Cryptocurrencies and Dependence: A Nonparametric View

Modeling dependence between cryptocurrency returns has received great attention in the last years. Most papers use Pearson’s correlation measure to quantify the degree of dependence, while other use parametric copula methods together with sophisticated linear or nonlinear regression methods to unearth complicated dependence patterns between them. In this article, we show that simple dependence measures might lead to misleading results and provide alternative and more robust measures of dependence, some quite recent ones. In addition, utilizing the flexibility of copula functions in characterizing the joint dependence, we employ sophisticated methods in estimating and plotting the copula density function non-parametrically, therefore avoiding issues of model misspecification. Using almost 2 years of daily return data of the 10 most traded cryptocurrencies we have first properly quantified the degree of dependence between different pairs and by visualizing the non-parametrically estimated bivariate copula density for most pairs, we have uncovered asymmetric tail dependence between them, hence shedding further light on this issue.