Dependence structures in financial time series: a chaos-theoretic approach

Of much interest in financial econometrics is the recovery of joint distributional behaviour of collections of contemporaneous financial time series, e.g., two related commodity price series, or two asset returns series. An approach to model their joint behaviour is to use copulas. Essentially, copulas are selected on the basis of a measure of correlation between the two series and are made to match their marginal properties. Of course, generalisations exist for more than two series. A possible limitation of this approach is that only linear correlations between series might be captured. We consider incorporating more general dependence structures, through the use of the correlation integral (as in the BDS test), as a means to refine the choice of candidate copulas in an empirical situation.