Fuzzy clustering of financial time series based on volatility spillovers

In this paper we propose a framework for fuzzy clustering of time series based on directional volatility spillovers. In the case of financial time series, detecting clusters of volatility spillovers provides insights into the market structure, which can be useful to both portfolio managers and policy makers. We measure directional—i.e. “From” and “To” the others—volatility spillovers with a methodology based on the generalized forecast-error variance decomposition. Then, we propose a weighted fuzzy clustering model for grouping stocks with a similar degree of directional spillovers. By using a weighted approach, we allow the algorithm to decide which dimension of spillover is more relevant for clustering. Moreover, a robust clustering model is also proposed to alleviate the effect of possible outlier stocks. We apply the proposed clustering model for the analysis of spillover effects in the Italian stock market.