Flexible Decomposition of Price Promotion Effects Using Store-Level Scanner Data

We propose a model that estimates standard, enhanced, and flexible decompositions of price promotion effects based on store-level data. At present, decompositions of sales effects due to price promotions are available only for household models. These models decompose the sales elasticity into a category incidence elasticity, a brand choice elasticity and a quantity elasticity. We introduce a method that allows a unit sales effect, estimated from store data, to be decomposed into its constituent sources. The standard decomposition divides the own-brand unit sales effect into a cross-brand effect, a stockpiling effect, and a category expansion effect. The enhanced decomposition includes two extensions. One extension is the separation of the category expansion effect into a cross-store effect and a market expansion effect. This extension requires data on stores of multiple chains in geographic proximity. The other extension assumes the availability of item-level data so that a cross-item effect can be decomposed further into within-brand (cannibalization) effects and between-brand effects. For improved understanding of promotion effects, we propose flexible decompositions. We allow the decomposition to depend on the type of promotion support, such as feature and/or display. A second type of flexibility is that we allow the decomposition to depend on the magnitude of the discount. We use a flexible, nonparametric, method to accommodate the latter dependency. We apply the models to four store-level scanner data sets. The decomposition results show that, on average, the secondary demand effect, captured by the sales decrease for other brands, is about 33 percent. This contrasts strongly with the finding from models of household data that about 75 percent of the sales elasticity is attributable, on average, to the brand choice elasticity. We also find that the contribution of the cross-brand effect decreases for increasing price discounts. In addition, except for feature-supported discounts, the stockpiling effect tends to decrease as a percent of the sales effect as the discount increases, while the contribution of the 2 category expansion effect tends to increase. We find that the flexible decomposition of promotion effects, separately for four types of support, provides superior insight into marketplace phenomena. The methodology and the empirical results are relevant to marketing scientists in academia and in the practice.