Paper: arXiv 2306.12964

Authors: Shuo Yu, Hongyan Xue, Xiang Ao, Feiyang Pan, Jia He, Dandan Tu, Qing He

Abstract

In the field of quantitative trading, it is common practice to transform raw historical stock data into indicative signals for the market trend. Such signals are called alpha factors. Alphas in formula forms are more interpretable and thus favored by practitioners concerned with risk. In practice, a set of formulaic alphas is often used together for better modeling precision, so we need to find synergistic formulaic alpha sets that work well together. However, most traditional alpha generators mine alphas one by one separately, overlooking the fact that the alphas would be combined later. In this paper, we propose a new alpha-mining framework that prioritizes mining a synergistic set of alphas, i.e., it directly uses the performance of the downstream combination model to optimize the alpha generator. Our framework also leverages the strong exploratory capabilities of reinforcement learning~(RL) to better explore the vast search space of formulaic alphas. The contribution to the combination models’ performance is assigned to be the return used in the RL process, driving the alpha generator to find better alphas that improve upon the current set. Experimental evaluations on real-world stock market data demonstrate both the effectiveness and the efficiency of our framework for stock trend forecasting. The investment simulation results show that our framework is able to achieve higher returns compared to previous approaches.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced reinforcement learning and policy gradient algorithms for a complex search space (high math), and evaluates the framework on real-world stock data with investment simulations and comparative benchmarks (high rigor).

Research Flowchart

  flowchart TD
  A["<b>Research Goal:</b><br>Develop a synergistic alpha mining framework<br>using RL for quantitative trading"] --> B["<b>Data Input:</b><br>Historical Stock Market Data"]
  B --> C["<b>Methodology:</b><br>Reinforcement Learning Agent<br>explores vast alpha formula space"]
  C --> D["<b>Computational Process:</b><br>Generate & combine alpha sets,<br>evaluate via downstream model performance"]
  D --> E["<b>Key Finding 1:</b><br>RL framework finds synergistic<br>alpha sets effectively"]
  D --> F["<b>Key Finding 2:</b><br>Higher investment returns vs.<br>traditional single-alpha mining"]

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 a new reinforcement‑learning based alpha‑mining framework that directly optimizes the performance of a downstream combination model to generate synergistic formulaic alphas.

Source sentence
we propose a new alpha‑mining framework that prioritizes mining a synergistic set of alphas, i.e., it directly uses the performance of the downstream combination model to optimize the alpha generator.

Assumes 4

  • The market consists of a fixed set of n stocks over T trading days. market
    Source sentence
    We consider a stock market with 𝑛stocks in a period of 𝑇trading days in total.
  • Alphas are combined linearly with weights learned by a linear model. model
    Source sentence
    In this paper, we use a linear model to combine the alphas.
  • The target return is the 20‑day return calculated from closing prices, implying trades are executed at closing price without explicit slippage modeling. execution
    Source sentence
    The target is set to be the 20‑day return of the stocks, selling/buying at the closing price (Ref(close, −20)/close − 1).
  • All alpha and target vectors are normalized to mean 0 and length 1 before further computation. statistical
    Source sentence
    For the rest of this paper, we will assume that all the 𝑓(𝑋) evaluations and the targets 𝑦 are normalized to have a mean of 0 and a length of 1 before subsequent computations.

Data

  • Chinese A‑shares market raw data market-prices unknown
    2009/01/01‑2021/12/31 · CSI300 and CSI500 constituents
    Source sentence
    The dataset is split by date into a training set (2009/01/01 to 2018/12/31), a validation set (2019/01/01 to 2019/12/31), and a test set (2020/01/01 to 2021/12/31). In the following experiments, we will use the constituent stocks of the CSI300 and the CSI500 indices of China A‑shares as the stock set.

Recognised sources: BaoStock

Costs

Partial Costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers.

Source sentence
stocks on each day to reduce excessive trading costs.

Finds 3

  • The proposed RL‑based framework achieves higher returns than prior alpha‑generation approaches. performance
    Source sentence
    our framework is able to achieve higher returns compared to previous approaches.
  • Increasing the alpha pool size yields substantial synergy improvements for the proposed method, unlike baseline generators. performance
    Source sentence
    This leads to a substantial improvement when the pool size increases, meaning that our method can produce alpha sets with great synergy.
  • In back‑testing on the CSI300 test period, the framework generates the greatest profit among all compared methods. performance
    Source sentence
    Our framework is able to gain the most profit compared to the other methods.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 11 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 21 papers in the archive (20 by independent authors): Alpha Discovery via Grammar-Guided Learning and Search (used as a baseline); AllocFlow: End-to-End Portfolio Trading Action Generation … (used as a baseline); QuantFactor REINFORCE: Mining Steady Formulaic Alpha Factors … (used as a baseline); AlphaSAGE: Structure-Aware Alpha Mining via GFlowNets for … (used as a baseline); FactorBench: A Portfolio-Aware Benchmark for Automated … (uses its method or data)

Code

https://github.com/ICT-FinD-Lab/alphagen — 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.