Improving data efficiency for analyzing global exchange rate fluctuations based on nonlinear causal network-based...

Area of Science:

• Computational Finance
• Network Science
• Information Theory

Background:

• Predicting currency value fluctuations is crucial for financial markets.
• Traditional methods often overlook complex nonlinear causal relationships between currencies.

Purpose of the Study:

• To predict currency value fluctuations using information and network theory.
• To analyze and model nonlinear causal relationships between 48 currencies over 25 years.

Main Methods:

• Calculated causal relationships using logarithmic return (log-return) and entropic value-at-risk (EVaR).
• Quantified causal relationships via transfer entropy.
• Modeled and analyzed information flow as a network.
• Classified currencies using hierarchical clustering.
• Predicted fluctuations with machine learning based on network topology.

Main Results:

• Information flow-based nonlinear causal relationships differ from the established key currency order.
• Network analysis revealed distinct currency communities based on causal relationships.
• Machine learning models showed improved currency fluctuation predictions using data from these communities.

Conclusions:

• Statistically significant nonlinear causal relationships provide a novel basis for currency classification.
• Leveraging network topology and information flow enhances data efficiency in currency prediction models.
• This approach offers a more nuanced understanding of inter-currency dynamics.