Independent Component Analysis and Feature Extraction of Financial Time Series

Abstract

This thesis discusses the application of a modern signal processing technique known as independent component analysis (ICA) or blind source separation to univariate time series. To perform single channel ICA on this univariate time series, we work within the embedding framework, using Takens’ delay coordinate maps. After a brief presentation of the results obtained with PCA (Signal/Noise decomposition, dimensionality reduction), we show that the same kind of experiments can be done with ICA. Studies done so far have yielded encouraging results among which the following emerge as the most noteworthy: - ICA, just like PCA, preserves the possibility to perform a Signal/Noise decomposition. - Independent components (ICs) reveal evidence of clustering amongst them. - The possibility to efficiently rank the ICs. Using all these results, we show that the time series can be reconstructed surprisingly well by using a small number of weighted ICs. Independent component analysis seems to be a promising powerful method of analyzing and understanding driving mechanisms in financial markets.