Demand estimation and pricing of perishable food

Demand estimation of perishable food products is challenging because freshness is typically not recorded in available data. To address this measurement challenge, I collect novel customer-transaction and batch-level-production panels—enabling direct observation of freshness for each individual unit in both consumer purchases and choice sets—from the prepared fried-food category of a Japanese grocery chain. With this unique data, I study consumer willingness-to-pay for freshness, its joint heterogeneous distribution with price sensitivity, and grocers’ optimal pricing by freshness using improved demand estimation. I build and estimate a model of consumer batch choice (i.e., freshness) for perishable goods, capturing heterogeneous freshness preferences and price sensitivity. I find that consumer demand for freshness is substantial, with an additional hour of freshness for an average product equivalent to 4.0% of its full price, but freshness preferences also correlate with price sensitivity, implying an absence of purely price-driven consumers. I then model the retailer’s dynamic markdown problem—trading-off discount losses against future stockout losses—and show that optimal markdowns can be zero under observed heterogeneity and the production process. I implement the proposed policy in the focal store; the structural model predicts a focal profit increase of 2.24% (95% posterior predictive interval [1.91, 2.60]), and a field implementation yields a profit increase of about 6.3%, consistent with this prediction. My findings can explain cross-grocer differences in pricing-by-freshness policies as a function of consumer heterogeneity.