Self-organizing maps as a foundation for charting or geometric pattern recognition in financial time series

Abstract

For a long time technical analysts have detected trading signals with charts. Nonetheless, from a scientific viewpoint, charts are somewhat subjective objects. Using Kohonen`s self-organizing maps (SOMs), the research presented proposes a systematic and automatic approach to charting, or more generally stated, geometric pattern recognition. It is found that the charts discovered using SOM in empirical time series do transmit useful information, and that it is hard for such information to be captured by ordinary econometric methods.