SCRO / Journées de l'optimisation

HEC Montréal, 29-31 mai 2023


HEC Montréal, 29 — 31 mai 2023

Horaire Auteurs Mon horaire

PAOIV Pricing and Assortment Optimization IV

31 mai 2023 13h30 – 15h10

Salle: Cogeco (bleu)

Présidée par Lidiia Shchichko

4 présentations

  • 13h30 - 13h55

    Conservative Dynamic Pricing with Demand Learning

    • Amin Shahmardan, prés., DeGroote School of Business, McMaster University
    • Mahmut Parlar, DeGroote School of Business, McMaster University
    • Yun Zhou, DeGroote School of Business, McMaster University

    Dynamic pricing algorithms often lead to unexpected price changes and significant loss of revenue due to early exploration to learn the demand function. As a result, managers may prefer subjective pricing decisions over algorithmic ones.

    This paper studies dynamic pricing with demand learning, using a generalized linear demand model with unknown parameters. We develop a conservative UCB-based pricing algorithm under a stage-wise safety constraint, which requires a pricing policy's expected revenue to be at least a specific fraction of the baseline policy's revenue in each period. The algorithm minimizes the total regret, which is made up of two components: the regret of the usual UCB-based pricing algorithm, and the regret based on the number of times the baseline policy is used. We demonstrate that the second term exhibits sublinear growth and is on the order of a logarithmic scale.

    We extend the problem to a case where the expected cumulative revenue gained by the algorithm should be at least a fraction of that of the baseline policy. In this case, the algorithm implements the baseline policy for finitely many times. We show that the regret is lower than that under the stage-wise safety constraint.
    Keywords: Dynamic Pricing; Demand Learning; Conservative Pricing; Multi-Armed Bandit Problem

  • 13h55 - 14h20

    Economies of Scope Contracts to Coordinate Assortment Planning in Omni-Channel Retail Supply Chains

    • Amin Aslani, prés., Haskayne School of Business, University of Calgary
    • Osman Alp, Haskayne School of Business, University of Calgary

    We investigate the operations of an omni-channel retail supply chain (RSC) consisting of an online sales website and a physical store. First, we consider a decentralized structure in which the online sales channel is operated by a manufacturer that supplies their products to the physical retailer. The retailer makes their assortment decisions based on the wholesale price quoted by the manufacturer, through a leader-follower Stackelberg game. We also consider the effect of product returns on the profitability of the parties, which is a norm in modern retailing but also a factor in lost profit. Next, we consider the RSC in a centralized structure in which both sales channels are owned by one party. The optimal assortment and wholesale price decisions are obtained under both structures. We show that the optimal assortment decision in the centralized structure can consist of either a limited or higher variety of products compared to the decentralized structure, but the total profit of the RSC is always greater in the centralized structure. We propose (dis)economies of scope contracts that are shown to be instrumental in coordinating the RSC and eliminating double marginalization, such that each party will be better off compared to the decentralized structure.

  • 14h20 - 14h45

    Improving supply chain efficiency using the optimal combination of inventory policy and contract structure

    • Maryam Afzalabadi, prés., Lazaridis School of Business and Economics, Wilfrid Laurier University

    There are always complicated interactions in a supply chain between suppliers, vendors, and retailers. Each party tries to synchronize its demands and orders to minimize costs and maximize benefits. The situation might get more complicated when revenue is shared between two or more parties.
    In this paper, we design a revenue-sharing contract to coordinate pricing and inventory control decisions in a serial supply chain consisting of a supplier, a manufacturer, and a retailer. The retailer applies a one-for-one period policy in which he constantly orders one unit of product to the manufacturer in a predetermined time interval, resulting in a deterministic demand for the manufacturer. Solution procedures are developed to find the equilibrium solution in the Vendor Managed Inventory (VMI) program.
    Besides the supply chain structure and the kind of interactions and coordination between supply chain members, we show that the replenishment policy significantly affects supply chain efficiency. In many cases, by choosing an appropriate combination of replenishment policy and coordination mechanism, we can increase supply chain efficiency, and even it may lead to perfect coordination.

  • 14h45 - 15h10

    On the Online Demand Fulfillment in an Omnichannel Retail Chain

    • Lidiia Shchichko, prés., Concordia University
    • Satyaveer S. Chauhan, Concordia University
    • Navneet Vidyarthi, Concordia University

    In this study, we examine an online order fulfillment problem for omnichannel retailers to determine the optimal quantity of products to be shipped from each in-house facility to each customer region under congestion caused by the limited number of employees who are engaged with product packing. Congestion affects the supply chain performance, and it is modelled as a nonlinear function of the total assigned orders. We, therefore, present the "mixed-integer nonlinear program" (MINLP) and propose three different methods to solve it. The first method is to use the non-commercial solver's SCIP default algorithm. The second method is a piecewise linear approximation where the breakpoints are chosen only once – static approximation. And the last solution method is a piecewise linear approximation where the breakpoints are added iteratively – dynamic approximation. An extensive number of experiments are conducted to compare the solution quality of each method. We find that both approximation algorithms significantly outperform the non-commercial solver SCIP default settings for MINLP. The proposed model is readily extendible to the cost minimization problem if delivery and capacity costs are not to be disregarded.