The cost of decentralized persuasion

Abstract

This thesis delves into the exploration of cost of decentralization in systems with exogenous and endogenous states as well as stochastic formulations within the context of the Scheduled Service Network Design Problem (SSNDP). It is structured around three key chapters, each offering unique insights into these complex dynamics.In the initial chapter, the focus is on understanding the cost of decentralization in a service system with exogenous states. The system comprises multiple service centers, each with a state that is independent from each other and unobservable to the customer. An arriving customer should make entry decisions to the system at the time of her entry relying on signals received from the system about its state. Notably, the system state is determined externally and remains unaffected by the internal dynamics of the system, such as customer’s entry decision. The chapter introduces inventive formulations for centralized and decentralized signaling problems, discusses optimal signaling policies for both scenarios, and presents a method for solving the non-linear formulation of the decentralized signaling problem. The chapter concludes by evaluating the cost of decentralization in the described system, establishing an upper bound on that and introduces a system design capable of attaining this upper bound. Moving to the second chapter, the research extends from systems characterized by exogenous states to those featuring endogenous states. Unlike exogenous system, here the state is determined by internal dynamics of the system. Introducing a unique system featuring multiple identical components each with a binary state, the chapter explores optimal signaling policies, with a specific focus on a decentralized signaling mechanism where each component sends a binary signal to the customers. Customers arrive dynamically over time, and upon arrival, they receive a signal about the system state. This prompts customers to decide whether to join a component or leave without engaging, thus contributing to the system’s endogenous evolution. Despite the initial intent of studying this system to identify conditions under which the cost of decentralization is high, the findings reveal that, compared to exogenous systems, it is relatively small. This prompts a critical question: whether the cost of decentralization in systems with endogenous states can be high, a question that remains to be answered and suggests a direction for future research efforts. The final chapter introduces two novel stochastic formulations for the Scheduled service network design problem, SSNDP, including the stochastic time-expanded and consolidation- based formulations. These formulations address the challenges of managing a network of Less Than Truckload (LTL) carriers that are transporting commodities with uncertain sizes throughout the network. A comparative analysis demonstrates the superiority of the stochastic consolidation-based formulation in terms of efficiency and ease of solution.