AlphaForge: A Framework to Mine and Dynamically Combine Formulaic Alpha Factors

Hao Shi,Weili Song,Xinting Zhang,Jiahe Shi,Cuicui Luo,Xiang Ao,Hamid Arian,Luis Seco

2024-08-28

Abstract:The complexity of financial data, characterized by its variability and low signal-to-noise ratio, necessitates advanced methods in quantitative investment that prioritize both performance and interpretability.Transitioning from early manual extraction to genetic programming, the most advanced approach in the alpha factor mining domain currently employs reinforcement learning to mine a set of combination factors with fixed weights. However, the performance of resultant alpha factors exhibits inconsistency, and the inflexibility of fixed factor weights proves insufficient in adapting to the dynamic nature of financial markets. To address this issue, this paper proposes a two-stage formulaic alpha generating framework AlphaForge, for alpha factor mining and factor combination. This framework employs a generative-predictive neural network to generate factors, leveraging the robust spatial exploration capabilities inherent in deep learning while concurrently preserving diversity. The combination model within the framework incorporates the temporal performance of factors for selection and dynamically adjusts the weights assigned to each component alpha factor. Experiments conducted on real-world datasets demonstrate that our proposed model outperforms contemporary benchmarks in formulaic alpha factor mining. Furthermore, our model exhibits a notable enhancement in portfolio returns within the realm of quantitative investment and real money investment.

Computational Finance,Artificial Intelligence

What problem does this paper attempt to address?

### The Problem the Paper Attempts to Solve

The paper aims to address the complexity of financial data, particularly its variability and low signal-to-noise ratio characteristics, which necessitate advanced methods in the field of quantitative investment to balance performance and interpretability. The current state-of-the-art alpha factor mining methods employ reinforcement learning techniques to generate a set of composite factors with fixed weights. However, this approach has the following issues:

1. **Inconsistent Performance**: The generated alpha factors exhibit unstable performance under different market conditions.
2. **Fixed Weights**: The fixed-weight factor combinations cannot adapt to the dynamic nature of financial markets.

To solve these problems, the paper proposes a two-stage formalized alpha generation framework—AlphaForge, for alpha factor mining and combination. Specifically:

- **Factor Mining Model**: Utilizes a generate-predict neural network to generate factors, leveraging the powerful spatial exploration capabilities of deep learning while maintaining factor diversity.
- **Factor Combination Model**: Selects factors based on their temporal performance and dynamically adjusts the weights of each component alpha factor to adapt to market changes.

Experimental validation shows that this framework outperforms existing benchmark methods on real-world datasets, achieving significant improvements in formalized alpha factor mining and demonstrating higher investment returns in quantitative investment and actual fund investment.