Learning Multiple Stock Trading Patterns with Temporal Routing Adaptor and Optimal Transport (2106.12950v2)

Abstract: Successful quantitative investment usually relies on precise predictions of the future movement of the stock price. Recently, machine learning based solutions have shown their capacity to give more accurate stock prediction and become indispensable components in modern quantitative investment systems. However, the i.i.d. assumption behind existing methods is inconsistent with the existence of diverse trading patterns in the stock market, which inevitably limits their ability to achieve better stock prediction performance. In this paper, we propose a novel architecture, Temporal Routing Adaptor (TRA), to empower existing stock prediction models with the ability to model multiple stock trading patterns. Essentially, TRA is a lightweight module that consists of a set of independent predictors for learning multiple patterns as well as a router to dispatch samples to different predictors. Nevertheless, the lack of explicit pattern identifiers makes it quite challenging to train an effective TRA-based model. To tackle this challenge, we further design a learning algorithm based on Optimal Transport (OT) to obtain the optimal sample to predictor assignment and effectively optimize the router with such assignment through an auxiliary loss term. Experiments on the real-world stock ranking task show that compared to the state-of-the-art baselines, e.g., Attention LSTM and Transformer, the proposed method can improve information coefficient (IC) from 0.053 to 0.059 and 0.051 to 0.056 respectively. Our dataset and code used in this work are publicly available: https://github.com/microsoft/qlib/tree/main/examples/benchmarks/TRA.

• The paper presents TRA, an extension module that integrates multiple independent predictors to capture diverse trading patterns in stock markets.
• It employs an optimal transport formulation to dynamically assign samples to predictors, preventing trivial dominance and ensuring balanced learning.
• Empirical evaluations show that TRA significantly improves metrics like the information coefficient, annualized returns, and Sharpe ratios over benchmark models.

Learning Multiple Stock Trading Patterns with Temporal Routing Adaptor and Optimal Transport

The paper presents a novel framework aiming to enhance stock prediction by acknowledging the complexity and diversity of trading patterns in financial markets. Traditional stock prediction models typically rely on the i.i.d assumption, which may not accurately capture the multiple trading strategies investors employ. This paper introduces the Temporal Routing Adaptor (TRA), a significant advancement in the field of machine learning-based quantitative finance, designed to tackle these limitations.

Core Contributions

The proposed TRA architecture is an extension module that integrates seamlessly with existing stock prediction models to effectively manage and predict using multiple trading patterns. This is achieved through two main components: a collection of independent predictors and a sophisticated routing mechanism. The predictors cater to modeling diverse patterns, while the router dynamically assigns samples to these predictors to optimize prediction accuracy.

The research addresses a common problem in stock prediction—multiple observed phenomena often contradict each other, such as momentum and reversal effects. By modeling different patterns independently, TRA provides a more nuanced understanding of stock market movements.

Technical Framework

TRA leverages a unique learning algorithm based on Optimal Transport (OT) to facilitate optimal assignment of samples to predictors. The OT formulation is critical for ensuring that the sample distribution across predictors matches the underlying pattern distribution, thus improving prediction performance.

This innovative approach helps avoid trivial solutions where one predictor dominates, a common pitfall in similar architectures. The inclusion of an auxiliary regularization term derived from OT solutions aids in stabilizing the training phase and enhancing pattern discovery.

Empirical Evaluation

The authors conduct extensive experiments using real-world stock market data to benchmark TRA against state-of-the-art models like Attention LSTM and Transformer. Notably, TRA shows a marked improvement in information coefficient (IC), with achievable gains from 0.053 to 0.059 for Attention LSTM and 0.051 to 0.056 for Transformer. Such results underscore TRA's effectiveness in handling stock prediction tasks by accounting for multiple trading patterns.

Additionally, the research evaluates portfolio performance metrics, such as annualized returns and Sharpe ratios. Here, TRA also outperforms, further validating its applicability to real-world investment strategies.

Implications and Future Directions

The methodological advancements presented in this paper have several implications for the development of AI systems in finance. By moving beyond the classical single-pattern approach, the TRA framework sets a precedent for future systems aimed at capturing the inherent multi-faceted nature of financial data. This work opens avenues for integrating more sophisticated pattern recognition and allocation mechanisms, which could lead to even more robust stock prediction models.

Future developments could explore real-time adaptation to emerging patterns, bridging the gap between offline learning and online market dynamics. Additionally, expanding the framework to handle proprietary data sources could yield further insights and improvements for developed markets, where public information is less predictive.

In conclusion, the paper provides a comprehensive exploration of multi-pattern learning in stock prediction, offering a significant contribution to both the theoretical foundations and practical implementations in financial AI systems.