Analysis of stock markets using complex network approach

Abstract

Stock market behavior analysis is one of the most challenging studies. Many people believe that there is no simple way to predict the trends of the stock market. To deal with the complexity of market movement, this paper aims at using complex network theory to model the behavior of listed stocks and studies its characteristics, such as Average Centrality, Average Degree, and Average Eigenvector Centrality. Furthermore, these topology features would be combined with the machine learning framework and predict the next-day movement (up or drop) of the Hong Kong Hang Seng Index (HSI). To construct the network, the daily close price of 185 noticeable stocks in the Hong Kong stock market from 01-2012 to 12-2019, in a total of 1972 trading days, were collected. Each trading day was a network where the link weights were measured by Mutual Information score. To implement a machine learning algorithm, the data between 01-2012 and 12-2018 were the training samples and data from 01-2019 to 12-2019 were the testing samples. The result shows that the predicting model had greater than 60% accuracy on next-day movement over the year of 2019, in total of 246 trading day, by using the topology feature, Average Eigenvector Centrality, as the learning input.