Attention Factors for Statistical Arbitrage
Researchers present an 'Attention Factor' model for equities trading that uses deep learning and attention mechanisms to spot profitable trades among similar stocks. Unlike older methods, it learns asset similarity and trading strategy together, optimizing for risk-adjusted returns after costs. The model achieves an out-of-sample Sharpe ratio above 4, and 2.3 after costs, an 84 percent improvement. It also reveals that weak factors matter for arbitrage, but requires heavy computation and struggles with sudden market shifts.
What it examines
This paper introduces an Attention Factor model that uses deep learning and attention mechanisms to find similar stocks, detect mispricing, and create trading strategies that maximize profits after trading costs. The approach jointly learns factors and trading policies, aiming to improve statistical arbitrage in U.S. equities.
What it concludes
The model achieves record-high Sharpe ratios, even after trading costs, outperforming previous methods. Its interpretable factors align with industry sectors. Applications include designing better trading strategies and risk management tools. Future work could explore other markets or improve adaptability to changing market conditions. Limitations include reliance on historical data.
Evidence objects
A new 'Attention Factor' model for equities trading uses deep learning and attention mechanisms to jointly learn asset similarity and trading strategies, optimizing risk-adjusted returns after transaction costs.
key_findings bullet 1 · key_findings · validation V0
The model achieves a remarkable out-of-sample Sharpe ratio above $4$, and $2.3$ after trading costsan $84%$ improvement over previous best models, setting a new benchmark for machine learning in finance.
key_findings bullet 2 · key_findings · validation V0
Leveraging 39 firm characteristics and a Long Convolution sequence model, it finds even weak factors crucial for arbitrage; however, it requires heavy computation and struggles with sudden market shifts like COVID-19.
key_findings bullet 3 · key_findings · validation V0
This paper presents the original 'Attention Factor Model,' uniquely integrating arbitrage factor discovery and trading allocation in a single deep learning framework. Leveraging attention mechanisms and sequence models, it achieves interpretable, high-performing results, with out-of-sample Sharpe ratios above 4 ($2.3$ net costs), significantly surpassing benchmarksdemonstrating compelling methodological and practical advances.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
Abstract: Statistical arbitrage exploits temporal price differences between similar assets. We develop a framework to jointly identify similar assets through factors, identify mispricing and form a trading policy that maximizes risk-adjusted performance after trading costs. Our Attention Factors are conditional latent factors that are the most useful for arbitrage trading. They are learned from firm charact… ▽ More Statistical arbitrage exploits temporal price differences between similar assets. We develop a framework to jointly identify similar assets through factors, identify mispricing and form a trading policy that maximizes risk-adjusted performance after trading costs. Our Attention Factors are conditional latent factors that are the most useful for arbitrage trading. They are learned from firm characteristic embeddings that allow for complex interactions. We identify time-series signals from the residual portfolios of our factors with a general sequence model. Estimating factors and the arbitrage trading strategy jointly is crucial to maximize profitability after trading costs. In a comprehensive empirical study we show that our Attention Factor model achieves an out-of-sample Sharpe ratio above 4 on the largest U.S. equities over a 24-year period. Our one-step solution yields an unprecedented Sharpe ratio of 2.3 net of transaction costs. We show that weak factors are important for arbitrage trading. △ Less
Source row: 262 · abstract type: unknown