Paper: arXiv 2406.16505

Authors: Feng Xu, Yan Yin, Xinyu Zhang, Tianyuan Liu, Shengyi Jiang, Zongzhang Zhang

Abstract

Alphas are pivotal in providing signals for quantitative trading. The industry highly values the discovery of formulaic alphas for their interpretability and ease of analysis, compared with the expressive yet overfitting-prone black-box alphas. In this work, we focus on discovering formulaic alphas. Prior studies on automatically generating a collection of formulaic alphas were mostly based on genetic programming (GP), which is known to suffer from the problems of being sensitive to the initial population, converting to local optima, and slow computation speed. Recent efforts employing deep reinforcement learning (DRL) for alpha discovery have not fully addressed key practical considerations such as alpha correlations and validity, which are crucial for their effectiveness. In this work, we propose a novel framework for alpha discovery using DRL by formulating the alpha discovery process as program construction. Our agent, \(\text{“Alpha”}^2\), assembles an alpha program optimized for an evaluation metric. A search algorithm guided by DRL navigates through the search space based on value estimates for potential alpha outcomes. The evaluation metric encourages both the performance and the diversity of alphas for a better final trading strategy. Our formulation of searching alphas also brings the advantage of pre-calculation dimensional analysis, ensuring the logical soundness of alphas, and pruning the vast search space to a large extent. Empirical experiments on real-world stock markets demonstrates \(\text{“Alpha”}^2\)’s capability to identify a diverse set of logical and effective alphas, which significantly improves the performance of the final trading strategy. The code of our method is available at https://github.com/x35f/alpha2.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)

Why this score: The paper introduces advanced mathematical concepts like program construction and dimensional analysis for alpha discovery, while also demonstrating high empirical rigor through real-world stock market experiments and a publicly available codebase for backtesting.

Research Flowchart

  flowchart TD
  Goal["Research Goal<br>Discover Interpretability &<br>Effective Formulaic Alphas"] --> Input["Input Data<br>Historical Market Data"]
  Input --> Method["Methodology: DRL Agent Alpha²<br>Formulates Alpha Discovery as Program Construction"]
  Method --> Search["Search Process<br>DRL-Guided Search Space Navigation<br>+ Dimensional Analysis & Pruning"]
  Search --> Eval["Evaluation<br>Metrics for Performance & Diversity"]
  Eval --> Outcomes["Outcomes<br>Logical, Diverse Alphas<br>+ Improved Trading Strategy"]
  Outcomes --> Comparison["Result: Outperforms<br>Genetic Programming Baselines"]
  
  style Goal fill:#f9f,stroke:#333,stroke-width:2px
  style Outcomes fill:#bbf,stroke:#333,stroke-width:2px

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes Alpha2, a novel deep‑reinforcement‑learning framework that discovers formulaic alphas by treating alpha generation as a program‑construction task.

Source sentence
In this work, we propose a novel framework for alpha discovery using DRL by formulating the alpha discovery process as program construction.

Assumes 6

  • The study assumes a day‑frequency trading strategy over a stock market with n distinct stocks and T trading days. market
    Source sentence
    In this study, we focus on finding a day-frequency trading strategy in a stock market consisting of n distinct stocks spanning T trading days.
  • Alphas are evaluated primarily by Information Correlation (IC) with the 20‑day return. statistical
    Source sentence
    method is to find alphas that have high IC with the 20-day return of the stocks.
  • The data is split into training (2009/01/01‑2018/12/31), validation (2019/01/01‑2020/12/31) and test (2021/01/01‑2023/12/31) periods. data
    Source sentence
    The dataset is split into a training set (2009/01/01 to 2018/12/31), a validation set (2019/01/01 to 2020/12/31), and a test set (2021/01/01 to 2023/12/31).
  • The universe of stocks is limited to constituents of the CSI300 and CSI500 indices of China A‑shares. market
    Source sentence
    We use the constituents of the CSI300 and CSI500 indices of China A-shares as the stock set.
Show 2 more
  • Alpha combination is performed by training an XGBoost model on the top 20 generated alphas ranked by IC. model
    Source sentence
    The XGBoost model is trained to fit the alpha signals to the return signals on the training dataset, using the top 20 generated alphas ranked by IC.
  • The back‑test uses a top‑k/drop‑n strategy with k=50 and at most n=5 stocks traded per day, investing evenly across selected stocks. execution
    Source sentence
    On each trading day, stocks are first sorted according to the alpha values, then the top‑k stocks are selected to trade. With at most n stocks traded every day, we try to invest evenly across the k stocks. In out experiment, k = 50 and n = 5.

Data

Backtest sample: 2021-01 – 2023-12 · Chinese A-shares (CSI300, CSI500) · equities · daily

  • Chinese A‑shares market via baostock · baostock market-prices public
    2009‑2023 · CSI300 and CSI500 constituents
    Source sentence
    ... The dataset is split into a training set (2009/01/01 to 2018/12/31), a validation set (2019/01/01 to 2020/12/31), and a test set (2021/01/01 to 2023/12/31). We use the constituents of the CSI300 and CSI500 indices of China A-shares as the stock set.

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

  • Alpha2 identifies a diverse set of logical and effective alphas that significantly improve the performance of the final trading strategy. performance
    Source sentence
    Empirical experiments on real-world stock markets demonstrates Alpha2’s capability to identify a diverse set of logical and effective alphas, which significantly improves the performance of the final trading strategy.
  • Alpha2 shows superior back‑test performance on the CSI300 stock market compared to baselines. performance
    Source sentence
    Our method demonstrates superior performance on the CSI300 stock market.
  • Alpha2 generates alphas with lower mean correlation than gplearn, leading to a more robust trading strategy. performance
    Source sentence
    With the more diverse and better alpha set, Alpha2 has greater potential to generate a more robust trading strategy.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (2 by independent authors)

Code

https://github.com/x35f/alpha2 — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (22)

  1. Generating Synergistic Formulaic Alpha Collections via Reinforcement Learning — → 2306.12964
  2. Generating Synergistic Formulaic Alpha Collections via Reinforcement Learning — → 2306.12964
  3. Auto A lpha: An efficient hierarchical evolutionary algorithm for mining alpha factors in quantitative investment (2020) — link
  4. 101 formulaic alphas (2016)
  5. Faster sorting algorithms discovered using deep reinforcement learning (2023)
  6. Mastering the game of G o without human knowledge (2017) — link
  7. Mastering the game of go without human knowledge (2017)
  8. Mastering the game of Go with deep neural networks and tree search (2016) — link
and 14 more
  1. Mastering the game of G o with deep neural networks and tree search (2016)
  2. Ray: A Distributed Framework for Emerging AI Applications (2018) — link
  3. Practical Massively Parallel M onte- C arlo Tree Search Applied to Molecular Design (2021) — link
  4. Light GBM : A Highly Efficient Gradient Boosting Decision Tree (2017)
  5. Alpha E volve: A learning framework to discover novel alphas in quantitative investment (2021)
  6. Generalized mean estimation in M onte- C arlo tree search (2019) — link
  7. Deep generative symbolic regression with M onte- C arlo-tree-search (2023)
  8. Learning equations for extrapolation and control (2018)
  9. Genetic P rogramming as a means for programming computers by natural selection (1994)
  10. Symbolic regression via neural-guided genetic programming population seeding (2021) — link
  11. Finding Alphas: A Quantitative Approach to Building Trading Strategies (2019)
  12. Multi-armed bandits with episode context (2011)
  13. XGB oost: A scalable tree boosting system (2016)
  14. Qlib: An AI -oriented quantitative investment platform (2020) — link

2 of 22 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.