MULTIVARIATE DECOMPOSITIONS FOR VALUE AT RISK MODELLING

This paper presents the application of independent component analysis (ICA) for value at risk modelling (VaR). The probabilistic models fitted to hidden components from the time series help to identify the independent factors influencing the portfolio value. An important issue here is the choice of the ICA algorithm, especially taking into account the characteristics of the instruments with respect to higher-order statistics. The proposed ICA-VaR concept has been tested on transactional data of selected stocks listed on Warsaw Stock Exchange.