End-to-End Autoencoder Policy Learning for Statistical Arbitrage

Summary

This study explores autoencoders for statistical arbitrage in US stocks. In a conventional two-stage workflow, a pricing or principal-component model identifies a synthetic asset and a separate mean-reversion strategy produces trading signals. The authors first train a standard autoencoder on stock returns and build strategies around an Ornstein-Uhlenbeck process. They then embed the autoencoder in a neural representation of portfolio policies that produces allocations directly.

The combined model is trained end to end by backpropagating risk-adjusted returns, connecting representation learning to portfolio decisions. The reported findings are that this approach simplifies strategy development and achieves higher gross returns than the compared alternatives. The abstract does not provide the evaluation period, risk-adjusted or net-of-cost results, comparison details, or evidence of performance outside the study setting. Its reported advantage is therefore a research finding, not a guarantee of live profitability.

Key ideas

• The paper applies autoencoders trained on US stock returns to statistical arbitrage.
• A baseline approach uses an autoencoder to identify a synthetic asset and an Ornstein-Uhlenbeck mean-reversion strategy.
• The end-to-end architecture embeds the autoencoder in a network that outputs portfolio allocations directly.
• Training uses backpropagation of risk-adjusted portfolio returns.
• The authors report higher gross returns than competitors, while the abstract does not establish net or live performance.

Tags

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


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









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.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.