Quantile-based fuzzy C-means clustering of multivariate time series: Robust techniques

Robust fuzzy clustering of multivariate time series is addressed when the purpose grouping together generated from similar stochastic processes. Robustness to presence anomalous attained by considering three well-known robust versions a C -means model based on spectral dissimilarity measure with high discriminatory power. The compares principal component scores obtained estimates quantile cross-spectral densities, and techniques follow so-called metric, noise trimmed approaches. metric approach incorporates in objective function distance aimed at neutralizing effect outliers, builds an artificial cluster expected contain outlying series, removes most atypical dataset. As result, proposed methods take advantage both nature these capability density identify complex dependence structures. An extensive simulation study including linear, nonlinear GARCH processes shows that algorithms are substantially effective coping clearly outperforming other alternative procedures. Two specific applications regarding financial environmental illustrate usefulness presented methods.