|
Abstract:
|
We use an epidemic-type point process model to describe state-dependence, i.e. the impact of past purchases on future purchases, in a novel way. We look at household panel data that records purchase timing, quantity and brand purchased as well as marketing mix variables such as price and promotional activity for one product category. Brand choice and purchase timing are jointly dependent variables conditional on past purchases, purchase quantity, and marketing mix variables. In our model the baseline rate is expressed as a function of time-varying marketing mix variables (price, display and feature) and is brand specific. We consider both the impact of past purchases of the same brand (auto-correlated effect) and of past purchases of other brands (cross-correlated effect) on future brand choice and purchase timing. Inference is based on MCMC techniques.
|
