An alternative EM-algorithm for fitting the skew-t mixture model

An alternative EM-algorithm to obtain maximum likelihood estimates of the finite mixture of skew-t distributions is introduced. Adopting the new algorithm is simpler respect to exiting algorithms, by reducing the number of conditional expectations in E-step and the number of equations to solve in M-step. The average CPU times shows that this simplification consequence is a considerably reduction of the computational time. Some Monte Carlo simulations are presented to show the maximum likelihood estimates based on the alternative EM-algorithm, provides good asymptotic properties. We illustrate the usefulness of the proposed methodology by analyzing one real data set.

Keywords:

EM-type algorithm , Skew-t mixtures , Truncated normal distribution