Paper: arXiv 2402.07080

Authors: Tao Ren, Ruihan Zhou, Jinyang Jiang, Jiafeng Liang, Qinghao Wang, Yijie Peng

Abstract

The formulaic alphas are mathematical formulas that transform raw stock data into indicated signals. In the industry, a collection of formulaic alphas is combined to enhance modeling accuracy. Existing alpha mining only employs the neural network agent, unable to utilize the structural information of the solution space. Moreover, they didn’t consider the correlation between alphas in the collection, which limits the synergistic performance. To address these problems, we propose a novel alpha mining framework, which formulates the alpha mining problems as a reward-dense Markov Decision Process (MDP) and solves the MDP by the risk-seeking Monte Carlo Tree Search (MCTS). The MCTS-based agent fully exploits the structural information of discrete solution space and the risk-seeking policy explicitly optimizes the best-case performance rather than average outcomes. Comprehensive experiments are conducted to demonstrate the efficiency of our framework. Our method outperforms all state-of-the-art benchmarks on two real-world stock sets under various metrics. Backtest experiments show that our alphas achieve the most profitable results under a realistic trading setting.

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 employs advanced concepts like MCTS, reward-dense MDPs, and risk-seeking policy optimization, indicating high math complexity, while demonstrating empirical rigor through comprehensive experiments on real-world datasets, backtests, and ablation studies.

Research Flowchart

  flowchart TD
  A["Research Goal: Discover formulaic alphas<br>that generate high returns<br>via Risk-Seeking MCTS"] --> B["Data Input:<br>Raw Stock Data<br>Preprocessing"]
  B --> C["Core Methodology:<br>Reward-Dense MDP Formulation"]
  C --> D["MCTS Agent:<br>Risk-Seeking Policy<br>Optimizes Best-Case Scenarios"]
  D --> E["Alpha Generation:<br>Formulaic Alphas<br>discovered by MCTS"]
  E --> F["Experiments & Backtest:<br>Performance Evaluation"]
  F --> G["Key Findings:<br>Outperforms SOTA benchmarks<br>Most profitable in trading settings"]

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 RiskMiner, a novel alpha mining framework that combines a reward‑dense MDP, Monte Carlo Tree Search and a risk‑seeking policy to discover synergistic formulaic alphas.

Source sentence
To address these problems, we propose a novel alpha mining framework, which formulates the alpha mining problems as a reward-dense Markov Decision Process (MDP) and solves the MDP by the risk-seeking Monte Carlo Tree Search (MCTS).

Assumes 5

  • The backtest uses a simple and widely used trading strategy under a realistic trading setting. execution
    Source sentence
    Employing a simple and widely used trading strategy, our methods achieve the most profitable outcomes among all compared methods.
  • Alphas are evaluated on daily price and volume features (open, high, low, close, volume, vwap) from the A‑share market. data
    Source sentence
    Six features on stock price and volume are used in our experiments: open, high, low, close, volume, vwap(volume‑weighted average price).
  • The reward‑dense MDP provides intermediate and terminal rewards to stabilize learning. model
    Source sentence
    Our approach involves a reward‑dense MDP, which is structured with both intermediate and terminal rewards, thereby providing more frequent and informative feedback.
  • The discrete solution space of formulaic alphas is huge, making MCTS appropriate for exploration. model
    Source sentence
    the discrete solution space for the formulas is huge and PPO is not the optimal method to
Show 1 more
  • The market period used for training, validation and testing is 2010‑2022, split as specified. data
    Source sentence
    The dataset contains 13 years of daily data and is split into a training set (2010/01/01 to 2019/12/31), a validation set (2020/01/01 to 2020/12/31), a test set (2021/01/01 to 2022/12/31).

Data

  • China A‑share market market-prices unknown
    13 years of daily data (2010‑2022) · CSI300 and CSI500 constituent stocks
    Source sentence
    Two popular stock sets are used in the experiments: constituent stocks on CSI300 index and CSI500 index.

