Building Dynamic Correlation Network for Financial Asset Returns

Abstract

This paper studies the dynamic correlation matrix estimation of highly volatile financial returns, which is used to build a dynamic correlation network. The widely used method of calculating time-dependent linear correlation matrices by moving window of a fixed sample period can have fundamental problems when applied to fat-tailed returns. A multivariate volatility model, DCC-GARCH, is employed to filter the fat-tailed returns and estimate the dynamic correlation of returns in order to overcome such difficulties. The time-dependent correlation matrices are calculated and compared with the ones that are calculated by the traditional calculation method to highlight the advantages of the proposed dynamic correlation based method. As a case study, the model is fitted to the Japanese stock returns to analyze dynamic changes in the correlation matrix. The method is not limited to financial returns, but can also be applied to build a dynamic correlation network of other time series data with high volatility.