[Paper Review] When Knowing Early Matters: Gossip, Percolation and Nash Equilibria

David Aldous|arXiv (Cornell University)|May 26, 2010

Complex Network Analysis Techniques6 references4 citations

TL;DR

This paper studies information diffusion in networks where agents receive decreasing rewards based on their rank in learning order, modeling gossip and insider trading. Using first-passage percolation and Nash equilibrium analysis, it shows that selfish behavior leads to inefficient outcomes—ranging from wasteful to totally wasteful equilibria—depending on network structure, even though the social optimum is achieved by communicating arbitrarily slowly.

ABSTRACT

Continually arriving information is communicated through a network of $n$ agents, with the value of information to the $j$'th recipient being a decreasing function of $j/n$, and communication costs paid by recipient. Regardless of details of network and communication costs, the social optimum policy is to communicate arbitrarily slowly. But selfish agent behavior leads to Nash equilibria which (in the $n o \infty$ limit) may be efficient (Nash payoff $=$ social optimum payoff) or wasteful ($0 < $ Nash payoff $

Motivation & Objective

• To understand how selfish agent behavior in information diffusion networks leads to Nash equilibria that may deviate from the social optimum.
• To examine the efficiency of Nash equilibria in different network topologies, including complete graphs and grids with nearest-neighbor or distance-dependent communication.
• To explore how reward structures based on an agent's rank in receiving information affect strategic incentives and equilibrium outcomes.
• To analyze the role of first-passage percolation and SI epidemic models in characterizing the timing and spread of information in large networks.
• To identify conditions under which Nash equilibria are efficient, wasteful, or completely inefficient (payoff = 0) in the n→∞ limit.

Proposed method

• Models information spread via a Poisson process of new information items arriving at random agents, with communication costs paid by recipients.
• Applies a rank-based reward function R(j/n), where R(u) is decreasing, R(1)=0, and ∫₀¹ R(u)du < ∞, capturing decreasing marginal value of information.
• Analyzes the distribution of receipt times using integral equations: 1−Fθ(t) = ∫_{t−1/θ}^t ∏ᵢ(1−Fθ(s−i/θ)) θ ds for Poisson or regular call strategies.
• Uses first-passage percolation and SI epidemic models to study the 'window width'—the time interval during which most agents learn the information.
• Compares Nash equilibria under symmetric strategies (same θ for all agents) to the social optimum, where communication is slowed to near-zero cost.
• Evaluates payoff differences under Rule 2 (rank-based reward) and Rule 3 (reward per transmission), showing structural differences in equilibrium behavior.

Experimental results

Research questions

• RQ1 Under what network structures does selfish agent behavior lead to Nash equilibria that are efficient, wasteful, or completely inefficient?
• RQ2 How does the choice of communication strategy (Poisson vs. regular calls) affect the efficiency of Nash equilibria in complete graphs and grids?
• RQ3 What is the impact of reward structure—based on rank of reception versus number of transmissions—on the existence and efficiency of Nash equilibria?
• RQ4 How does the 'window width' of information percolation relate to the efficiency of Nash equilibria in different network topologies?
• RQ5 In the n→∞ limit, can the Nash equilibrium rate θ^Nash be characterized, and does it scale as N⁻¹ (efficient) or 1 (wasteful) in grid networks?

Key findings

• The social optimum is achieved by communicating arbitrarily slowly, yielding a payoff arbitrarily close to ∫₀¹ R(u)du, as communication costs can be made negligible.
• In the complete graph with rank-based rewards, the Nash equilibrium is wasteful: 0 < Nash payoff < social optimum payoff, due to strategic incentives to learn early.
• On the N×N grid with nearest-neighbor communication, the Nash equilibrium rate θ^Nash scales as N⁻¹, indicating an efficient equilibrium in the limit.
• When reward is based on number of transmissions (Rule 3), the Nash equilibrium on the grid becomes wasteful, with θ^Nash scaling as 1 (constant), not N⁻¹.
• For the N×N torus with short and long-range interactions, rigorous results from Chatterjee and Durrett confirm the asymptotic behavior predicted in this paper’s heuristic analysis.
• The analysis shows that the same qualitative behavior—efficient, wasteful, or totally wasteful equilibria—can emerge depending on network structure and reward function, even under symmetric strategies.