A functional approach to estimation of the parameters of generalized negative binomial and gamma distributions

TL;DR

This paper introduces a novel functional approach for estimating parameters of generalized negative binomial and gamma distributions, avoiding complex grid methods and providing density estimates instead of point estimates.

Contribution

It proposes a new methodology based on minimizing $\\ell^p$-distances and $L^p$-metrics, offering a more flexible and direct way to estimate distribution parameters.

Findings

Provides a method to estimate parameters without grid algorithms.

Yields density function estimates rather than point estimates.

Applicable to a wide class of distributions including Poisson, NB, and Weibull-Poisson.

Abstract

The generalized negative binomial distribution (GNB) is a new flexible family of discrete distributions that are mixed Poisson laws with the mixing generalized gamma (GG) distributions. This family of discrete distributions is very wide and embraces Poisson distributions, negative binomial distributions, Sichel distributions, Weibull--Poisson distributions and many other types of distributions supplying descriptive statistics with many flexible models. These distributions seem to be very promising for the statistical description of many real phenomena. GG distributions are widely applied in signal and image processing and other practical problems. The statistical estimation of the parameters of GNB and GG distributions is quite complicated. To find estimates, the methods of moments or maximum likelihood can be used as well as two-stage grid EM-algorithms. The paper presents a methodology based on the search for the best distribution using the minimization of $\ell^p$-distances and $L^p$-metrics for GNB and GG distributions, respectively. This approach, first, allows to obtain parameter estimates without using grid methods and solving systems of nonlinear equations and, second, yields not point estimates as the methods of moments or maximum likelihood do, but the estimate for the density function. In other words, within this approach the set of decisions is not a Euclidean space, but a functional space.