Résumé en anglais

We examine the unconditional distribution of the realized variance of three European stock market indexes obtained from intraday transaction prices. We find that they share common distributional features: a significant mass close to zero, a sharp decrease afterwards and a significant right tail. Their important differences, however, compel us to model them non-parametrically through lognormal kernel estimators. We then move to the analysis of their dependence structure and find strong evidence of asymmetry. Hence, unlike common practice, we resort to non-exchangeable copula models. Such a characterization also allows us to assess the direction of greater contamination among stock market variances.