Modified k-means algorithm for clustering stock market companies

In recent years, there has been a lot of interest in the database community in mining time series data, especially in finance markets. Partitioning assets into natural groups or identifying assets with similar properties are natural problems in finance. In this paper, we proposed a modified k-means clustering algorithm to cluster stock market companies, based on similarity measure between time series. This algorithm utilize maximum information compression (MIC) index as similarity measure for clustering them and its comparison with two other similarity measures, namely correlation coefficient and least-square regression error are made. Appling this algorithm leads to a natural partition of the data, as companies belonging to the same industrial branch are often grouped together. This algorithm is applied to the analysis of the Dow Jones (DJ) index companies, in order to identify similar temporal behavior of the traded stock prices. The identification of clusters of companies of a given stock market index can be exploited in the portfolio optimization strategies.

Keywords:

clustering , feature similarity , feature selection , stock market index