A0342
Title: Copula-based clustering via evidence accumulation: An application to portfolio diversification
Authors: Andrea Mecchina - University of Trieste (Italy) [presenting]
Roberta Pappada - University of Trieste (Italy)
Nicola Torelli - University of Trieste (Italy)
Abstract: Understanding the dependence structure of the time series of financial returns is fundamental to risk assessment and portfolio diversification. To this end, relevant information lies in the pairwise association among asset returns in the left tail of their joint distribution. Copula models allow the measurement of extreme dependencies using finite (lower) tail dependence coefficients. Such coefficients can be used to define copula-based dissimilarities that allow the identification of clusters of time series accounting for their co-movement, regardless of marginal modeling. The resulting clusters might depend critically on the selected copula model and on the other clustering-specific choices. Hence, a novel approach is proposed in which evidence is accumulated from multiple classifications obtained from different copula models and clustering methods. In particular, a co-occurrence matrix of assets in the same cluster is derived, from which the final partition is obtained. Such a partition proved to be more robust and less dependent on specific design choices. Assets in the same cluster are expected to perform similarly in risky scenarios, thus making such information valuable to risk-averse investors. An empirical demonstration of the benefits of a clustering-based diversification strategy is presented using data from the EURO STOXX 50 index and assessed through performance metrics.
