Noise Undressing and Information Identifying of the Financial Correlation Matrix

Abstract

We apply the random-matrix approach to undress the noise of the cross correlation matrix constructed from Shanghai Stock Exchange (SSE) for the period 2001-2008. The empirical evidence shows that, about 7.4% of the eigenvalues fall out the RMT bounds, and the eigenvalues within the bounds agree with the universal properties of random matrix, implying a large degree of noise in the correlation matrix. We also find that SSE has a particularly high value of the largest eigenvalues of 209.26, which is significantly different from other exchanges.