||The attention that quantile regression models have received in the recent years show that it is a hot-topic in financial econometris. The assumption of multivariate ellipticity has been the foundation of risk models for many decades. However, if this assumption is incorrect the implied correlation analysis may give very misleading results. For this reason we aim to study the dependence of the joint distribution of financial time series focusing on the quantiles dependence. The method we employ involves copula quantile regressions and high frequency data. It represents a special case of nonlinear quantile regression. This method is able to characterize the whole distribution and allows different relationship between the regressor and the dependent variable at different quantiles. These properties are very appealing, especially in risk management and portfolio modelling. All, finance practitioners and beyond, are interested to know the type of dependence in the financial data, but mostly they are interested in the lower tail dependence of the joint distribution i.e. where the losses occur. We believe that the copula quantile regression method will give more useful information than (what is usually used in literature) the mean regression.