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.
What problem does this paper attempt to address?
This paper attempts to solve the problem of volatility prediction in financial markets, especially how to more accurately capture and model the complex, non - linear interdependencies among multiple market indices. Traditional models such as the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model can effectively capture the volatility clustering phenomenon, but perform poorly when dealing with complex non - linear relationships in multi - asset or global financial markets.
### Main Objectives of the Paper
1. **Improve Volatility Prediction**: By introducing Graph Neural Networks (GNNs), especially the Temporal Graph Attention Network (Temporal GAT), to enhance the ability to predict volatility.
2. **Capture Volatility Spillover Effects**: Use the volatility spillover index to construct a directed graph to better reflect the dynamic influence relationships between different markets.
3. **Combine Temporal and Structural Data**: Combine time - series data with the structural relationships among market indices to form a comprehensive learning model.
4. **Adapt to Market Dynamic Changes**: Study how the dynamic graph structure reflects real - time market changes, enabling the model to adapt to changing market conditions.
### Specific Problem Description
- **Volatility Clustering Phenomenon**: Volatility in financial markets often exhibits clustering characteristics, that is, large price changes may be followed by further large changes, and small changes may be followed by further small changes. This characteristic is very important for investors and risk managers, but it is difficult for traditional models to fully capture.
- **Complex Non - linear Relationships**: In the globally interconnected financial markets, the fluctuation of one market may affect other markets through complex spillover effects. Traditional models usually assume that each asset is independent, ignoring such complex relationships.
- **Temporal Dependence and Structural Dependence**: Fluctuations in financial markets are affected not only by historical data, but also by the structured relationships between different markets. Therefore, a method is needed that can consider these two factors simultaneously.
### Solution
The paper proposes a method based on Graph Neural Networks, which specifically includes the following aspects:
1. **Dynamic Graph Representation**: Represent the global financial market as a dynamic graph, where each market index is a node and the relationships between nodes are edges. In this way, the complex relationships between markets can be better captured.
2. **Temporal Graph Attention Network (Temporal GAT)**: Combine Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs), and dynamically adjust the influence of past events through the temporal attention mechanism, thereby capturing temporal dependence and structural dependence.
3. **Volatility Spillover Index**: Use the volatility spillover index to measure the degree of volatility transmission between different markets and construct a more meaningful graph structure.
### Experimental Results
Through an empirical study of eight major global indices over a period of 15 years, the results show that the Temporal GAT model significantly outperforms the traditional GARCH model and other machine - learning methods in short - to - medium - term predictions. Sensitivity and scenario analyses further verify the effectiveness and robustness of this model.
In conclusion, by introducing GNNs and the volatility spillover index, this paper provides a new tool to capture and model the complex behavior of the global financial market, providing valuable insights for financial analysts, risk managers, and investors.