returns and aligning with the literature on the optimal number of latent factors (roughly 10–15). The study highlights the potential of integrated, end-to-end training to reduce modeling risk and streamline StatArb pipeline design, while outlining concrete avenues for improving after-cost performance and intraday applicability.

Abstract

In Statistical Arbitrage (StatArb), classical mean reversion trading strategies typically hinge on asset-pricing or PCA based models to identify the mean of a synthetic asset. Once such a (linear) model is identified, a separate mean reversion strategy is then devised to generate a trading signal. With a view of generalising such an approach and turning it truly data-driven, we study the utility of Autoencoder architectures in StatArb. As a first approach, we employ a standard Autoencoder trained on US stock returns to derive trading strategies based on the Ornstein-Uhlenbeck (OU) process. To further enhance this model, we take a policy-learning approach and embed the Autoencoder network into a neural network representation of a space of portfolio trading policies. This integration outputs portfolio allocations directly and is end-to-end trainable by backpropagation of the risk-adjusted returns of the neural policy. Our findings demonstrate that this innovative end-to-end policy learning approach not only simplifies the strategy development process, but also yields superior gross returns over its competitors illustrating the potential of end-to-end training over classical two-stage approaches.

End-to-End Policy Learning of a Statistical Arbitrage Autoencoder Architecture

TL;DR

This work generalises statistical arbitrage by replacing traditional PCA-based factor models with Autoencoder-derived residuals and embedding them in an end-to-end neural trading policy. It systematically compares Autoencoder residual generation against Fama-French and PCA baselines and introduces an end-to-end AE-StatArb policy that optimizes both the residual representation and risk-adjusted returns. Across US equity data (2000–2022), Autoencoder approaches achieve competitive pre-cost performance, with the learned policy showing superior risk-adjusted returns and aligning with the literature on the optimal number of latent factors (roughly 10–15). The study highlights the potential of integrated, end-to-end training to reduce modeling risk and streamline StatArb pipeline design, while outlining concrete avenues for improving after-cost performance and intraday applicability.

Abstract

In Statistical Arbitrage (StatArb), classical mean reversion trading strategies typically hinge on asset-pricing or PCA based models to identify the mean of a synthetic asset. Once such a (linear) model is identified, a separate mean reversion strategy is then devised to generate a trading signal. With a view of generalising such an approach and turning it truly data-driven, we study the utility of Autoencoder architectures in StatArb. As a first approach, we employ a standard Autoencoder trained on US stock returns to derive trading strategies based on the Ornstein-Uhlenbeck (OU) process. To further enhance this model, we take a policy-learning approach and embed the Autoencoder network into a neural network representation of a space of portfolio trading policies. This integration outputs portfolio allocations directly and is end-to-end trainable by backpropagation of the risk-adjusted returns of the neural policy. Our findings demonstrate that this innovative end-to-end policy learning approach not only simplifies the strategy development process, but also yields superior gross returns over its competitors illustrating the potential of end-to-end training over classical two-stage approaches.
Paper Structure (21 sections, 18 equations, 1 figure, 5 tables)

This paper contains 21 sections, 18 equations, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Statistical Arbitrage
  4. Asset Pricing Models
  5. Arbitrage Signal Extraction
  6. Arbitrage Portfolio Construction
  7. Methodology
  8. Data
  9. Universe of Eligible Stocks
  10. Benchmarks for Asset Pricing Models
  11. Fama French
  12. Principal Component Analysis
  13. Autoencoder Residual Generation
  14. Benchmarks for Arbitrage Signal Extraction
  15. Ornstein Uhlenbeck
  16. ...and 6 more sections

Figures (1)

  • Figure 1: Performance comparison in compounded percentage returns from 2002 to end of 2022.