Development of Stock Correlation Network Models Using Maximum Likelihood Method and Stock Big Data

Stock correlation network uses stock return to study the relationship between different stocks traded in the stock market. The method of general threshold is a basic idea to use the highest values of correlation coefficients to develop stock network. However, there are two disadvantages. First, the correlation coefficient can only quantify linear relationship. In real case, there would be more non-linear relationships. Additionally, the general threshold favours strong correlations, which will lead to neglect of some information. To address these issues, this work will introduce a new method to measure the relationship between two variables. And maximum likelihood method will be applied to select the optional correlation relationship. Using 280 stocks traded at the Shanghai Stocks Exchange in China during from year 2014 to 2016, we first compare the function of correlation coefficient and mutual information to measure the proximity between two objects. Based on mutual information and correlation coefficient, we then develop two stock networks by two different construction mechanism and study the topological properties of these networks.