How to Predict Marketplace Demand Quantities Using Volumetric Choice Experiments

Abstract

Demand predictions are important for product and pricing decisions. Brand managers in packaged goods categories need to understand drivers of both primary and secondary demand in order to increase sales.

While there are many models dedicated to explaining secondary demand, we focus on the challenge of accurate primary demand predictions.

Primary demand is not only influenced by the composition, but also the size of a choice set. Specifically, larger choice sets typically lead to larger levels of primary demand, because they allow consumers to satisfy their demand for variety. Demand models for packaged goods need to be able to explain demand quantities across choice sets of different sizes, as assortment sizes are subject can vary substantially across time or between stores.The divergence in choice set size is even greater when brand managers need to predict marketplace demand using choice experiments.

While choice experiments involve choices from relatively small choice sets with 2-10 choice alternatives, marketplace offerings typically comprise 100 or more unique products or stock-keeping units. Extant models either ignore or exaggerate the effect of choice-set size on demand.

We propose a new model that allows heterogeneous response to varying choice-set sizes. In our empirical application in the chocolate bar category, we find that our model significantly improves predictions of marketplace demand quantities. A common benchmark volumetric demand model overpredicts marketplace demand by a factor of almost 2, our model enables realistic marketplace demand predictions.

We validate our predictions using information about aggregate sales at the market and brand level.

Keywords: Choice Models, Demand Analysis, Volumetric Demand, Multiple Discrete Continuous Models, Bayesian Estimation

JEL Classification: M3, C8, C9