HIDDEN MARKOV INDEPENDENT COMPONENT ANALYSIS AS A MEASURE OF COUPLING IN MULTIVARIATE FINANCIAL TIME SERIES
Nauman Shah, Stephen Roberts · 2008
Modelling the dynamics of financial markets has been an area of active research in recent years. This paper presents a time series analysis model which can be used to infer patterns within financial data, in order to better understan d the dynamics of financial markets. The focus of the paper is on finding causal and time-scale relationships between financial time series. Wavelets are used to extract useful time-scale information from financial data at different frequencies and mutual information between time series is used as the canonical measure of coupling. A Hidden Markov Independent Component Analysis (HMICA) model is used to infer a series of hidden states and it is shown that these hidden states are indicative of changes in mutual information between time series at various different time scales.