Chapter 6: Discovering the Co-Movement Structure of Chinese Stock Market By Space With Em Algorithm

Co-movement structure of stocks and its change along with time is important for investment and risk management purposes. In this chapter, the co-movement among stocks is measured by partial correlation coefficients of continuously compounded daily returns, and characterized by conditional independence graphs. Traditionally, studies on co-movement are limited to a few stocks or market indices due to numerical difficulties and poor prediction accuracy with high dimensional data. The SPACE (Sparse Partial Correlation Estimation) model by Peng, Wang, Zhou, and Zhu (2009) makes it possible to estimate the partial correlation coefficients among stocks with rather high dimension. However, the model does not allow omission of data; therefore the SPCE model cannot be immediately applied to financial data. This is because missing data can occur very often for any stocks in a day. In this chapter, we modify the SPACE model with EM (Expectation-Maximization) algorithm (SPACE.MISS) to make full use of available data. Simulation study shows that the SPACE.MISS converges fast and estimates accurately under a moderate amount of missing data. The application of SPACE.MISS is also exhibited for the Chinese stock market, and it successfully discovers the co-movement structure changes after special social and natural events.