Clustering Heteroskedastic Time Series by Model-Based Procedures