An Alternative Approach to Visualizing Stock Market Correlation Matrices- An Empirical study of forming portfolios that contain only small numbers of stocks using both existing and newly discovered visualization methods

The core of stock portfolio diversification is to pick stocks from different correlation clusters when forming portfolios. The result is that the chosen stocks will be only weakly correlated with each other. However, since correlation matrices are high dimensional, it is close to impossible to determine correlation clusters by simply looking at a correlation matrix. It is therefore common to regard industry groups as correlation clusters. In this thesis, we used three visualization methods namely Hierarchical Cluster Trees, Minimum Spanning Trees and neighbor-Net splits graphs to “collapse” correlation matrices’ high dimensional structures onto two-dimensional planes, and then assign stocks into different clusters to create the correlation clusters. We then simulated sets of portfolios where each set contains 1000 portfolios, and stocks in each of the portfolio were picked from the correlation clusters suggested by each of the three visualization methods and industry groups (another way of determine correlation clusters). The mean and variance distribution of each set of 1000 simulated portfolios gives us an indication of how well those clusters were determined. The examinations were conducted on two sets of financial data. The first one is the 30 stocks in the Dow Jones Industrial average which contains relatively small number of stocks and the second one is the ASX 200 which contains relatively larger number of stocks. We found none of the methods studied consistently defined correlation clusters more efficiently than others in out-of-sample testing. The thesis does contribute the finance literature in two ways. Firstly, it introduces the neighbor-Net method as an alternative way to visualize financial data’s underlying structures. Secondly, it used a novel “visualizatio