Feature Clustering with Self-organizing Maps and an Application to Financial Time-series for Portfolio Selection

The portfolio selection is an important technique to decrease the risk in stock investment. In portfolio selection, the investor’s property is distributed among a set of stocks in order to minimize the financial risk in market downturns. With this in mind, and aiming to develop a tool to assist the investor in finding balanced portoflios, we achieved a generic method for feature clustering with Self-Organizing Maps (SOM). This method relies on component planes comparison with the modified Rv coefficient and hierarchical clustering of the obtained distances. The ability of neural networks to discover nonlinear relationships in input data makes them ideal for modeling dynamic systems as the stock market.