Enhancing Exchange-Traded Fund Price Predictions: Insights from Information-Theoretic Networks and Node Embeddings

Area of Science:

• Quantitative Finance
• Financial Econometrics
• Network Science

Background:

• Predicting financial market movements, particularly for Exchange Traded Funds (ETFs), is crucial for investment strategies.
• Traditional methods often struggle to capture complex, nonlinear relationships within market data.
• Understanding the dynamics of U.S. sector index ETFs requires advanced analytical techniques.

Purpose of the Study:

• To develop a novel approach for predicting price fluctuations in U.S. sector index ETFs.
• To identify and quantify nonlinear dependencies between key financial variables.
• To enhance the accuracy and explainability of financial market forecasting models.

Main Methods:

• Utilized information-theoretic measures, including mutual information and transfer entropy, to construct threshold networks.
• Derived network-based features such as centrality measures and node embeddings.
• Integrated these derived features into gradient-boosting algorithm-based predictive models.

Main Results:

• The constructed threshold networks effectively highlighted nonlinear dependencies between log returns and trading volume rate changes.
• Network-derived features provided unique insights into ETF dynamics.
• The integration of these features led to a significant enhancement in predictive accuracy for ETF price fluctuations.

Conclusions:

• The proposed information-theoretic and network-based approach offers superior predictive performance for U.S. sector index futures.
• This method provides enhanced explainability compared to existing financial forecasting models.
• The study demonstrates the value of leveraging nonlinear dynamics and network structures in quantitative finance.