Fig. 1
From: Predicting stock market movements using network science: an information theoretic approach
Analysis Process a gather time-series stock price records of the S&P 500 underlying companies, and split the records by one hour records, b calculate mutual information of the stock pairs, c construct networks using the mutual information as link weights, d computing the strength distribution for each network, e build metrics with the strength distribution data and other network measurements (average, median and maximum values of eigenvector and betweenness centralities are used). The plotted centralities are maximum values of centralities, f predict the amplitude of S&P 500 changes by forming a linear combination of top performing metrics, and forecast actual S&P 500 index by building ARIMA models with network measurements