Global Games With Noisy Information Sharing

Hessam Mahdavifar, Ahmad Beirami, Behrouz Touri, Jeff S. Shamma

TL;DR

This paper investigates "global games" where agents share information in a noisy social network. They demonstrate that a common "intuitive" threshold policy, where agents act based solely on their belief about an underlying fundamental, fails to be an equilibrium when information is shared. However, they prove that a generalized threshold policy does exist and is unique under certain noise conditions, offering a new framework for understanding coordination in noisy environments.

What Problem Does This Paper Address?

Global games are a powerful tool for modeling situations where agents make decisions based on incomplete information about an underlying "fundamental" (like the strength of a political regime or a currency's value). In these games, agents receive noisy private signals about this fundamental. A key finding in simpler versions of global games is the existence of a "threshold policy" equilibrium: agents act based on whether their belief about the fundamental crosses a certain threshold.

This paper tackles a crucial extension: what happens when agents share their noisy observations with each other before making their decisions? This is a more realistic scenario for social networks and collaborative systems. The authors question whether the established threshold policy results still hold when this information sharing occurs, especially since the sharing itself is noisy. They aim to understand if and how equilibria can be established in such complex, interconnected information environments.

Methods

The core of the paper involves:

  • Challenging the Intuitive Policy: The authors first mathematically demonstrate that a simple threshold policy, where an agent's decision depends only on their own updated belief about the fundamental (derived from their private signal and shared noisy information), does not generally lead to a Bayesian Nash Equilibrium in the presence of noisy information sharing. This is a significant departure from prior work.
  • Generalizing Threshold Policies: To overcome this, they introduce a more general notion of a "threshold policy" that can depend on a multi-dimensional information vector (the agent's private signal plus noisy observations of others).
  • Functional Equations and Fixed Points: The existence of these generalized threshold policies is framed as solving a functional equation. They leverage the Banach fixed-point theorem, a powerful mathematical tool, to prove the existence of a unique solution (a threshold function) under specific conditions. This involves defining a "contraction mapping" that, when iterated, converges to the desired equilibrium strategy.
  • Lipschitz Continuity: To ensure the convergence and stability of their iterative method, they analyze the Lipschitz continuity of the functions involved, establishing conditions on the noise parameters (σ and τ) that guarantee this property.

Datasets

Not discussed in the paper. The paper focuses on theoretical modeling and analysis rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • Fragility of Simple Thresholds: The most striking insight is that the intuitive threshold policy, so effective in simpler global games, breaks down when agents share information. This highlights the critical role of the information structure in determining equilibrium properties.
  • Noise as a Stabilizer (in a sense): Counter-intuitively, the paper shows that sufficiently noisy information sharing can actually enable the existence of a unique, generalized threshold equilibrium. This suggests that a moderate level of noise can prevent coordination failures that might arise from perfect, but potentially misleading, information sharing.
  • Iterative Convergence: The authors provide a constructive proof for the existence of the generalized threshold policy by showing that an iterative process (applying a contraction mapping) converges to the equilibrium strategy. This not only proves existence but also offers a practical method for approximating the equilibrium.
  • Generalization is Key: The paper emphasizes that to find equilibria in these more complex settings, one must move beyond simple threshold policies based on individual beliefs and consider more sophisticated strategies that account for the shared, noisy information.

Key Results

  • Non-existence Theorem: The paper formally proves that intuitive threshold policies (based on an agent's expected value of the fundamental given their information) do not constitute a Bayesian Nash Equilibrium in global games with noisy information sharing.
  • Existence and Uniqueness Theorem: They establish that a symmetric, generalized threshold policy equilibrium does exist and is unique under certain conditions on the noise parameters (σ and τ) and the number of agents (n). This condition is characterized by the noise variance parameters satisfying a specific inequality (wn < τ).
  • Convergence Guarantee: The iterative method proposed is shown to converge to the unique threshold function, providing an approximation mechanism for finding the equilibrium strategy.

Limitations and Future Work

  • Sufficient, Not Necessary Conditions: The paper acknowledges that the derived conditions for the existence of the threshold-type equilibrium policy are sufficient but not necessarily necessary. Characterizing the entire set of parameters for which such equilibria exist remains an open problem.
  • Specific Noise Model: The analysis relies on additive Gaussian noise. Extending the framework to other noise distributions or models could be a future direction.
  • Static Game: The current model is static. Analyzing dynamic versions of global games with noisy information sharing, where agents update their strategies over time, would be a natural extension.
  • General Utility Functions: While the paper mentions extensions to more general utility functions, the core results are derived for a specific form. Exploring the implications for broader classes of utility functions could be valuable.

Summary updated

Abstract

Global games form a subclass of games with incomplete information, where a set of agents decide actions against a regime with an underlying fundamental θ representing its power. Each agent has access to an independent noisy observation of θ. In order to capture the behavior of agents in a social network of information exchange, we assume that agents share their observation in a noisy environment prior to making their decisions. We show that global games with noisy sharing of information do not admit an intuitive type of threshold policy, which only depends on agents' belief about the underlying θ. This is in contrast to the existing results on the threshold policy for the conventional setup of global games. Motivated by this result, we investigate the existence of equilibrium strategies in a more general collection of threshold-type policies and show that such equilibrium strategies exist and are unique if the sharing of information happens over a sufficiently noisy environment.
BibTeX
@article{mahdavifar2018global,
  author = {Hessam Mahdavifar and Ahmad Beirami and Behrouz Touri and Jeff S. Shamma},
  title = {{Global Games With Noisy Information Sharing}},
  journal = {IEEE Trans. Signal Inf. Process. Over Networks},
  year = {2018},
  url = {https://openreview.net/forum?id=cQTo1qvm9C},
}
Figure