Network Analysis of the Stock Market

In this study, we built a network for the US stock market based on the correlation of different stock returns. Community detection techniques were then applied to the constructed correlation network. The resulting communities were consistent with the identified market sections using Standard Industrial Classification code, which demonstrates that performances of public stocks within the same sector tend to have similar patterns. Furthermore, we used an open-source network analysis and visualization software, Gephi, to generate visualizations of the return correlations among various public stocks. The visualization results offer a very intuitive way to look at the overall correlation structure of different public stocks and identify key market segments, which could be very useful for real practice such as market monitoring. For network applications, We first looked at the US credit crisis spreading represented by the spread of negative stock performances between Jul/2007 and Feb/2009. Stock performance were classified into three categories based on the rate of return (-20%, 0%, and 20%), then a sequence of snapshots of the stock network, colored by the three defined categories, show the cascading behavior of stock performances. We observed that the cascading starts from some stocks in different communities, and then spread from the initially infected (negative return) stocks. This observation indicates that the established stock network could potentially be very useful for stock market performance prediction from macroeconomic factors. Then we investigated the application of network in portfolio management. Traditional approaches of portfolio management often rely on certain statistical properties, such as expected return and price variance. These properties, however, generally represent the local behavior of the stocks and are thus not able to represent the stock characteristics in terms of whole stock market. One advantage of creating a network characterizing different stocks within a market is that some important global properties of the stock within the network can be extracted, such as degree centrality, betweenness centrality, and closeness centrality. In this work, we did some preliminary studies of creating investment portfolios using the network properties of different stocks, and then compare the performance with some standard market index such as NASDAQ and S&P 500. Results show that with proper weights given to the top centrality nodes (i.e. stocks), we could outperform the S &P 500 for certain periods.