Paper: arXiv 2502.05218

Authors: Yitong Duan, Weiran Wang, Jian Li

Abstract

As a fundamental method in economics and finance, the factor model has been extensively utilized in quantitative investment. In recent years, there has been a paradigm shift from traditional linear models with expert-designed factors to more flexible nonlinear machine learning-based models with data-driven factors, aiming to enhance the effectiveness of these factor models. However, due to the low signal-to-noise ratio in market data, mining effective factors in data-driven models remains challenging. In this work, we propose a hypergraph-based factor model with temporal residual contrastive learning (FactorGCL) that employs a hypergraph structure to better capture high-order nonlinear relationships among stock returns and factors. To mine hidden factors that supplement human-designed prior factors for predicting stock returns, we design a cascading residual hypergraph architecture, in which the hidden factors are extracted from the residual information after removing the influence of prior factors. Additionally, we propose a temporal residual contrastive learning method to guide the extraction of effective and comprehensive hidden factors by contrasting stock-specific residual information over different time periods. Our extensive experiments on real stock market data demonstrate that FactorGCL not only outperforms existing state-of-the-art methods but also mines effective hidden factors for predicting stock returns.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs only partially 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 involves advanced mathematics including hypergraph convolutional neural networks (HyperGCN) with detailed matrix operations, non-linear activations, and a novel contrastive learning framework, making it highly mathematically dense. It also demonstrates strong empirical rigor with extensive experiments on real stock market data, comparing against state-of-the-art methods and reporting performance improvements, indicating backtest-ready implementation.

Research Flowchart

  flowchart TD
  A["Research Goal: Enhance Stock Prediction<br>via Data-Driven Factor Mining"] --> B["Input Data: Historical<br>Stock Returns & Prior Factors"]
  B --> C["Cascading Residual Hypergraph<br>Removes Prior Factor Effects"]
  C --> D["Temporal Residual Contrastive Learning<br>Extracts Hidden Factors from Residuals"]
  D --> E["Hypergraph Integration<br>Captures High-Order Nonlinear Relationships"]
  E --> F["Stock Returns Prediction Model"]
  F --> G["Key Findings: State-of-the-art Performance<br>& Effective Hidden Factors"]

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 FactorGCL, a hypergraph-based factor model with temporal residual contrastive learning that mines hidden factors to improve stock return prediction.

Source sentence
we propose a hypergraph-based factor model with temporal residual contrastive learning (FactorGCL) that employs a hypergraph structure to better capture high-order nonlinear relationships among stock returns and factors.

Assumes 3

  • The dataset is split temporally into training, validation, and test sets with a 5‑year, 1‑year, 2‑year split. data
    Source sentence
    We follow the temporal order to split the dataset into training set, validation set and test set, where the time length is 5 years : 1 year : 2 years
  • Secondary industry factors (83 industries) are used as prior factors with binary exposures. model
    Source sentence
    We adopt secondary industry factors as the prior factors (83 industries, if a stock belongs to the industry, the value of corresponding factor exposure is 1, otherwise 0).
  • Transaction costs are set to 0.3% in the investment simulation. execution
    Source sentence
    set TopK = 30, ∆t = 10, and the transaction cost to 0.3%.

Data

Backtest sample: 2020-01 – 2023-06 · CSI 300 and CSI 500 stocks · equities · daily

  • China A‑shares market dataset market-prices unknown
    01/01/2014 to 06/30/2023 · 5028 China A‑shares stocks
    Source sentence
    We conduct our experiments on the China A‑shares market, utilizing a dataset spanning from 01/01/2014 to 06/30/2023. This dataset includes 5028 stocks, excluding suspended or otherwise abnormal stocks, and is constructed from a sequence of market data comprising day‑level price‑volume data
  • CSI300 benchmark portfolio benchmark unknown
    Source sentence
    We use the equally weighted CSI300 and CSI500 portfolio as the benchmark
  • CSI500 benchmark portfolio benchmark unknown
    Source sentence
    We use the equally weighted CSI300 and CSI500 portfolio as the benchmark

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Partial

Paper text extraction: Modeled - transaction cost set to 0.3% per trade in the back‑test

Source sentence (the authors' words)
set TopK = 30, ∆t = 10, and the transaction cost to 0.3%.

Finds 3

  • FactorGCL achieves the highest information coefficient (IC) and information‑ratio (ICIR) among all compared methods on the test dataset. predictive-accuracy
    Source sentence
    Our method achieves the highest IC and ICIR among all the compared methods.
  • In investment simulations, FactorGCL outperforms all baselines across annualized return, information ratio, and return‑over‑maximum‑drawdown metrics. performance
    Source sentence
    The investment simulation results show that our method achieves the best performance across all metrics, indicating that our model can achieve profitable investments in the real market.
  • Ablation studies confirm that removing any module (prior beta, hidden beta, etc.) degrades performance, validating the contribution of each component. statistical
    Source sentence
    To verify the effectiveness of different modules in our framework, we build four variants of proposed model by removing the prior beta module, hid…

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 21 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (3 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.