Costs

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

Finds 3

  • RiskMiner outperforms all state‑of‑the‑art benchmarks on two real‑world stock sets across every reported metric. performance
    Source sentence
    Our method outperforms all state‑of‑the‑art benchmarks on two real‑world stock sets under various metrics.
  • In realistic backtests, RiskMiner achieves the most profitable results among all compared methods. performance
    Source sentence
    Backtest experiments show that our alphas achieve the most profitable results under a realistic trading setting.
  • Ablation study indicates that MCTS contributes significantly to overall performance, with the risk‑seeking policy providing additional gains. performance
    Source sentence
    The MCTS has a slight advantage over the Risk‑seeking policy. We can infer that MCTS contributes significantly to the overall performance. By integrating with a risk‑seeking policy, our methods can elevate performance to a new tier.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 18 related papers). Coverage is partial, so this is neither replication nor refutation.

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

Code

No repository by the authors found in the paper.

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

Cited References (36)

  1. Capital asset prices: A theory of market equilibrium under conditions of risk (1964)
  2. A five-factor asset pricing model (2015)
  3. Generating Synergistic Formulaic Alpha Collections via Reinforcement Learning — → 2306.12964
  4. Finding Alphas: A Quantitative Approach to Building Trading Strategies (2015)
  5. 101 Formulaic Alphas (2016) — link
  6. Qlib: An AI-oriented Quantitative Investment Platform (2020) — link
  7. AlphaEvolve: A Learning Framework to Discover Novel Alphas in Quantitative Investment (2021)
  8. AutoAlpha: an Efficient Hierarchical Evolutionary Algorithm for Mining Alpha Factors in Quantitative Investment (2020) — link
and 28 more
  1. Stock Alpha Mining Based On Genetic Algorithm (2019) — link
  2. Revisiting Stock Alpha Mining Based On Genetic Algorithm (2019) — link
  3. Factorvae: A probabilistic dynamic factor model based on variational autoencoder for predicting cross-sectional stock returns (2022)
  4. Adaptive long-short pattern transformer for stock investment selection (2022)
  5. Hierarchical Multi-Scale Gaussian Transformer for Stock Movement Prediction. (2020)
  6. Attention is all you need (2017)
  7. Stock Price Prediction via Discovering Multi-Frequency Trading Patterns (2017)
  8. HIST: A Graph-based Framework for Stock Trend Forecasting via Mining Concept-Oriented Shared Information (2021) — link
  9. Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction — → 2309.00073
  10. DoubleAdapt: A Meta-learning Approach to Incremental Learning for Stock Trend Forecasting — → 2306.09862
  11. Spotlight news driven quantitative trading based on trajectory optimization (2023)
  12. Knowledge-driven stock trend prediction and explanation via temporal convolutional network (2019)
  13. Listening to chaotic whispers: A deep learning framework for news-oriented stock trend prediction (2018)
  14. FinGPT: Open-Source Financial Large Language Models — → 2306.06031
  15. FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design — → 2311.13743
  16. Mastering atari, go, chess and shogi by planning with a learned model (2020)
  17. A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play (2018)
  18. Mastering atari, go, chess and shogi by planning with a learned model (2020)
  19. Discovering faster matrix multiplication algorithms with reinforcement learning (2022)
  20. Faster sorting algorithms discovered using deep reinforcement learning (2023)
  21. Proximal Policy Optimization Algorithms (2017) — link
  22. Towards human-level bimanual dexterous manipulation with reinforcement learning (2022)
  23. FinRL: A deep reinforcement learning library for automated stock trading in quantitative finance (2020) — link
  24. Cumulative prospect theory meets reinforcement learning: Prediction and control (2016)
  25. A stochastic approximation method for simulation-based quantile optimization (2022)
  26. XGBoost: A Scalable Tree Boosting System (2016)
  27. LightGBM: A Highly Efficient Gradient Boosting Decision Tree (2017)
  28. CatBoost: unbiased boosting with categorical features (2018)

5 of 36 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.