Repeated Multimarket Contact with Private Monitoring: A Belief-Free Approach

Atsushi Iwasaki, Tadashi Sekiguchi, Shun Yamamoto, Makoto YokooAAAI 2020 · OpenReview

Abstract

This paper studies repeated games where two players play multiple duopolistic games simultaneously (multimarket contact). A key assumption is that each player receives a noisy and private signal about the other's actions (private monitoring or observation errors). There has been no game-theoretic support that multimarket contact facilitates collusion or not, in the sense that more collusive equilibria in terms of per-market profits exist than those under a benchmark case of one market. An equilibrium candidate under the benchmark case is belief-free strategies. We are the first to construct a non-trivial class of strategies that exhibits the effect of multimarket contact from the perspectives of simplicity and mild punishment. Strategies must be simple because firms in a cartel must coordinate each other with no communication. Punishment must be mild to an extent that it does not hurt even the minimum required profits in the cartel. We thus focus on two-state automaton strategies such that the players are cooperative in at least one market even when he or she punishes a traitor. Furthermore, we identify an additional condition (partial indifference), under which the collusive equilibrium yields the optimal payoff.

Introduction

In both economics and artificial intelligence, competition is frequently modeled as a Repeated Prisoners’ Dilemma (PD). A common real-world phenomenon is "multimarket contact," where firms (e.g., ride-sharing services or airlines) compete across multiple geographic or product markets simultaneously. Empirical evidence suggests that such contact facilitates collusion, keeping prices high. However, classical game theory—specifically the work of Bernheim and Whinston (1990)—suggests that under perfect monitoring, multimarket contact does not enhance the ability to collude beyond what is possible in a single market.

This paper addresses this theoretical gap by introducing private monitoring. In realistic settings, firms cannot perfectly observe a rival's actions; they instead receive noisy signals, such as their own sales volumes. Analyzing "belief-free" equilibria in these settings—where a player's optimal action is independent of their belief about the opponent's history—becomes increasingly complex as the number of markets grows.

Methodology: The gEV Strategy

The authors utilize a belief-free approach to construct strategies for two-player, $M$-market repeated games. The core of their approach is a two-state automaton, which simplifies the coordination problem inherent in cartels where communication is prohibited.

Two-State Automaton and Benchmarks

The strategies are defined using a simple machine with two states: Cooperation (R) and Punishment (P). To establish a baseline, the authors reference the standard single-market benchmark known as the EV strategy (Ely-Välimäki).

The single-market benchmark (EV strategy) represented as a two-state automaton.

The Generalized EV (gEV) Strategy

The paper's primary contribution is the Generalized EV (gEV) strategy. Unlike the linear transition probabilities in standard models, gEV employs a non-linear approach. The probability of switching from cooperation to punishment depends on the specific combination of signals across markets.

The proposed multimarket gEV strategy for two markets (M=2), illustrating transitions based on joint signals.

Key innovations in the gEV strategy include:

  • Partial Indifference: The $M$ markets are divided into two sets (Set $A$ and Set $B$). The equilibrium is constructed such that a player is indifferent among actions in Set $B$ markets, provided they remain cooperative in Set $A$.
  • Mild Punishment: Instead of a "grim trigger" that collapses all cooperation, gEV uses mild punishment where players continue to cooperate in some markets even while punishing in others.
  • Signal Non-linearity: Transition probabilities jump sharply if "bad" signals are received from specific subsets of markets, effectively managing the temptation to defect.

Key Results and Insights

The theoretical analysis yields several significant findings:

  • Payoff Improvement: Theorem 2 proves that the gEV strategy achieves a higher per-market average payoff than the best-known single-market strategies (EV and sEV).
  • Threshold for Effect: The multimarket contact effect becomes mathematically significant when the number of markets $M$ is greater than three.
  • Optimal Payoffs: Under the "partial indifference" condition, the gEV equilibrium yields the optimal payoff within its class of belief-free strategies.
  • Simplicity and Robustness: By focusing on two-state strategies, the authors ensure the model remains relevant to real-world cartels that must coordinate without explicit communication.

Limitations and Future Work

While providing a mathematical "missing link" for the multimarket contact effect, the study has limitations:

  1. Symmetry: The model assumes identical (symmetric) markets. Future research will explore asymmetric markets with varying sizes and profitabilities.
  2. Market Selection: The division of markets into cooperative and punishment sets is currently fixed; optimizing this selection is a goal for future work.
  3. Global Optimality: While gEV outperforms benchmarks, it is not yet proven to be the absolute optimal strategy among all possible complex equilibria.

Summary updated

Abstract

This paper studies repeated games where two players play multiple duopolistic games simultaneously (multimarket contact). A key assumption is that each player receives a noisy and private signal about the other's actions (private monitoring or observation errors). There has been no game-theoretic support that multimarket contact facilitates collusion or not, in the sense that more collusive equilibria in terms of per-market profits exist than those under a benchmark case of one market. An equilibrium candidate under the benchmark case is belief-free strategies. We are the first to construct a non-trivial class of strategies that exhibits the effect of multimarket contact from the perspectives of simplicity and mild punishment. Strategies must be simple because firms in a cartel must coordinate each other with no communication. Punishment must be mild to an extent that it does not hurt even the minimum required profits in the cartel. We thus focus on two-state automaton strategies such that the players are cooperative in at least one market even when he or she punishes a traitor. Furthermore, we identify an additional condition (partial indifference), under which the collusive equilibrium yields the optimal payoff.
BibTeX
@inproceedings{iwasaki2020repeated,
  author = {Atsushi Iwasaki and Tadashi Sekiguchi and Shun Yamamoto and Makoto Yokoo},
  title = {{Repeated Multimarket Contact with Private Monitoring: A Belief-Free Approach}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2020},
  url = {https://openreview.net/forum?id=1gOxu1_PxPp},
}
Figure