Joint Inventory and Markdown Management for Perishable Products with Censored Demand Data

Project: Research

View graph of relations


At the end of each sales period, inventory managers often offer price discounts on perishable products to clean up the leftover inventories. For example, bakery stores offer price discounts near the evening for those day-old unsold breads. Consequently, the sales period is composed of both the regular-sales phase and the clearance phase. Customers, on the other hand, can be divided into two types. One type highly cares about the breads' quality and is willing to pay high price for the freshness. The other type is price-sensitive and would like to check the availability of discounted products at the clearance phase to save money. Only when discounted products are stocked-out would the second-type customers consider the products at the regular price, causing one-way intertemporal substitution. Managers need to decide not only the optimal ordering quantity for perishable products in each sales period but also the amount of inventory to be marked down at the end of the regular-sales phase within each sales period. Such joint inventory and markdown decisions are explored by some researchers in recent years. However, in those studies, customer demand distribution is exogenously given. In this project, we aim to study a more realistic situation where demand distribution is unknown. Managers have to make the joint inventory and markdown decisions by learning about demand and intertemporal substitution rate through observed sales data. We plan to consider three information scenarios: fully unobservable, partially observable and censored observation. Under the first scenario, not only the realized demands for both regular-sales and clearance phases are known, but also the amount of customers who choose the regular-price products due to the stockout of clearance products (intertemporal substitution) is also known. Under the second scenario, the realized demands are known but the amount of intertemporal substitution is unknown. Under the last scenario, managers only observe the sales amount for each phase. Thus, the composition of the two types of demands for the regular sales is unknown. When stockout occurs, the lost sales are unobservable. We plan to use Bayesian updating to estimate both the demand parameters and the substitution rate. Based on that, we shall build up a dynamic programming decision model for the joint inventory and markdown decisions. This research will contribute at least in two aspects: the joint inventory and markdown decisions with demand learning and the estimation of demand and intertemporal substitution rate with censored demand data.  


Project number9043109
Grant typeGRF
Effective start/end date1/01/1922/12/21