Dynamic correlation network analysis of financial asset returns with network clustering.

Area of Science:

• Quantitative Finance
• Network Science
• Financial Econometrics

Background:

• Analyzing dynamic correlation networks of highly volatile financial assets presents challenges due to high dimensionality.
• Existing business sector classifications may not adequately capture evolving market interdependencies.
• Volatility fluctuations can distort the true correlation between individual financial assets.

Purpose of the Study:

• To propose a novel network clustering approach for analyzing dynamic correlation networks in financial markets.
• To address high dimensionality issues in financial time series data.
• To provide a framework applicable to various volatile financial and non-financial time series.

Main Methods:

• Employed hierarchical recursive network clustering to group stocks based on filtered returns, reducing dimensionality.
• Filtered stock returns to mitigate volatility effects on correlation.
• Utilized a model-based correlation estimation method to create a dynamic correlation network from adjacency matrices.
• Applied time-axis clustering to summarize the dynamic network into representative sub-period networks.

Main Results:

• Transformed individual stock return correlations into group-based portfolio return correlations.
• Successfully reduced the dimensionality of the dynamic correlation network.
• Identified three representative correlation networks through time-axis clustering, enabling intertemporal comparisons.
• Demonstrated the framework's applicability to Japanese stock returns.

Conclusions:

• The proposed network clustering method effectively handles high dimensionality in dynamic correlation network analysis.
• The framework provides a robust method for understanding market-level correlation dynamics.
• The approach is versatile and can be extended to diverse financial and non-financial volatile time series data.