Title: Estimating an Inverse Gamma distribution

Abstract: In this paper we introduce three different algorithms based on method of moments, maximum likelihood and full Bayesian estimation for learning the parameters of the Inverse Gamma distribution. We also provide an expression for the KL divergence for Inverse Gamma distributions which allows us to quantify the estimation accuracy of each of the algorithms. All the presented algorithms are novel. We consider most relevant the introduction of a conjugate prior for the Inverse Gamma shape parameter which enables the Bayesian inference procedure; in order to compute expectations under the proposed distribution we use the Laplace approximation. The introduction of this new prior allows analytical inference of the parameter estimation problem and also opens the possibility of including Inverse Gamma distributions into more complex Bayesian structures, e.g. mixture models. The algorithms introduced in this paper are computationally compared using synthetic data and interesting relationships between the maximum likelihood and the Bayesian approach are derived.