Paper: arXiv 2305.08740

Authors: Sheng Xiang, Dawei Cheng, Chencheng Shang, Ying Zhang, Yuqi Liang

Abstract

The price movement prediction of stock market has been a classical yet challenging problem, with the attention of both economists and computer scientists. In recent years, graph neural network has significantly improved the prediction performance by employing deep learning on company relations. However, existing relation graphs are usually constructed by handcraft human labeling or nature language processing, which are suffering from heavy resource requirement and low accuracy. Besides, they cannot effectively response to the dynamic changes in relation graphs. Therefore, in this paper, we propose a temporal and heterogeneous graph neural network-based (THGNN) approach to learn the dynamic relations among price movements in financial time series. In particular, we first generate the company relation graph for each trading day according to their historic price. Then we leverage a transformer encoder to encode the price movement information into temporal representations. Afterward, we propose a heterogeneous graph attention network to jointly optimize the embeddings of the financial time series data by transformer encoder and infer the probability of target movements. Finally, we conduct extensive experiments on the stock market in the United States and China. The results demonstrate the effectiveness and superior performance of our proposed methods compared with state-of-the-art baselines. Moreover, we also deploy the proposed THGNN in a real-world quantitative algorithm trading system, the accumulated portfolio return obtained by our method significantly outperforms other baselines.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper employs advanced deep learning architectures like transformers and heterogeneous graph neural networks, involving complex mathematical formulations and attention mechanisms, which justifies a high math score. It demonstrates empirical rigor through extensive experiments on US and China stock markets, a real-world trading system deployment, and the inclusion of ablation studies, though specific code or detailed backtest metrics are not visible in the provided excerpt.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Stock Price Movement Prediction"] --> B["Data Input<br>US & China Market Historical Prices"]
  B --> C["Step 1: Graph Construction<br>Dynamic Company Relation Graphs per Day"]
  C --> D["Step 2: Feature Encoding<br>Transformer Encoder for Temporal Representations"]
  D --> E["Step 3: Core Modeling<br>THGNN: Heterogeneous Graph Attention Network"]
  E --> F["Prediction Output<br>Target Movement Probability"]
  F --> G["Real-World Deployment<br>Quantitative Algorithm Trading System"]
  G --> H["Key Findings<br>Superior Performance & High Portfolio Returns vs. Baselines"]

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 temporal and heterogeneous graph neural network (THGNN) to learn dynamic relations among price movements for financial time series prediction.

Source sentence
we propose a temporal and heterogeneous graph neural network-based (THGNN) approach to learn the dynamic relations among price movements in financial time series.

Assumes 3

  • Traders spend the same amount on each trading day (e.g., $50,000) to eliminate time dependence. execution
    Source sentence
    Traders spend the same amount on each trading day (for example, $50,000). We made this assumption to eliminate the time dependence of the testing process in order to make a fair comparison.
  • There is always enough liquidity to satisfy opening price orders on the next day. market
    Source sentence
    There is always enough liquidity in the market to satisfy the opening price of the order on the
  • Transaction costs are ignored because they are relatively cheap. execution
    Source sentence
    Transaction costs are ignored because the cost of trading US stocks through brokers is relatively cheap, whether on a transaction-by-transaction basis or on a stock-by-stock basis.

Data

  • S&P 500 constituents market-prices unknown
    2016-2021 · S&P 500
    Source sentence
    We thoroughly evaluate our approach on both the S&P 500
  • CSI 300 constituents market-prices unknown
    2016-2021 · CSI 300
    Source sentence
    We thoroughly evaluate our approach on both the S&P 500

Costs

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

Source sentence
Transaction costs are ignored because the cost of trading US stocks through brokers is relatively cheap, whether on a transaction-by-transaction basis or on a stock-by-stock basis.

Finds 2

  • THGNN achieves significantly higher accumulated portfolio returns than all baselines. performance
    Source sentence
    the accumulated portfolio return obtained by our method significantly outperforms other baselines.
  • Experimental results show that THGNN significantly outperforms state-of-the-art baselines across multiple metrics. performance
    Source sentence
    The experimental results show that our method significantly outperforms state-of-the-art baselines.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 39 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 11 papers in the archive (10 by independent authors): CaT-GNN: Enhancing Credit Card Fraud Detection via Causal … (extends it); R&D-Agent-Quant: A Multi-Agent Framework for Data-Centric … (uses its method or data)

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.