Paper: arXiv 2610.04959
Authors: Yanzheng Jin, Pengyang Shao, Yunshan Ma, Haowen Pan, Naixin Zhai, Chen-Hui Song, Fei Shen, Kenji Kawaguchi
Abstract
Formulaic alpha discovery seeks symbolic expressions that predict cross-sectional asset returns. In deployment, multiple formulas are combined into an alpha pool, where each formula is valued through the complementary information it contributes to joint predictive performance. While Reinforcement Learning and Generative Flow Networks have emerged as promising paradigms for generating formulaic alphas, existing frameworks face three related challenges. First, generating formulas individually leaves pool context and inter-formula complementarity outside the generative state. Second, formula-wise generation lacks a unified mechanism for preserving and revising structures at different levels. Third, pool-level rewards jointly reflect predictive performance and redundancy but cannot be differentiated directly through symbolic evaluation to train the generator. To overcome these challenges, we introduce AlphaPADI (Formulaic Alpha Discovery via Pool-Aware Hierarchical Discrete Diffusion), a novel framework built around three components: (1) grammar-constrained buffer initialization that constructs syntactically valid pool candidates, (2) pool-aware hierarchical diffusion that reconstructs complete pools at multiple structural scales under the current pool context, and (3) reward-guided pool refinement that evaluates joint predictive performance and inner diversity, updates the elite buffer, and trains the reverse model through reconstruction and preference learning. Empirical results on the Chinese and U.S. stock markets demonstrate that AlphaPADI outperforms the evaluated baselines in both predictive and portfolio performance, thereby validating pool-aware generation as an effective framework for automated alpha discovery.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; reported Sharpe ratio of 5.8423 is implausibly high for daily or lower-frequency data; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)
Why this score: This paper presents a highly novel approach to alpha discovery using diffusion models, addressing key limitations of prior work. It demonstrates strong empirical results on real-world datasets, indicating robust methodology. The mathematical underpinnings of hierarchical discrete diffusion and preference learning are advanced.
Research Flowchart
flowchart TD
A["Research Goal: Formulaic Alpha Discovery"] --> B{"Challenges in Alpha Discovery"};
B --> C["AlphaPADI Framework: Novel Solution"];
C -- Components --> C1["1. Grammar-Constrained Buffer Initialization"] & C2["2. Pool-Aware Hierarchical Diffusion"] & C3["3. Reward-Guided Pool Refinement"];
C1 & C2 & C3 --> D{"Data/Inputs: Chinese & U.S. Stock Markets"};
D --> E["Computational Processes: Reconstruction & Preference Learning"];
E --> F["Outcomes: Superior Predictive & Portfolio Performance"];
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 AlphaPADI, a framework for formulaic alpha discovery that treats the alpha pool as the common unit of generation, evaluation, and learning.
Source sentence
we introduce AlphaPADI (Formulaic Alpha Discovery via Pool-Aware Hierarchical Discrete Diffusion)
Assumes 3
- Pool combination is performed with a ridge combiner fitted once on the training period and then frozen for validation and testing. model
Source sentence
We evaluate every pool, including those returned by the baselines, with the ridge combiner in Eq. 3, fitted once on the training period and frozen during validation and testing.
- Data are split chronologically into training (2016‑01‑01–2020‑12‑31), validation (2021‑01‑01–2022‑12‑31), and testing (2023‑01‑01–2025‑12‑31). data
Source sentence
three universes share a chronological split: 2016-01-01–2020-12-31 for training, 2021-01-01–2022-12-31 for validation, and 2023-01-01–2025-12-31 for testing.
- Trading costs are not accounted for in the reported performance results. execution
Source sentence
Deploying a discovered strategy would require further assessment of execution costs and financial risk.
Data
2023-01 – 2025-12 · China and US equities · equities · daily
- CSI300 market-prices public
Source sentence
Datasets. We evaluate CSI300, CSI500, and S&P500, representing large‑cap Chinese, mid‑cap Chinese, and large‑cap U.S. equities, respectively.
- CSI500 market-prices public
Source sentence
Datasets. We evaluate CSI300, CSI500, and S&P500, representing large‑cap Chinese, mid‑cap Chinese, and large‑cap U.S. equities, respectively.
- S&P500 market-prices public
Source sentence
Datasets. We evaluate CSI300, CSI500, and S&P500, representing large‑cap Chinese, mid‑cap Chinese, and large‑cap U.S. equities, respectively.
Qlib datasets
Costs
Not modeled
Finds 2
Sharpe 5.84 · annual return 20.6% · max drawdown 8.6%
- AlphaPADI ranks first on every reported predictive metric while achieving the highest annualized return and Sharpe ratio, as well as the lowest maximum drawdown across all three datasets. performance
Source sentence
AlphaPADI ranks first on every reported predictive metric while achieving the highest annualized return and Sharpe ratio, as well as the lowest maximum drawdown, across all three datasets.
- AlphaPADI leads at both pool sizes (K=20 and K=50), maintaining superior predictive and portfolio performance compared to baselines. performance
Source sentence
AlphaPADI leads at both pool sizes. At both K = 20 and 50, AlphaPADI attai…
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.
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 (33)
- Qlib: An ai-oriented quantitative investment platform (2020) — link
- Empirical asset pricing via machine learning (2020)
- The unprecedented stock market reaction to COVID-19 (2020)
- Alphasage: Structure-aware alpha mining via gflownets for robust exploration — → 2509.25055
- Generating synergistic formulaic alpha collections via reinforcement learning — → 2306.12964
- Alphaqcm: Alpha discovery in finance with distributional reinforcement learning (2025)
- Alphaforge: A framework to mine and dynamically combine formulaic alpha factors — → 2406.18394
- Alphaagent: Llm-driven alpha mining with regularized exploration to counteract alpha decay (2025)
and 25 more
- Autoalpha: an efficient hierarchical evolutionary algorithm for mining alpha factors in quantitative investment (2020) — link
- Structured denoising diffusion models in discrete state-spaces (2021)
- Simple and effective masked diffusion language models (2024)
- Digress: Discrete denoising diffusion for graph generation (2022) — link
- Direct preference optimization: Your language model is secretly a reward model (2023)
- Diffusion model alignment using direct preference optimization (2024)
- Preference-based alignment of discrete diffusion models (2025) — link
- Discrete diffusion modeling by estimating the ratios of the data distribution (2023) — link
- Large language diffusion models (2026)
- Diffusion on syntax trees for program synthesis (2025)
- Training diffusion models with reinforcement learning (2024)
- Fine-tuning discrete diffusion models via reward optimization with applications to dna and protein design (2025)
- Llada 1.5: Variance-reduced preference optimization for large language diffusion models (2026)
- 101 formulaic alphas (2016)
- Alphaevolve: A learning framework to discover novel alphas in quantitative investment (2021)
- Genetic programming as a means for programming computers by natural selection (1994)
- QuantFactor REINFORCE: mining steady formulaic alpha factors with variance-bounded REINFORCE — → 2409.05144
- Navigating the alpha jungle: an LLM-Powered MCTS framework for formulaic factor mining (2025) — link
- Multilayer perceptrons for classification and regression (1991)
- Xgboost: A scalable tree boosting system (2016)
- Common risk factors in the returns on stocks and bonds (1993)
- Flow network based generative models for non-iterative diverse candidate generation (2021)
- Grammar variational autoencoder (2017)
- Ridge regression: Biased estimation for nonorthogonal problems (1970)
- An empirical study of financial factor mining based on gene expression programming (2021)
4 of 33 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2601 linked papers.