Mining financial time series: New insights from model-based clustering methods Jos´ e G. Dias * Department of Quantitative Methods & UNIDE, ISCTE – Lisbon University Institute, Portugal Jeroen K. Vermunt Department of Methodology and Statistics, Tilburg University, The Netherlands Sofia Ramos Department of Finance & UNIDE, ISCTE – Lisbon University Institute, Portugal Abstract In recent years large amounts of financial data have become available for analysis. We propose to explore returns from 21 European stock markets by model-based clustering of regime switching models. These models allow the relaxation of traditional assumptions such as conditional Gaussian re- turns. The data mining approach handles simultaneously the heterogeneity across stock markets and time, i.e., time-constant and time-varying discrete latent variables capture unobserved heterogeneity between and within stock markets, respectively. The results show a clear distinction between groups of stock markets, each one characterized by different regime switching dynamics that correspond to different expected return-risk patterns. Keywords: Data mining, hidden Markov model, stock indexes, cluster analysis, model-based clustering, latent class model * Corresponding author. Tel.: 351 217 903 208 Fax.: 351 217 903 004 Email addresses: jose.dias@iscte.pt (Jos´ e G. Dias), J.K.Vermunt@uvt.nl (Jeroen K. Vermunt), sofia.ramos@iscte.pt (Sofia Ramos) Preprint submitted to European Journal of Operational Research December 14, 2010