Independent component analysis and resolution pursuit with wavelet and cosine packets

Abstract

We present an application of Independent Component Analysis combined with Multiresolution Analysis and the Matching Pursuit algorithm for exploring high frequency financial time series. We show that these are powerful instruments in the finance data mining domain, especially when dealing with signal decomposition and approximation obtained through wavelet and cosine packet dictionaries. With intra-daily temporal series, some features in the underlying stochastic processes may remain undetected by standard models applied to the observed data; thus, capturing the latent dependencies and extracting features such as hidden periodic components represent crucial tasks. Independent Component Analysis results to be particularly relevant for the scopes of suggesting a better compromise for the time and frequency resolution pursuit and a better efficiency and accuracy of the Matching Pursuit performance.