Distributionally robust multi-item newsvendor problem with covariate information

Abstract

This dissertation investigates robust optimization for use in demand forecasting. Techniques of robust optimization such as construction of ambiguity set, robust counterpart and affine recourse approximation are carefully studied. In addition, we’ve included the use of machine-learning technique in our ambiguity set construction and evaluated methods of machine-learning such as K-means Clustering and Classification and Regression Tree (CART). In our project, we considered the problem of a manager selling multiple product in a single period model. We evaluated cases where the seller considers/include uncertain covariates and/or cross-price elasticity using two different linear decision rule, i.e. Partial Affine Recourse Approximation (PARA) and Full Affine Recourse Approximation (FARA).