Market Response Modeling
A preview of the dataset
Goodbelly is a line of probiotic juice products produced by Colorado based Next Foods, Inc. The product was launched in January of 2008 and is available nationwide at retailers such as Whole Foods Market and Safeway, Inc. In May-July of 2010 the firm spent money on in store product demonstrations in select Whole Foods regions. Goodbelly is trying to understand what, if any, impact the in store demonstrations might have had on sales and profitability.
To help address these questions we have data from the Rocky Mountain (RM) and Northeast Regions (NE). The data include units sold, retail price, a dummy code for demo (which is a 1 if the demo occurred in a particular store within each regain in a particular week) and a dummy code for demo1-3 which is a 1 if a store had a demo 1, 2, or 3 weeks ago.
First, I estimated sales response models for the RM and NE regions separately, specifying sales as a linear function of price, demo, and demo1-3.
SO WHAT?A current demo has a stronger effect on current sales than previous demos.
An increase in price has a significant and negative effect on sales in both regions.
Both demo and demo1-3 has a significant and positive effect on sales in both regions.
In order to provide statistical guidance on whether or not the price coefficient and demo/demo1-3 coefficients are statistically different across regions, I created these models to consider a dummy variable for RM vs NE and the interaction of the dummy with Price, Demo, and Demo 1-3.
Model 1 includes the NE Dummy.
Model 2 includes the NE dummy and the NE*Price interaction term. In both models, the NE dummy is significant and negative. The NE*Price dummy is significant and positive, indicating RM stores are more price sensitive relative to NE stores.
Model 3 shows the Demo and Demo 1-3 effects do not differ between RM and NE.
Model 4 includes interactions for NE with price, demo, and demo 1-3 which show that the demo effects do not differ between RM and NE stores. Price is not significant at the 95% confidence level but it is very close and significant at the 90% confidence level.
Based on this I chose Model 2.
Above is the use of Model 2 to test for the effect of running demos in all stores in the RM and NE markets for the week of July 20, 2010. The model includes the dynamic effects of the demonstrations estimated to be present in the data. To zero in on the demos, the retail weekly price in each store was assumed to remain the same at July 13, 2010 for the week of July 20 and the following weeks. Assuming 30% margin for retail and 50% margin for manufacturing,
the expected profits were $39,175.77.
If the model did not account for dynamic demo effects, the profit analysis would have underestimated the predicted profits by thousands.
