Paper: arXiv 2410.16858

Authors: Pulikandala Nithish Kumar, Nneka Umeorah, Alex Alochukwu

Abstract

Volatility forecasting is essential for risk management and decision-making in financial markets. Traditional models like Generalized Autoregressive Conditional Heteroskedasticity (GARCH) effectively capture volatility clustering but often fail to model complex, non-linear interdependencies between multiple indices. This paper proposes a novel approach using Graph Neural Networks (GNNs) to represent global financial markets as dynamic graphs. The Temporal Graph Attention Network (Temporal GAT) combines Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs) to capture the temporal and structural dynamics of volatility spillovers. By utilizing correlation-based and volatility spillover indices, the Temporal GAT constructs directed graphs that enhance the accuracy of volatility predictions. Empirical results from a 15-year study of eight major global indices show that the Temporal GAT outperforms traditional GARCH models and other machine learning methods, particularly in short- to mid-term forecasts. The sensitivity and scenario-based analysis over a range of parameters and hyperparameters further demonstrate the significance of the proposed technique. Hence, this work highlights the potential of GNNs in modeling complex market behaviors, providing valuable insights for financial analysts and investors.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced graph neural network theory and temporal attention mechanisms, indicating high math complexity, while the 15-year backtest on eight global indices with sensitivity analysis demonstrates strong empirical rigor.

Research Flowchart

  flowchart TD
  Start["Research Goal: Improve Volatility Prediction"] --> Method["Proposed Method: Temporal Graph Attention Network (GAT)"]
  Method --> Data["Data: 8 Major Global Indices (15 Years)"]
  Data --> Process["Process: Construct Dynamic Graphs using Correlation & Spillover Indices"]
  Process --> Compute["Compute: Apply GNNs to Capture Non-Linear Dependencies"]
  Compute --> Outcome["Outcome: Outperforms GARCH & ML Models"]
  Outcome --> Result["Result: Enhanced Accuracy for Short-Mid Term Forecasts"]

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 Graph Attention Network (Temporal GAT) that combines Graph Convolutional Networks and Graph Attention Networks to model volatility spillovers as dynamic graphs.

Source sentence
This paper proposes a novel approach using Graph Neural Networks (GNNs) to represent global financial markets as dynamic graphs. The Temporal Graph Attention Network (Temporal GAT) combines Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs) to capture the temporal and structural dynamics of volatil

Assumes 4

  • Volatility spillover indices accurately capture the directional relationships between market indices. model
    Source sentence
    The spillover index matrix provides a directional measure of volatility transfer, illustrating which indices are net transmitters or receivers of volatility.
  • Realized volatility can be computed reliably from daily adjusted closing prices. data
    Source sentence
    The realized volatility (RV) data was computed using daily adjusted closing prices, following the approach suggested by Andersen et al.
  • The graph structure is static during model training. model
    Source sentence
    assuming static graph structures during training may not fully capture the evolving nature of financial markets during structural shifts.
  • The selected eight indices are representative of global financial market dynamics. market
    Source sentence
    These indices were selected due to their significant influence on global financial markets.

Data

Backtest sample: global stock indices · equities · daily

  • Eight major global market indices · Yahoo Finance market-prices public
    Nov 2007 - Jun 2022 · S&P 500, DAX, CAC 40, FTSE 100, Nifty 50, Nikkei 225, KOSPI, Hang Seng
    Source sentence
    The dataset was sourced from Yahoo Finance and spans November 2007 to June 2022.

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Temporal GAT significantly outperforms traditional GARCH and MLP models, especially for short- to mid-term volatility forecasts. performance
    Source sentence
    The results indicate that the Temporal GAT significantly outperforms traditional models, particularly in short- to mid-term forecasts.
  • Temporal GAT achieves a markedly lower Mean Absolute Forecast Error (MAFE) than baseline models. predictive-accuracy
    Source sentence
    Notably, the MAFE for the Temporal GAT was markedly lower than that of baseline models.
  • Using a volatility spillover index to construct the graph yields better predictive performance than correlation‑based graph construction. performance
    Source sentence
    The volatility spillover index outperformed simpler correlation-based methods in constructing the graph, capturing the propagation of shocks across markets.
  • Leave‑one‑out sensitivity analysis shows that removing major indices like the S&P 500 or DAX substantially reduces forecast accuracy. statistical
    Source sentence
    Their removal significantly reduced predictive accuracy, while indices with weaker spillover effects, like the HANG SENG, tended to add noise.
  • Temporal GAT remains robust during periods of heightened market volatility, such as the COVID‑19 pandemic, maintaining superior accuracy. performance
    Source sentence
    The Temporal GAT also showed robustness during heightened market volatility, such as the COVID-19 pandemic, maintaining superior accuracy compared to traditional models.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 46 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 4 papers in the archive (4 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 (50)

