Clustering Stock Market values with a Self-Organized feature Map

Stock market values do not follow a normal distribution. Instead, market observations show evidence of heavy-tail histograms that represent much more extreme events that what would be expected by a normal distribution. In this paper we will discuss some optimizations to the Self-Organizing Map that can take better account of this extreme data. We describe the process of pattern extraction from raw stock prices time series and present a new weighted-Euclidean metric for competitive learning when dealing with nancial patterns. Results support the fact that stock prices are not normally distributed and other ndings are also reported, namely what can be inferred from trained maps.