Matrix Variate Regression as a Tool for Insurers

In matrix variate or three-way datasets, each single observation of a response variable constitutes a realization of a random matrix. This situation arises in time series or spatial datasets where each observation of a response contains multiple metrics measured across time or space, or when a response variable can be stratified across three different dimensions. In this paper, we present the matrix variate regression modelling approach developed in recent years, which is able to intuitively model three-way data responses given a matrix of covariates. We illustrate its potential use in P&C insurance by applying the modeling methodology to analyze the distribution of the losses from several natural hazard types over both space and time. Address for Correspondence: eshunt@asu.edu