Forecasting Warranty Claims Considering Dynamic Over-Dispersion

Forecasting warranty claims are vitally important for manufacturers in preparing their fiscal plans as well as in managing their inventory. One of the widely used forecasting models is the non-homogeneous Poisson process (NHPP), which assumes that the mean and the variance of the numbers of warranty claims at any given time interval are equal. However, this is not always the case. Warranty claim data often exhibit a phenomenon known as over-dispersion, which implies that the variance to mean ratio is larger than one. Furthermore, this ratio might change over time and can have a trend or a clearly discernible functional form, which has not yet been considered in the existing literature on warranty claims forecasting. This paper presents a warranty claim forecasting approach that tackles the problem of the dynamic over-dispersion exhibited in warranty claims data. It considers the application of both mixed NHPP and Cox process models to warranty claims and assumes that the intensity of the mixed NHPP follows a gamma distribution and the intensity of the Cox process follows a gamma process. Warranty claim data collected from an electronics product manufacturer are used validate the models, which show that these models outperform conventional NHPP models. © 2012 Elsevier B.V. All rights reserved.