Abstract
Finite mixture distributions provide a flexible method for high-dimensional data modeling. They are widely used in many disciplines such as astronomy and genetics. One reason for their popularity is their flexibility and straightforward implementation. The interest increase in multivariate mixtures and their applicability when they are combined with clustering methods motivated us to opt for these methods to analyse financial markets' dynamics. An empirical investigation of a set of interest rate time series is performed using a new methodology. The objective of these analyses is to determine the similarities and specificities among the analysed financial time series. Interest rates are subsequently classified according to their behavior. Such a study has been widely exploited for stock price changes using traditional methods, whereas interest rates have been less considered. The major advantage of cluster analysis is that it gives the study more realism, since it represents a functional relationship between independent and dependent variables. The EM algorithm improves clustering in the sense that it considers a stochastic relationship among variables taking missing data into account.