Abstract

Vine copulas are a well-known and widely accepted tool for modelling and forecasting financial time series. Their popularity stems from their flexibility in modelling the dependence between series, as well as their tractability even in high dimensions. However, despite the good results obtained in empirical applications, the construction of vine copulas depends on the ordering of the time series under consideration, which can significantly impact the results. In this work, we highlight some practical consequences of this issue and propose an easy-to-implement solution. Using data from the seven major cryptocurrencies, we illustrate the consequences of neglecting this issue when applying vine copulas to two popular financial problems: risk measure forecasting and portfolio allocation. Finally, we demonstrate the superior performance of our proposal using the same dataset.