Strategic customer behavior in a queueing system with alternating information structure

Yiannis Dimitrakopoulos, Antonis Economou, Stefanos LeonardosEur. J. Oper. Res 2021 · DOI · OpenReview

Abstract

This paper investigates the strategic behavior of customers in an M/M/1 queueing system characterized by an alternating information structure. Unlike traditional models that assume either full observability or complete opacity, this study introduces a system that switches between these two states. Using a game-theoretic framework and matrix-analytic methods, the authors identify unique equilibrium strategies and evaluate system performance. The results indicate that an intermediate level of information revelation can optimize both throughput and social welfare, often outperforming the extremes of constant transparency or total unobservability.

Introduction

In queueing theory, information availability significantly influences customer decision-making. Standard models typically focus on two extremes: observable queues, where customers know the exact queue length, and unobservable queues, where decisions are based on statistical averages. However, real-world systems often operate in an "in-between" state due to periodic updates, delayed information, or intermittent system failures.

This research addresses the gap in existing literature by proposing a unified framework for systems with alternating information states. The goal is to understand how strategic customers adapt to varying levels of information and how system administrators can leverage information flow as a control variable to enhance performance.

Methods

The system is modeled as an M/M/1 queue where the information state alternates between Mode 1 (observable) and Mode 0 (unobservable). The durations of these periods follow exponential distributions. Customers are assumed to be risk-neutral strategic actors aiming to maximize their net benefit, defined as the service value minus waiting costs and fees.

The authors define a Potentially Equilibrium Strategy (PES) characterized by three parameters: the joining threshold during observable periods ($n_e$), the threshold for staying in the queue when the system switches from unobservable to observable ($n_s$), and the probability of joining during unobservable periods ($q$).

Table of system parameters and strategy thresholds.

To analyze the system, the authors utilize Continuous-Time Markov Chains (CTMC) and solve for steady-state distributions using Quasi-Birth-Death (QBD) processes. Partial generating functions are derived to compute key performance measures such as throughput and social welfare. The study relies on theoretical proofs and numerical experiments with synthetic parameters to validate the model.

Key Insights and Ideas

The "Sweet Spot" of Ignorance

A central finding of the study is that throughput and social welfare are often unimodal functions of the information revelation rate. This suggests an "ideal" frequency of information updates that maximizes system utility. Providing too much information can lead to excessive balking in high-traffic scenarios, while providing too little can cause customers to waste time in excessively long lines.

Reneging as a Strategic Tool

The model uniquely accounts for strategic reneging. Customers who join the queue while it is hidden may choose to leave (renege) the moment the system becomes observable if they discover the queue length exceeds their threshold $n_s$. This behavior highlights the dynamic nature of customer decisions under changing information states.

Information as a Control Variable

For system managers, the rate of "announcements" (switching to the observable state) serves as a powerful lever. Increasing this rate can encourage joining by reducing uncertainty, but it also increases the risk of later abandonment.

Key Results

The authors prove the existence of a unique customer equilibrium strategy $(n_e, n_s, q_e)$ for any given set of system parameters. Numerical results demonstrate that for specific arrival rates, the alternating system achieves higher throughput and social welfare than either the fully observable or fully unobservable M/M/1 benchmarks.

Equilibrium performance measures.

Furthermore, performance measures are found to be sensitive to economic factors, generally increasing with the service valuation ($R$) and the refundability of entrance fees.

Limitations and Future Work

While the model provides a robust framework, it relies on exponential assumptions for service times and information cycles. Future research could explore more complex distributions, such as constant information update intervals. Additionally, the current model assumes a homogeneous customer population; incorporating heterogeneous valuations and waiting costs would provide further realism. Finally, investigating dynamic information structures—where the revelation of queue length is state-dependent—remains a promising avenue for future study.

Summary updated

BibTeX
@article{dimitrakopoulos2021strategic,
  author = {Yiannis Dimitrakopoulos and Antonis Economou and Stefanos Leonardos},
  title = {{Strategic customer behavior in a queueing system with alternating information structure}},
  journal = {Eur. J. Oper. Res.},
  year = {2021},
  doi = {10.1016/j.ejor.2020.10.054},
}
Figure