Unique Ergodicity in the Interconnections of Ensembles with Applications to Two-Sided Markets

Ramen Ghosh, Wynita M. Griggs, Jakub Marecek, Robert N. Shorten

Abstract

This paper introduces a novel discrete-time feedback framework for analyzing the long-term stability and fairness of multi-sided markets, such as ride-sharing platforms (e.g., Uber, Lyft). By modeling agent populations as ensembles and utilizing Iterated Random Functions (IRFs), the authors establish mathematical conditions for unique ergodicity. This property ensures that the system converges to a unique invariant probability measure, providing a rigorous foundation for predictable resource allocation and market fairness that is independent of initial conditions.

Introduction

The rapid growth of the gig economy has made two-sided and multi-sided platforms ubiquitous. These markets connect distinct groups of participants, such as workers and customers, through centralized digital platforms. However, analyzing the long-term stability, predictability, and fairness of these systems is challenging due to the non-linear and uncertain responses of human agents to platform signals like surge pricing. This research addresses a fundamental control-theory question: Under what conditions can a platform guarantee that a market stabilizes and allocates resources fairly over the long run, regardless of the system's starting state?

Modeling Multi-Sided Markets

The authors represent the market as a closed-loop, discrete-time dynamical system. The primary components are "ensembles" of agents (e.g., populations of drivers and riders), central controllers (pricing algorithms), and filters. To account for the inherent randomness in human behavior, the framework utilizes Iterated Random Functions (IRFs) with state-dependent probabilities. While IRFs are a class of Markov processes less common in traditional economics, they are highly effective for establishing stability guarantees in stochastic systems.

Block diagram of a two-sided market model consisting of two interconnected ensembles of agents.

As shown in the figure above, the two-sided market is modeled as two interconnected feedback loops where the output of one ensemble (e.g., rider demand) influences the controller of the other (e.g., driver supply pricing).

Establishing Unique Ergodicity

The central mathematical contribution of this work is the proof of unique ergodicity for these interconnected systems. A system is uniquely ergodic if it converges to a unique, attractive invariant probability measure. In practical terms, this means the long-term average behavior of the system stabilizes and becomes independent of the initial state of the agents.

Generalized block diagram of the interconnected system used for ergodicity analysis.

The authors prove that if agent responses satisfy an "average contractivity" condition—ensuring the system state remains bounded—the entire interconnected market will be uniquely ergodic. This result holds for linear time-invariant systems, non-linear systems, and systems with discrete finite states. This property provides a mathematical definition of fairness: every agent is guaranteed a predictable long-term average allocation of resources, regardless of their initial advantages or starting state.

Results and Simulations

The theoretical proofs were validated using simulated "toy examples" rather than empirical datasets:

  • Fixed Demand Simulation: In a two-sided market with 150 drivers and a fixed demand of 120 rides, the system stabilized effectively. Controllers adjusted prices to meet demand, keeping the error signal bounded near zero over 1,800 time steps.
  • Elastic Demand Simulation: When demand was made elastic—where prices influence both rider demand and driver supply—the system again converged predictably, successfully balancing the competing populations.

A key finding is that the choice of controller is critical; for instance, a standard Proportional-Integral (PI) controller can destroy ergodicity, whereas a lag approximant preserves it.

Discussion and Future Work

While the model provides a robust theoretical framework, it currently assumes a constant number of registered participants and abstracts away spatial complexities like driver relocation. Future research aims to design "ergodic controls" specifically built to deliver guaranteed quality of service across large-scale networks, bridging the gap between macroscopic fluid approximations and microscopic agent-level models.

Summary updated

Abstract

There has been much recent interest in two-sided markets and dynamics thereof. In a rather a general discrete-time feedback model, which we show conditions that assure that for each agent, there exists the limit of a long-run average allocation of a resource to the agent, which is independent of any initial conditions. We call this property the unique ergodicity. Our model encompasses two-sided markets and more complicated interconnections of workers and customers, such as in a supply chain. It allows for non-linearity of the response functions of market participants. Finally, it allows for uncertainty in the response of market participants by considering a set of the possible responses to either price or other signals and a measure to sample from these.
BibTeX
@article{ghosh2025unique,
  author = {Ramen Ghosh and Wynita M. Griggs and Jakub Marecek and Robert N. Shorten},
  title = {{Unique Ergodicity in the Interconnections of Ensembles with Applications to Two-Sided Markets}},
  journal = {Int. J. Control},
  year = {2025},
  doi = {10.1080/00207179.2025.2469281},
}
Figure