Dominick's Price Optimization

Ayan Saraf, Bo Suk Yoon, Lazuli Abel, Peter Laughlin MSBA 46894 M3 - Team G B. Store Model: The store model estimates separate parameters for each store, allowing for store-specific effects on sales. Assumption: each potential buyer is, on average, different for each store but identical within that store’s market. C. Mixed Model: The mixed model combines fixed effects that are common across all stores and random, store-specific effects. Assumption: some plain potential buyer’s habits can be slightly tweaked to reflect all potential buyers across all store markets. D. Improved Model: Built from the hierarchical Mixed Model, this model includes the feature and display variables for products 4, 5, and 11. Assumption: features and displays of competitive products will affect consumer behavior in a statistically significant way. We expect an inverse relationship between price and quantity sold. This relationship would be captured by negative price coefficients. If products are substitutes, an increase in the price of one product should lead to an increase in the quantity sold of the other product. This relationship is reflected in positive cross-price elasticities between substitutes in all models. Investigating Model Error For each model, we interpret the three error types as follows: - Mean Error: On average, do the predictions match the actual values? If so, expect 0. - Root Mean Squared Error (RMSE): When predictions are incorrect, is the distance between the predicted and actual values small? If so, expect a value close to 0. - Mean Absolute Error: When predictions are incorrect, is the distance between the predicted and actual values small? If so, expect a value close to 0 and below RMSE. Single Pooled Store Mixed Improved Training Test Training Test Training Test Training Test Mean Error 0.00 0.00 0.00 0.00 0.00 0.005 0.00 0.00 RMSE 0.48 0.47 0.41 0.52 0.45 0.46 0.443 0.446 Mean Abs Error 0.58 0.57 0.38 0.37 0.35 0.35 0.340 0.345 Based on predictive accuracy, the Improved Model emerges as the best-performing model for forecasting orange juice sales because it exhibits the lowest RMSE and Mean 3