InvariantStock: Learning Invariant Features for Mastering the Shifting Market

TL;DR

This study presents InvariantStock, a designed learning framework comprising two key modules: an environment-aware prediction module and an environment-agnostic module that can learn invariant features across different environments in a straightforward manner, thereby enhancing robustness against distribution shifts.

Abstract

Accurately predicting stock returns is crucial for effective portfolio management. However, existing methods often overlook a fundamental issue in the market, namely, distribution shifts, making them less practical for predicting future markets or newly listed stocks. This study introduces a novel approach to address this challenge by focusing on the acquisition of invariant features across various environments, thereby enhancing robustness against distribution shifts. Specifically, we present InvariantStock, a designed learning framework comprising two key modules: an environment-aware prediction module and an environment-agnostic module. Through the designed learning of these two modules, the proposed method can learn invariant features across different environments in a straightforward manner, significantly improving its ability to handle distribution shifts in diverse market settings. Our results demonstrate that the proposed InvariantStock not only delivers robust and accurate predictions but also outperforms existing baseline methods in both prediction tasks and backtesting within the dynamically changing markets of China and the United States.

Figures (8)

• Figure 1: The structural design of InvariantStock is delineated, where the green region symbolizes the feature selection module, comprising a mask model and a reconstruction model. The red and blue regions depict the environment-agnostic prediction module and the environment-aware prediction module, respectively. Each prediction module is composed of a state extractor, an encoder, a decoder, and a predictor.
• Figure 2: Comparative analysis of cumulative returns by different methods in the China stock market highlighting the superior performance of InvariantStock amidst market shifts.
• Figure 3: The cumulative returns from various methods in the US Stock Market indicate that with a strategically chosen number of stocks in the portfolio, approaches like InvariantStock, DoubleAdapt, and FactorVAE are capable of yielding improved results. However, these methods consistently encounter a higher maximum drawdown, suggesting an increased risk factor in their performance.
• Figure 4: Visualization of the mean value of the mask across all test data, indicating feature significance. Lighter colours represent features of higher prominence. The features are specified in Table \ref{['tab:features']}. The selection model demonstrates a preference for most fundamental features, while price-related features are generally disfavored.
• Figure 5: The historical trend of SSEC depicting the distribution shifting in China stock market.