📈 economics

Strategic Exit and Unilateral Control

This paper demonstrates that in repeated games where opponents can unilaterally decide to exit after each round, global payoff control collapses into local control, forcing controllers to use memoryless strategies and reducing enforceable relations to a single dimension linking payoffs to expected game duration rather than payoffs alone.

Original authors: Alexander Kangas

Published 2026-07-27

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Alexander Kangas

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are playing a game of chess, but with a twist: after every single move, your opponent gets to decide if the game continues or if they just walk away. In the world of game theory—the study of how people make decisions when their outcomes depend on others—this changes everything. Usually, players assume the game will go on for a long time, allowing them to use "long-term strategies." They might take a small loss today, hoping to make a huge profit tomorrow, trusting that the relationship will last long enough for the math to work out. This is the logic behind "repeated games," where players try to find a balance that keeps them both happy over time. But what happens when the other person holds the "off" switch? Can you still force a specific outcome if your partner can quit whenever they feel like it? This is the big question: Does the power to control a relationship vanish the moment the other person can walk away, or is there a way to stay in charge even when the future is uncertain?

This paper, written by Alexander Kangas, dives into that exact puzzle. It asks: If you are trying to control the total score of a game, but your opponent can stop the game at any moment after seeing the results, what kind of control do you actually have left? The author finds that the old way of playing—where you sacrifice a little now to win big later—completely falls apart. You can't rely on "future promises" to balance out "current losses" because your opponent might quit before those promises come true. Instead, the paper proves that the only way to maintain control is to make sure every single round is fair and balanced on its own, right then and there. It's like trying to keep a scale balanced: if your opponent can pull the rug out from under you at any second, you can't rely on a heavy weight you plan to add later. You have to make sure the scale is perfectly balanced with the weights you have right now.

The paper shows that in a classic game called the Prisoner's Dilemma (where two people have to decide whether to cooperate or betray each other), this rule is strict. If you want to enforce a specific relationship between your total score, your opponent's total score, and how long the game lasts, you have to play the exact same way in every single round. You can't change your strategy based on what happened before. The only "magic formula" that survives this kind of strategic quitting is a specific equation that links your total points, your opponent's total points, and the expected number of rounds played. Interestingly, the paper proves you cannot control just the points alone; the length of the game must be part of the deal. If you try to force a rule that only cares about the score without caring about how long the game lasts, your opponent can simply quit to break your rule.

To understand this, think of a "long-term strategy" like a student promising to study hard all week to get an A on a final exam. If the teacher (the opponent) can decide to cancel the exam after Tuesday, the student's promise to study on Wednesday and Thursday becomes useless. The teacher might say, "I'll stop the game right after Tuesday," and the student's plan to balance their effort over time fails. The paper argues that the only way to guarantee a result in this scenario is to make sure the student is studying just enough every single day to get the grade they want, regardless of whether the exam happens tomorrow or next year. You have to be perfect in the present moment because the future is no longer guaranteed.

The author uses math to prove that if you want to enforce a rule where the total rewards are linked in a specific way, every single round of the game must be "neutralized" locally. This means that no matter what your opponent does in that specific round, the expected gain or loss for you is exactly zero. It's like a tightrope walker who can't rely on a safety net that might be pulled away; they have to be perfectly balanced at every single step. The paper shows that in the Prisoner's Dilemma, this forces the controlling player to use a "mixed" strategy—basically, flipping a coin to decide whether to cooperate or defect with a fixed probability every time. They can't get clever and change their mind based on the history of the game.

One of the most surprising findings is that you cannot control the total score without also controlling the duration of the game. The paper shows that any rule you try to enforce will inevitably include the number of rounds played as a key ingredient. It's like trying to bake a cake where the recipe says, "The taste depends on the ingredients and how long you bake it." You can't just say, "The taste depends only on the ingredients," because if you stop baking too early, the cake is raw. In this game, the "baking time" (the duration) is inextricably linked to the "taste" (the payoff). The paper provides a specific formula for the Prisoner's Dilemma with scores of 5, 3, 1, and 0, showing that the relationship between the players' scores and the game length is fixed and unbreakable, as long as the controlling player sticks to their fixed strategy.

In short, the paper reveals that when the other person holds the power to quit, the only way to stay in control is to be consistent and local. You can't play the long game if the game might end tomorrow. You have to make every single move count on its own, balancing the scales in the present moment. This turns the idea of "unilateral control" (one player dictating the terms) from a complex, long-term dance into a simple, repetitive rhythm. The paper doesn't just suggest this; it proves it mathematically, showing that any attempt to rely on future cancellations or complex history-dependent tricks will fail against a smart opponent who knows how to walk away. The result is a clear, rigid rule: to control the outcome, you must control the present, round by round, and accept that the length of the game is part of the price you pay for that control.

Technical Summary: Strategic Exit and Unilateral Control

Problem Statement
The paper addresses the limits of unilateral control in repeated games when the continuation of the relationship is not fixed or exogenous, but rather a strategic decision made by the opponent after observing each round's outcome. While existing literature on Zero-Determinant (ZD) and autocratic strategies (e.g., Press and Dyson, 2012; Hilbe et al., 2013) demonstrates that a player can impose global linear payoff relations regardless of an opponent's behavior, these results rely on the assumption that the relationship continues with a fixed probability or schedule. This allows the controller to balance positive rewards at some histories against negative rewards at later histories (intertemporal cancellation).

