Modeling and Evaluating Insurance Losses Via Mixtures of Erlang Distributions

In this paper we suggest the use of mixtures of Erlang distributions with common scale parameter to model insurance losses. A modified expectation-maximization (EM) algorithm for parameter estimation tailored to this class of distributions is presented, and its computation efficiency is discussed. Goodness-of-fit tests are performed for data generated from some common parametric distributions and for catastrophic loss data in the United States. Formulas for value-at-risk and conditional tail expectation are provided for individual and aggregate losses.