讲座题目：Data-Driven Shelf-Stock Allocation
报告人： 赵嘉
中国科学院大学经济与管理学院
讲座时间：2022年11月25日(周五) 12:15-13:30
讲座地点：中国科学院大学中关村校区教学楼S406；
腾讯会议 ID：879-930-557
内容摘要
Consider a shelf-stock allocation problem where the demand distribution is unknown, yet a set of demand forecasts is available. The manager decides the placement and stock-levels of commodities on shelves to maximize the expected profit. We develop a data-driven robust model leveraging a decision-dependent Wasserstein metric that incorporates the demand forecasts and captures the vertical-location effect. Given locations, we develop a nested binary search for the shadow price and a closed-form optimal stock, leveraging a supergradient-based KKT condition. Furthermore, the optimal location can be determined by an improved L-shaped decomposition with a derived supergradient. Furthermore, impacts of ambiguity-aversion on the marginal expected profit, shadow price, and optimal stocking are analyzed. Numerical experiments with real-life sales data justify the value of our proposed approach in hedging against distributional uncertainty when the demand process heavily fluctuates and exploiting the forecasting information when the demand process is relatively stationary.
主讲人简介
赵嘉，中国科学院大学经济与管理学院博士研究生，研究方向：鲁棒优化及其应用
