Building a dynamic correlation network for fat-tailed financial asset returns.

Area of Science:

• Quantitative Finance
• Network Science
• Financial Econometrics

Background:

• Estimating dynamic correlation matrices for financial assets is crucial for risk management.
• Traditional methods using fixed moving windows struggle with the fat-tailed nature of financial returns, leading to spurious correlations.
• Existing approaches lack robustness in capturing genuine dynamic shifts in asset interdependencies.

Purpose of the Study:

• To present a novel filtering approach for constructing dynamic correlation networks of highly volatile financial asset returns.
• To address and mitigate the problem of spurious correlations in dynamic correlation matrix estimation.
• To provide a more reliable method for detecting significant changes in financial asset correlations.

Main Methods:

• A filtering approach is employed to preprocess financial asset returns.
• The multivariate volatility model, Dynamic Conditional Correlation-Generalized Autoregressive Conditional Heteroskedasticity (DCC-GARCH), is utilized for filtering fat-tailed returns.
• Network analysis, including topological measures and indicators like largest eigenvalue and cosine distance, is applied to correlation matrices.

Main Results:

• The proposed filtering method demonstrates superior reliability in detecting dynamic changes compared to traditional fixed moving window techniques.
• Analysis of Japanese stock returns reveals higher network density and lower heterogeneity during stress periods.
• The dynamic correlation network of stock returns exhibits robustness over time.

Conclusions:

• The novel filtering approach effectively overcomes spurious correlation issues in dynamic correlation network construction for volatile financial assets.
• The method provides a robust framework for analyzing financial market dynamics and identifying stress events.
• The approach is broadly applicable to various financial and non-financial high-volatility time series.