The central question is: What forms of unilateral control survive when the opponent can choose to exit the relationship at any point based on the realized public history? The paper investigates whether a player can enforce additive linear identities involving cumulative payoffs and expected duration against an opponent who controls both actions and the horizon of the interaction.

Methodology
The analysis is conducted within a framework of two players, $X$ and $Y$ , playing a finite stage game repeatedly.

Model Structure: Players draw actions from behavioral strategies based on public pre-action history $H_t$ . After observing the action profile, both players simultaneously decide whether to continue. The game ends ( $\tau$ ) if at least one player chooses to exit. The analysis focuses on the set of opponent strategies $\Sigma^{fin}_Y(\sigma_X)$ that result in a finite expected duration ( $E\tau < \infty$ ).
Control Definition: Player $X$ is said to enforce a coefficient vector $\theta = (\alpha, \beta, \gamma)$ if the expected cumulative controlled increment $\sum_{t=1}^\tau g_\theta(a^t_X, a^t_Y)$ is zero for all opponent strategies in $\Sigma^{fin}_Y(\sigma_X)$ , where $g_\theta$ is a linear combination of stage payoffs and a constant.
The Strategic-Stopping Principle: The core methodological contribution is a "one-more-round" comparison argument. The paper compares two opponent strategies: one that exits immediately before a specific reachable history $h$ , and another that reaches $h$ , plays a specific pure action $b$ , and then exits. Because the identity must hold for both, the difference in expected controlled rewards isolates the conditional expected increment of that single round.

Key Contributions and Results

The Strategic-Stopping Principle (Theorem 1):
The paper establishes that a cumulative linear identity is enforceable against every finite-duration opponent strategy if and only if the conditional expected stage increment vanishes at every reachable history.
- Implication: Global control via intertemporal cancellation is impossible under strategic exit. Control must be local. The controller cannot rely on future corrections because the opponent can selectively truncate the relationship before those corrections occur.
- Mathematical Condition: For every reachable history $h$ and every pure opponent action $b$ , the expected controlled increment $G_b(x_h)$ must be zero, where $x_h$ is the controller's mixed action at $h$ .
Behavioral Unconditionality (Corollary 2):
In games where the set of mixed actions satisfying the local zero-condition is a singleton (a unique local equalizer), the controller must use the same mixed action at every reachable history.
- This implies that "unconditional" behavior is not an assumption but a necessary consequence of robust unilateral control in the face of strategic exit. The controller cannot condition their strategy on history to maintain the identity; they must play a constant mixture.
Application to the Prisoner's Dilemma (Section 5):
Applying these results to the standard Prisoner's Dilemma with payoffs $(T, R, P, S) = (5, 3, 1, 0)$ :
- If player $X$ cooperates with a constant probability $p$ at every reachable history, they enforce the identity:
  $(1 + p)U_X + (4 - p)U_Y + 5(p^2 - 3p - 1)E\tau = 0$
- Duration is Indispensable: The coefficient of the expected duration term, $5(p^2 - 3p - 1)$ , is strictly negative for all $p \in [0, 1]$ . Consequently, no non-trivial payoff-only cumulative identity (where the duration coefficient is zero) can be enforced.
- Resulting Relation: The enforceable relation is strictly one-dimensional, linking a weighted sum of cumulative payoffs to the expected duration of the relationship. Longer expected relationships correspond to proportionally larger controlled payoff aggregates.
Two-Action Games (Section 4):
The paper characterizes the coefficient space for general two-action games. It shows that while payoff-duration control is generally available via unconditional mixed actions, payoff-only control is exceptional and requires specific determinant conditions on the payoff matrix that are not met in the standard Prisoner's Dilemma.

Significance and Claims
The paper claims to characterize the robustness of unilateral control when the horizon of the game is endogenous and controlled by the opponent.

Shift in Control Mechanism: The primary significance is the shift from "global" control (relying on telescoping sums and occupation measures over a fixed horizon) to "local" control (requiring the expected increment to be zero at every step).
Limitation of Autocracy: Strategic exit rules out the possibility of enforcing payoff-only relations in the Prisoner's Dilemma. The controller cannot dictate a specific ratio of payoffs independent of how long the relationship lasts.
Robustness: The results hold for arbitrary history-dependent opponent strategies, provided the expected duration is finite. The paper explicitly distinguishes its findings from models of exogenous continuation (where survival weights are fixed) and equilibrium selection models, focusing solely on the existence of enforceable identities under strategic stopping.

The paper concludes that while strategic exit eliminates the ability to enforce complex, history-dependent payoff relations, it leaves a specific, robust class of identities intact: those that are locally balanced at every step, which in the Prisoner's Dilemma manifests as a linear trade-off between cumulative payoffs and expected relationship duration.

Technical Summary: Strategic Exit and Unilateral Control

More like this