An application of matrix-variate mixtures to insurance data

Over the last years, there has been a growing interest in the analysis of matrix-variate data via mixture models. Quite often the tails of the matrix-variate normal distribution, used for the mixture components, are lighter than required, implying a bad fitting and the disruption of the underlying grouping structure. A solution to this issue consists in fitting mixtures of matrix-variate distributions with heavy tails. An example of such situation is here discussed by using a dataset concerning the non-life Italian insurance market. The fitting results of the matrix-variate normal mixture model are the worst, and the related data classification seems not realistic compared to the one produced by the heavy-tailed models.