Centrality Metrics’ Performance Comparisons on Stock Market Datasets

Abstract

The stock market is an essential sub-sector in the financial area. Both understanding and evaluating the mountains of collected stock data has become a challenge in relevant fields. Data visualisation techniques can offer a practical and engaging method to show the processed data in a meaningful way, with centrality measurements representing the significant variables in a network, through exploring the aspects of the exact definition of the metric. Here, in this study, we conducted an approach that combines data processing, graph visualisation and social network analysis methods, to develop deeper insights of complex stock data, with the ultimate aim of drawing the correct conclusions with the finalised graph models. We addressed the performance of centrality metrics methods such as betweenness, closeness, eigenvector, PageRank and weighted degree measurements, drawing comparisons between the experiments’ results and the actual top 300 shares in the Australian Stock Market. The outcomes showed consistent results. Although, in our experiments, the results of the top 300 stocks from those five centrality measurements’ rankings did not match the top 300 shares given by the ASX (Australian Securities Exchange) entirely, in which the weighted degree and PageRank metrics performed better than other three measurements such as betweenness, closeness and eigenvector. Potential reasons may include that we did not take into account the factor of stock’s market capitalisation in the methodology. This study only considers the stock price’s changing rates among every two shares and provides a relevant static pattern at this stage. Further research will include looking at cycles and symmetry in the stock market over chosen trading days, and these may assist stakeholder in grasping deep insights of those stocks.