  1. Statistical models for financial volatility (1993)
  2. Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation (1982)
  3. Graph neural networks: A review of methods and applications (2020)
  4. Certain speculative prices (1963) (1972)
  5. Deep attentive learning for stock movement prediction from social media text and company correlations (2020)
  6. Temporal and heterogeneous graph neural network for financial time series prediction — → 2305.08740
  7. Inductive representation learning on temporal graphs (2020) — link
  8. Graph convolutional networks: a comprehensive review (2019)
and 42 more
  1. Machine learning on graphs: A model and comprehensive taxonomy (2022)
  2. A comprehensive survey on graph neural networks (2020)
  3. Deep learning on graphs: A survey (2020)
  4. Geometric deep learning: going beyond euclidean data (2017)
  5. Generalized autoregressive conditional heteroskedasticity (1986)
  6. Conditional heteroskedasticity in asset returns: A new approach (1991)
  7. A general framework for adaptive processing of data structures (1998)
  8. Neural networks: a comprehensive foundation (1998)
  9. Better to give than to receive: Predictive directional measurement of volatility spillovers (2012)
  10. The anatomy of a large-scale hypertextual web search engine (1998)
  11. The graph neural network model (2008)
  12. Multi-graph convolutional network for relationship-driven stock movement prediction (2021)
  13. Graph-based lstm for anti-money laundering: Experimenting temporal graph convolutional network with bitcoin data (2023)
  14. Incorporating corporation relationship via graph convolutional neural networks for stock price prediction (2018)
  15. A novel graph convolutional feature based convolutional neural network for stock trend prediction (2021)
  16. Forecasting stock prices using stock correlation graph: A graph convolutional network approach (2021)
  17. Deeper insights into graph convolutional networks for semi-supervised learning (2018)
  18. How powerful are graph neural networks? (2018) — link
  19. Semi-supervised classification with graph convolutional networks (2016) — link
  20. Empirical properties of asset returns: stylized facts and statistical issues (2001)
  21. Threshold heteroskedastic models (1994)
  22. Advances in financial machine learning (2018)
  23. Econometric analysis of realized volatility and its use in estimating stochastic volatility models (2002)
  24. A new model for learning in graph domains (2005)
  25. Measuring financial asset return and volatility spillovers, with application to global equity markets (2009)
  26. The distribution of realized stock return volatility (2001)
  27. Neural machine translation by jointly learning to align and translate (2014) — link
  28. Forecasting stock index returns using ARIMA-SVM, ARIMA-ANN, and ARIMA-random forest hybrid models (2014)
  29. A comprehensive comparative study of artificial neural network (ANN) and support vector machines (SVM) on stock forecasting (2023)
  30. Hybrid deep learning and GARCH-family models for forecasting volatility of cryptocurrencies (2023)
  31. Novel volatility forecasting using deep learning–long short term memory recurrent neural networks (2019)
  32. Forecasting the volatility of stock price index: A hybrid model integrating LSTM with multiple GARCH-type models (2018)
  33. Financial risk measurement for financial risk management (2013)
  34. Forecasting global stock market volatility: The impact of volatility spillover index in spatial-temporal graph-based model (2023)
  35. Universal features of price formation in financial markets: perspectives from deep learning (2021)
  36. Commonality in liquidity (2000)
  37. Multifractality in asset returns: theory and evidence (2002)
  38. Empirical asset pricing via machine learning (2020)
  39. A semi-supervised graph attentive network for financial fraud detection (2019)
  40. Network Filtering of Spatial-temporal GNN for Multivariate Time-series Prediction (2022)
  41. Understanding commonality in liquidity around the world (2012)
  42. Graph attention networks (2017) — link

1 of 50 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.