Paper: arXiv 2502.15813

Authors: Meet Satishbhai Sonani, Atta Badii, Armin Moin

Abstract

This paper presents a novel hybrid model that integrates long-short-term memory (LSTM) networks and Graph Neural Networks (GNNs) to significantly enhance the accuracy of stock market predictions. The LSTM component adeptly captures temporal patterns in stock price data, effectively modeling the time series dynamics of financial markets. Concurrently, the GNN component leverages Pearson correlation and association analysis to model inter-stock relational data, capturing complex nonlinear polyadic dependencies influencing stock prices. The model is trained and evaluated using an expanding window validation approach, enabling continuous learning from increasing amounts of data and adaptation to evolving market conditions. Extensive experiments conducted on historical stock data demonstrate that our hybrid LSTM-GNN model achieves a mean square error (MSE) of 0.00144, representing a substantial reduction of 10.6% compared to the MSE of the standalone LSTM model of 0.00161. Furthermore, the hybrid model outperforms traditional and advanced benchmarks, including linear regression, convolutional neural networks (CNN), and dense networks. These compelling results underscore the significant potential of combining temporal and relational data through a hybrid approach, offering a powerful tool for real-time trading and financial analysis.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper combines advanced deep learning architectures (LSTM and GNN) requiring knowledge of sequential processing and graph theory, placing it on the higher end of math complexity. Empirical rigor is solid with an expanding window validation, MSE metrics, and benchmarking against multiple models, indicating backtest-ready methodology.

Research Flowchart

  flowchart TD
  A["Research Goal: Enhance Stock Price Prediction<br>using Temporal and Relational Data"] --> B["Data Input: Historical Stock Prices"]

  subgraph Methodology ["Methodology: Hybrid LSTM-GNN Model"]
      direction LR
      B --> C["LSTM Component<br>Captures Temporal Patterns"]
      B --> D["GNN Component<br>Models Inter-Stock Relations via Pearson Correlation"]
  end

  Methodology --> E["Training: Expanding Window Validation<br>Continuous Learning & Adaptation"]

  E --> F["Computational Process:<br>Integrate Time-Series & Graph-Based Analysis"]

  F --> G["Key Findings & Outcomes"]
  subgraph Outcomes
      direction LR
      G1["MSE: 0.00144<br>10.6% reduction vs. Standalone LSTM"]
      G2["Outperforms Benchmarks<br>Linear Regression, CNN, Dense Networks"]
  end

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 hybrid model that combines Long Short-Term Memory (LSTM) networks with Graph Neural Networks (GNNs) for stock price prediction.

Source sentence
This paper presents a novel hybrid model that integrates long-short-term memory (LSTM) networks and Graph Neural Networks (GNNs) to significantly enhance the accuracy of stock market predictions.

Assumes 3

  • Past inter‑stock relationships are assumed to persist into the future for the GNN component. market
    Source sentence
    assuming that past relationships persist into the future may not hold during unprecedented market events or structural economic changes
  • The expanding window training approach assumes continuous availability of new data for retraining. data
    Source sentence
    progressively increased the training dataset by incorporating new data as it became available
  • The model assumes that the selected ten stocks are representative of broader market dynamics. market
    Source sentence
    Ten prominent stocks representing diverse sectors were selected

Data

  • Kaggle via YFinance API · Yahoo Finance market-prices public
    2005-01-01 to 2023-12-31 · Apple, Microsoft, Comcast, Costco, Qualcomm, Adobe, Starbucks, Intuit, AMD, Intel
    Source sentence
    The dataset was obtained from Kaggle via the YFinance API

Recognised sources: Kaggle Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The hybrid LSTM‑GNN model achieved an MSE of 0.00144, a 10.6% reduction compared to the standalone LSTM MSE of 0.00161. performance
    Source sentence
    our hybrid LSTM-GNN model achieves a mean square error (MSE) of 0.00144, representing a substantial reduction of 10.6% compared to the MSE of the standalone LSTM model of 0.00161
  • The hybrid model outperformed linear regression (MSE 0.00224) and other benchmarks such as CNN and DNN. predictive-accuracy
    Source sentence
    The LSTM‑based models outperformed the DNN and CNN
  • Expanding window training improved adaptability, mitigated concept drift, and enhanced generalisation. statistical
    Source sentence
    The expanding window training approach offers several key advantages. Firstly, it enhances adaptability
  • The hybrid model incurs higher computational demands and sensitivity to hyper‑parameter settings. negative
    Source sentence
    combined with expanding window training, resulted in higher computational demands... The

Checked by others

Supported by

  • NeuralFactors: A Novel Factor Learning Approach to Generative Modeling of Equities (2024)
    On the question: Does using longer training windows improve equity price prediction performance and generalisation?
    Compare the findings

    This paper: Expanding window training improved adaptability, mitigated concept drift, and enhanced generalisation.

    That paper: Increasing training data length improves performance, contradicting the belief that non‑stationarity requires shorter windows.

    Both findings report that expanding the training window boosts performance, mitigates concept drift and enhances generalisation, contrary to the belief that short windows are needed.

  • HARd to Beat: The Overlooked Impact of Rolling Windows in the Era of Machine Learning (2024)
    On the question: Does increasing the training window size improve forecasting accuracy and adaptability?
    Compare the findings

    This paper: Expanding window training improved adaptability, mitigated concept drift, and enhanced generalisation.

    That paper: Larger training windows generally lead to lower RMSE for the HAR model.

    Both findings report that larger/expanding windows lower error (RMSE) and enhance generalisation, showing consistent benefits across HAR and LSTM‑GNN approaches.

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.