Know your Limits: Minimal Requirements for Pareto/NBD Distributed Data Sets
(A2019-8372)

AUTHORS

Lydia Simon, University of Duisburg-Essen; Jost Adler, University of Duisburg-Essen

KEYWORDS

Pareto NBD model; customer base analysis; synthetic data

ABSTRACT

The Pareto/NBD model is one of the best-known models in customer base analysis. When being applied to empirical data sets, the true distribution and parameters of the data are obviously not known. Here arises a lack in research on how many customers or which duration of calibration are necessary to achieve reliable estimation results. Furthermore, it was never examined how well the underlying parameters can be recovered at all. Therefore, we perform a broad simulation study on Pareto/NBD distributed data sets to find minimal requirements for its usage. The results show that a calibration period between 18 months and five years with at least 5,000 customers is recommended. Furthermore, the heterogeneity parameters are overrated which leads to an overestimation in the expected values of the purchase and dropout rate.