Modeling Discrete-Time Transactions Using the BG/BB Model

For a business, it's important to know the repeat purchasing behavior of an existing customer. However, in a non-contractual setting, businesses cannot easily observe whether a customer is still active. The Beta-Geometric/Beta Binomial (BG/BB) model developed by Fader and Hardie in 2004 serves as a natural framework to model these types of behaviors. We further evaluate the BG/BB model by applying two different versions of the model to a dataset of 1975 customers over a period of 16 years to evaluate the predictive abilities of the BG/BB, and compares it with the Binary Logistic Regression model to determine whether the BG/BB model can give accurate forecasts over a long period of time.