A Skewed-t Mixture for Bimodal Data

We develop a mixture of location-scale skewed-t mixtures for bimodal heavy-tailed data. Skewness is introduced by means of a procedure that can be extended to any symmetric density, so that the model-building procedure is very general. We derive the main properties of the distribution and show that maximum likelihood estimation can be carried out by means of the EM algorithm. Simulation experiments and a real-data application suggest that the approach is more effective than a recently proposed mixture of g-and-h distributions.