| Abstract |
Food waste is a significant concern for grocery chains, impacting both profitability and environmental sustainability. We examine how grocery stores learn over time to improve their management of perishable product categories. We begin by deriving empirical predictions from a Bayesian model in which stores solve the newsvendor problem under econometric uncertainty, learning about demand over time. We then test these predictions using a novel dataset of Japanese supermarket transactions, inventory, and waste records across 112 newly opened stores. We document three main findings: (1) Waste rates gradually decline after store openings and new product introductions, falling 61\% after three years. (2) The decreases are driven not just by stores learning how much inventory to stock, but also about which products to stock---60\% of products are dropped within five years. (3) The decreases cannot be fully explained by demand-side learning, changes in profit margins, or dynamic markdowns. Our results suggest that stores gain significant store--product-level demand information over time---hence, investment in superior demand-prediction systems represents a significant opportunity to reduce food waste and increase profits. (joint work with Robbert Evan Sanders and Kanishka Misra) |