Retail Multiproduct Pricing Under Demand-Model and Predictive Uncertainty: A Risk-Aware Optimization Framework

This paper develops a risk-aware optimization framework for retail multiproduct pricing when demand models disagree and their predictions remain uncertain. Using a controlled 18-week dataset for three products, linear-in-level and log-log specifications are estimated over 455 training subsets and screened on a three-week holdout. The retained models generate 1,000 admissible demand systems. The pricing objective balances expected gross margin, dispersion across systems, and predictive dispersion within systems. Thirty-six risk-aversion pairs are evaluated. The risk-neutral benchmark recommends prices (36.415, 39.589, 20.691) and expected gross margin 2205.984. Under a rule preserving at least 95% of this benchmark, the selected solution uses (λmodel, λpred) = (3, 1), recommends (31.753, 34.651, 20.862), and sacrifices 4.608% of expected margin while reducing model-disagreement dispersion by 50.501% and predictive dispersion by 19.770%. The results show that uncertainty can be incorporated directly into pricing decisions.