Time series analysis of statistical characteristics of CDS and stock markets

Abstract

Econophysics studies the economic and financial systems through tools and methodologies used in physics. This dissertation discusses the statistical characteristics of the U.S. CDS market and stock market through time series analysis. First, we investigate two markets using random matrix theory and network analysis. We compare distributions of the eigenvectors and eigenvalues of the empirical cross-correlation matrix with those of the random cross-correlation matrix. The deviating eigenvalues and the corresponding eigenvectors represent genuine information of the market. Time window analysis is applied for both markets to identify change of the largest eigenvalue which indicates the market-wide effect. We show that the change of the largest eigenvalue of CDS market precedes that of stock market in the subprime mortgage crisis period. Second, we analyze the information flow between industrial sectors for inter- and intra-structures of two markets through transfer entropy. When k = 3, the transfer entropy declines in the subprime mortgage crisis period, but increases in the European sovereign debt crisis period. When k = 7, the opposite pattern appears. Changing trends of the transfer entropy of the two markets are similar, but the change in the CDS market precedes that in the stock market. These results suggest that observations of the CDS market may help predict the stock market movements.