Alan Montgomery

Department of Marketing

Wharton School

University of Pennsylvania

Retailers face an immense and complicated problem in determining profitable pricing strategies. Typically a retailer is selling products in scores of stores operating in a regional chain and not simply in a single store. This paper shows how pricing policies can be customized to the store level, rather than adopting a uniform pricing policy for all stores. This customization process is known as micro-marketing. The full decision-theoretic profit maximization problem is posed and solved to reveal the value of micro-marketing pricing strategies.

This paper takes a general approach to modeling brand competition by constructing systems of demand equations for each category. We apply these demand models to store-level optical scanner data for the refrigerated orange juice category from Dominick's Finer Foods. (Dominick's is the second largest supermarket chain in the Chicago metropolitan area.) This modeling approach to price response at the store level avoids imposing specific substitution patterns on a category, but results in a general form with a huge number of parameters. This creates a formidable estimation problem since it becomes difficult to reliably measure store-level differences in price sensitivity.

Two extreme approaches to these measurement problems are to allow each store to have its own model or to pool all stores together. Individual store models uncover a great deal of differences between stores. But they have poor predictive ability and the standard errors for the parameter estimates are too large to be useful for setting pricing strategies. At the other extreme, the pooled chain-level model has fair predictive ability, but it ignores the heterogeneity across stores.

To improve our estimates we use cross-store information in a hierarchical Bayesian model. Shrinkage techniques allow the estimates to fall between the extremes of either pooling all stores or using individual store models. Formally this is a random coefficient model in which a component of the parameter variation is related to demographic and competitive characteristics of the store's trading area. To estimate this model we use Gibbs sampling, a recent advance in statistical computing, to compute the exact finite-sample posterior distribution of model parameters. An added benefit of the Gibbs Sampler is that we can easily compute the posterior distributions of the profit function. These posterior profit distributions fully reflect the uncertainty in the parameter estimates.

The price elasticity estimates are validated by comparison to actual randomized field experiments. In addition we compare the out-of-sample performance of the hierarchical model estimates to the individual linear models and find a 20% reduction in the mean-squared error of the predictions. The pooled model estimates, in contrast, improve the MSE by half this amount. Also the hierarchical model can show that the ``characteristics for the store's trading area''(1-interpretation of effects - relationship between national our price - effects of % of houses) are interesting predictors fo the store's price sensitivity profile. Stores in areas with larger families and more elderly consumers tend to be more price sensitive, while those stores located in areas with more expensive homes tend to be less price sensitive.

The results presented in this paper show that improved pricing strategies can increase chain-wide profits by over 20%. To better reflect competitive pressures and consumer responses, we introduce constraints. These constraints require total revenue and average price in each store to equal the values produced under a uniform pricing strategy. If these constraints are satisfied then we can find new micro-marketing strategies that retain the properties of the current uniform pricing strategy, which increases profits by 3 to 4%. The typical supermarket retailer's profit margin is less than 3% after allowing for selling costs. When we compare these micro-marketing profit increases to current profit returns, we find that micro-marketing pricing strategies could have a large impact upon retailer profitability.