Smart contracts and reaction-function games

Technical Summary: Smart Contracts and Reaction-Function Games

1. Problem and Motivation

The paper addresses the challenge of credible commitment in one-shot strategic interactions, particularly in environments characterized by trust deficits, coordination failures, and free-riding. Traditional game theory suggests that in one-shot games, players often fail to reach Pareto-efficient outcomes (e.g., the Prisoner's Dilemma or weakest-link coordination games) because they cannot credibly condition their actions on the actions of others.

The authors propose blockchain-based smart contracts as a mechanism to solve this. Unlike standard strategic games where players choose actions directly, smart contracts allow agents to commit to reaction functions—algorithms that prescribe an action as a function of the actions chosen by other players. Once deployed on a blockchain, these contracts are immutable and automatically execute, effectively transforming the game from a simultaneous move game into a "reaction-function game." The core problem is to formalize the equilibrium concepts for such games, determine which outcomes can be supported, and identify robust strategies that overcome the limitations of conventional Nash equilibria.

2. Methodology and Framework

The authors formalize a new class of games and solution concepts:

• Strategic Games vs. Reaction-Function Games:
  • In a standard strategic game $(N, A, (u_i))$, players choose actions $a_i \in A_i$.
  • In a reaction-function game, players choose reaction functions $R_i: A_{-i} \to A_i$. A profile $R = (R_1, \dots, R_n)$ defines a mapping from the space of action profiles to itself.
• Fixed Points and Outcomes:
  • An outcome is realized only if the chosen reaction functions are mutually consistent. Formally, an action profile $a$ is a fixed point of $R$ if $a_i = R_i(a_{-i})$ for all $i$.
  • The set of fixed points is denoted $E(R)$. If $E(R) = \emptyset$, the outcome is undefined (assigned payoff $-\infty$).
  • Players evaluate a profile $R$ based on their preferred fixed point within $E(R)$.
• Equilibrium Concept:
  • A Reaction-Function Equilibrium (RFE) is defined as an unambiguous profile $R$ where no player can unilaterally deviate to a different reaction function $R'_i$ to achieve a strictly higher payoff at the resulting fixed point.
  • Unambiguity requires that all players agree on the realized outcome (i.e., there is a Pareto-superior fixed point in $E(R)$).
• Refinements:
  • To address equilibrium multiplicity and implausible commitments, the authors introduce "Safe" reaction functions. A reaction is safe if it guarantees the player at least their maxmin payoff (security level) regardless of others' actions.
  • In symmetric investment games, the authors further refine the analysis using Norm-Proofness (resistance to conforming to arbitrary social norms) and Payoff Consistency (reacting identically to payoff-equivalent profiles).

3. Key Contributions and Results

A. Existence and Characterization of Equilibria

• Existence: Theorem 3.1 proves that a reaction-function equilibrium always exists in any finite game. This is established by mapping the reaction-function game to a sequential-move extensive-form game (where players move in a fixed order), which is known to have a Nash equilibrium via backward induction.
• Two-Player Games (Folk Theorem Analogue): Theorem 3.2 characterizes the set of supportable outcomes in two-player games. An outcome $a$ can be supported in an RFE if and only if $u_i(a) \geq \bar{v}_i$ (the maxmin payoff) for all players. This mirrors the logic of folk theorems in infinitely repeated games but applies to a one-shot setting. Players can use "promise and threat" reaction functions to enforce Pareto-efficient outcomes (like cooperation in the Prisoner's Dilemma) that are not Nash equilibria in the underlying strategic game.
• Many Players: Theorem 3.3 shows that with three or more players (depending on action space size), the set of supportable outcomes expands significantly, potentially allowing any outcome to be supported. This highlights the need for refinements like safe play to narrow down predictions.

B. Safe Reaction Functions

• The paper demonstrates that if a safe outcome (one yielding maxmin payoffs) is a Nash equilibrium in the strategic game, a safe RFE exists.
• Crucially, Theorem 4.1 states that if a safe RFE exists, a Pareto-efficient safe RFE also exists. This implies that players can achieve high-payoff outcomes without sacrificing the security of their maxmin guarantees, provided they use appropriate reaction functions (e.g., matching strategies in the Stag Hunt).

C. Application to Symmetric Investment Games
The authors apply the framework to two prominent classes of investment games, assuming monotone reaction functions (investing more as others invest more) and using smart contracts to escrow funds.

1. Weakest-Link Games:

   • Context: Payoffs depend on the minimum investment. Coordination failure is common.
   • Result: The Best-Reply (BR) reaction function ($R_i(a_{-i}) = \min_{j \neq i} a_j$) is identified as the unique norm-proof and payoff-consistent reaction.
   • High-Risk Case: In "high-risk" settings (where coordination gains are small relative to effort costs), the BR reaction is shown to be weakly dominant against monotone play. This provides a robust recommendation for players to match the minimum investment, effectively solving the trust/coordination problem without communication or central authority.

2. Public-Good Games:

   • Context: Investments benefit all but cost the individual; free-riding is dominant in standard games.
   • Result: The BR reaction (always 0) is ineffective. Instead, the authors identify reaction functions that match the average of others' investments.
   • Refinement: Between rounding up and rounding down the average, rounding down is superior. It is safe across all parameter values and supports high-investment equilibria.
   • High-Risk Case: For high-risk public-good games, the "round down" reaction is the unique welfare-maximizing strategy conditional on safe play. This strategy makes free-riding self-defeating: if everyone matches the average (rounded down), a unilateral deviation to zero investment collapses the fixed point to zero, punishing the deviator.

4. Significance and Claims

The paper claims that blockchain-based smart contracts offer a practical mechanism to implement reaction-function games, thereby extending the scope of credible commitment beyond traditional repeated-game settings.

• Overcoming Trust Barriers: In weakest-link games, reaction functions allow strangers to coordinate on efficient outcomes immediately by committing to match the minimum effort, removing the need for pre-play communication or trust.
• Disciplining Free-Riding: In public-good games, the proposed "round down" reaction function creates a self-enforcing mechanism where conditional collaboration is rational, even without altruism or external punishment institutions.
• Practicality vs. Theory: The authors distinguish their approach from "program equilibrium" (Tennenholtz, 2004) and "commitment games" (Kalai et al., 2010). While those frameworks allow players to condition actions on the code of opponents' programs (leading to infinite strategy spaces and fragility to code mismatches), reaction functions condition actions on the actions of opponents. This keeps the strategy space parsimonious, interpretable, and computationally tractable (e.g., via simple iterative fixed-point algorithms).
• Implementation: The paper outlines a practical implementation sketch where a coordinating contract collects reaction functions and deposits, computes the fixed point, and executes transfers. The authors note that for monotone functions on finite lattices, fixed-point computation is straightforward.

In summary, the paper argues that smart contracts transform the strategic landscape by allowing players to commit to conditional strategies, enabling the implementation of Pareto-improving outcomes that are otherwise unattainable in one-shot interactions, specifically through the use of safe, norm-proof, and payoff-consistent reaction functions.