Abstract
We study the convergence of Nash equilibria in a game of optimal stopping. If the associated mean field game has a unique equilibrium, any sequence of $n$-player equilibria converges to it as $n\to\infty$. However, both the finite and infinite player versions of the game often admit multiple equilibria. We show that mean field equilibria satisfying a transversality condition are limit points of $n$-player equilibria, but we also exhibit a remarkable class of mean field equilibria that are not limits, thus questioning their interpretation as "large $n$" equilibria.
Introduction
In game theory, solving $n$-player games becomes exponentially difficult as the number of participants increases—a phenomenon often called the "curse of dimensionality." Mean Field Games (MFGs) were developed to circumvent this by modeling an infinite number of identical, negligible agents. The underlying assumption is that an MFG solution serves as a reliable approximation for games with a sufficiently large but finite number of players.
However, many strategic scenarios—particularly those involving strategic complementarity, such as bank runs—exhibit multiple equilibria. This raises a fundamental stability question: do all equilibria found in the infinite-limit MFG actually correspond to the behavior of large finite groups? This paper investigates whether some MFG solutions are merely "mathematical ghosts"—artifacts of the infinite limit that cannot be reached by any sequence of finite-player games.
The Optimal Stopping Model
The authors utilize a case study of an optimal stopping game. In this model, $n$ players observe idiosyncratic signals and must decide when to "stop." A critical feature is the presence of positive externalities: as more players choose to stop, the incentive for the remaining players to stop increases.
In the mean field limit, the discrete interactions of $n$ players are replaced by a representative agent reacting to the statistical distribution of the entire population. The equilibrium of this system is characterized by the equation $u + F_t(r - cu) = 1$, where $u$ represents the proportion of players who have stopped.
Classification of Equilibria
The core of the analysis lies in classifying the solutions to the equilibrium equation. By examining the function $g_t(u)$, the authors identify three distinct types of crossings at the equilibrium line:
- Increasing-transversal ($i$): The function crosses the equilibrium line with a positive slope.
- Decreasing-transversal ($d$): The function crosses with a negative slope.
- Tangential ($t$): The function merely touches the equilibrium line without crossing it.
These classifications are vital because the "slope" at the crossing point determines the physical reality and stability of the equilibrium in the large-$n$ limit.
Convergence and "Ghost" Equilibria
The research establishes rigorous convergence theorems to distinguish between valid and invalid MFG equilibria.
- Stable Limits: Theorems 5.4 and 6.1 prove that "increasing-transversal" equilibria are guaranteed to be limits of $n$-player Nash equilibria. These represent robust behaviors that large groups of agents are likely to adopt.
- Mathematical Ghosts: Theorem 6.6 provides a striking counter-result: "strongly decreasing-transversal" equilibria are never the limit points of $n$-player games. Despite being mathematically valid solutions in the MFG framework, they do not represent the behavior of large finite populations.
- Emergent Randomization: The study also finds that $n$-player games may not converge to a single deterministic MFG equilibrium. Instead, they can converge to a mixture (a probability distribution) over multiple MFG equilibria. This explains how collective "herding" can lead to unpredictable macro-outcomes even when individual signals are clear.
Numerical simulations using Monte Carlo methods with $n=10,000$ players support these findings. In scenarios with a "tent-shaped" signal density (Example 6.5), the $n$-player games cluster around stable points (0 and 1), while the unstable MFG equilibrium (at 0.5) is rarely reached, often appearing only as part of a mixture.
Conclusion
This work highlights that not all solutions to Mean Field Games are created equal. Transversality serves as a litmus test for the physical relevance of an equilibrium. While MFGs remain a powerful tool for approximating large-scale systems, practitioners must be cautious of "ghost" equilibria that lack a finite-player counterpart.
Future research may extend these insights to broader classes of MFGs, such as those involving complex drift/diffusion control or "common noise"—shocks that affect all agents simultaneously—which could further shift the stability landscape of these strategic interactions.
Summary updated