Classification of Financial Returns According to Thresholds Exceedances

Abstract

Properties of a panel of financial time series arc explored, aiming at classifying market shares according to their extremal returns behaviour. Existing methods for optimal portfolio selection involve estimation of correlation coefficient, whose properties for measuring dependence in financial time series are questionable. Alternatively, for stationary processes of financial returns, the mean size of cluster of thresholds exceedances leads to define a measure of extremal dependence more accurate than correlation. Further functionals that might help to optimal portfolio selection, are, for instance, the total loss occurred to a stock during an extreme event or the time-length duration of a loss in a stress period. Combining functionals of financial returns it is possible to clustering shares properly and setting up a tool for portfolio selection. The performance of this method is assessed, through an application to real financial time series, by means of standard Markowitz theory of optimal selection of shares.