Iterated Admissibility does not refine Extensive-Form Rationalizability

Emiliano Catonini

DOI: https://doi.org/10.1093/ej/ueae032

2024-04-18

The Economic Journal

Abstract:Abstract I show that, in an extensive-form game, the outcomes that are consistent with Iterated Admissibility (i.e., with the iterated elimination of weakly dominated strategies) need not be consistent with Extensive-form Rationalizability.

economics

What problem does this paper attempt to address?

The Robustness of Extensive-Form Rationalizability

Jing Chen

2011-01-01

Abstract:As proved by Shimoji and Watson, a strategy of an extensive-form game is rationalizable in the sense of Pearce if and only if it survives the maximal elimination of conditionally dominated strategies. Briefly, this is a process of iteratively eliminating conditionally dominated strategies according to a specific order. We prove, however, that there is nothing special about this order, and that the notion of extensive-form rationalizability (EFR) is actually quite robust from an algorithmic point of view. Indeed, although possibly very different from each other, the sets of strategies surviving two arbitrary elimination orders are always equivalent in a very strong sense. That is, each strategy si of a player i surviving the first order can be identified with a strategy si of i surviving the second order, so that for every strategy profile (s1, . . . , sn) the corresponding profile (s1, . . . , s ′ n) not only generates the same payoff profile, but actually the same terminal node. To prove our results we put forward a new notion of dominance and an elementary characterization of EFR that may be of independent interest. We establish connections between EFR and other existing iterated strategy-elimination procedures, using our notion of dominance and our characterization of EFR.
Forcing Iterated Admissibility in Strategic Belief Models

Fernando Tohmè,Gianluca Caterina,Rocco Gangle

DOI: https://doi.org/10.48550/arXiv.1205.0444

2012-05-02

Logic

Abstract:Iterated admissibility (IA) can be seen as exhibiting a minimal criterion of rationality in games. In order to make this intuition more precise, the epistemic characterization of this game-theoretic solution has been actively investigated in recent times: it has been shown that strategies surviving m+1 rounds of iterated admissibility may be identified as those that are obtained under a condition called rationality and m assumption of rationality (RmAR) in complete lexicographic type structures. On the other hand, it has been shown that its limit condition, $R\infty AR$, might not be satisfied by any state in the epistemic structure, if the class of types is complete and the types are continuous. In this paper we introduce a weaker notion of completeness which is nonetheless sufficient to characterize IA in a highly general way as the class of strategies that indeed satisfy $R\infty AR$. The key methodological innovation involves defining a new notion of generic types and employing these in conjunction with Cohen's technique of forcing.
The Order Independence of Iterated Dominance in Extensive Games

Jing Chen,Silvio Micali

DOI: https://doi.org/10.3982/te942

2013-01-01

Theoretical Economics

Abstract:Shimoji and Watson (1998) prove that a strategy of an extensive game is rationalizable in the sense of Pearce if and only if it survives the maximal elimination of conditionally dominated strategies. Briefly, this process iteratively eliminates conditionally dominated strategies according to a specific order, which is also the start of an order of elimination of weakly dominated strategies. Since the final set of possible payoff profiles, or terminal nodes, surviving iterated elimination of weakly dominated strategies may be order-dependent, one may suspect that the same holds for conditional dominance.We prove that, although the sets of strategy profiles surviving two arbitrary elimination orders of conditional dominance may be very different from each other, they are equivalent in the following sense: for each player i and each pair of elimination orders, there exists a function phi(i) mapping each strategy of i surviving the first order to a strategy of i surviving the second order, such that, for every strategy profile s surviving the first order, the profile (phi(i)(s(i)))(i) induces the same terminal node as s does.To prove our results, we put forward a new notion of dominance and an elementary characterization of extensive-form rationalizability (EFR) that may be of independent interest. We also establish connections between EFR and other existing iterated dominance procedures, using our notion of dominance and our characterization of EFR.
Rationalizing Irrational Beliefs

Geoffrey Dunbar,Ruqu Wang,Xiaoting Wang

DOI: https://doi.org/10.4236/tel.2016.66115

2016-01-01

Theoretical Economics Letters

Abstract:In this paper we propose a “behavioral equilibrium” definition for a class of dynamic games of perfect information. We document various experimental studies of the Centipede Game in the literature that demonstrate that players rarely follow the subgame perfect equilibrium strategies. Although some theoretical modifications have been proposed to explain the outcomes of the experiments, we offer another: players can choose whether or not to believe that their opponents use subgame perfect equilibrium strategies. We define a “behavioral equilibrium” for this game; using this equilibrium concept, we can reproduce the outcomes of those experiments.
Uncoupled Bandit Learning towards Rationalizability: Benchmarks, Barriers, and Algorithms

Jibang Wu,Haifeng Xu,Fan Yao

2023-12-23

Abstract:Under the uncoupled learning setup, the last-iterate convergence guarantee towards Nash equilibrium is shown to be impossible in many games. This work studies the last-iterate convergence guarantee in general games toward rationalizability, a key solution concept in epistemic game theory that relaxes the stringent belief assumptions in both Nash and correlated equilibrium. This learning task naturally generalizes best arm identification problems, due to the intrinsic connections between rationalizable action profiles and the elimination of iteratively dominated actions. Despite a seemingly simple task, our first main result is a surprisingly negative one; that is, a large and natural class of no regret algorithms, including the entire family of Dual Averaging algorithms, provably take exponentially many rounds to reach rationalizability. Moreover, algorithms with the stronger no swap regret also suffer similar exponential inefficiency. To overcome these barriers, we develop a new algorithm that adjusts Exp3 with Diminishing Historical rewards (termed Exp3-DH); Exp3-DH gradually forgets history at carefully tailored rates. We prove that when all agents run Exp3-DH (a.k.a., self-play in multi-agent learning), all iteratively dominated actions can be eliminated within polynomially many rounds. Our experimental results further demonstrate the efficiency of Exp3-DH, and that state-of-the-art bandit algorithms, even those developed specifically for learning in games, fail to reach rationalizability efficiently.

Computer Science and Game Theory,Machine Learning,Multiagent Systems
The Complexity of Admissibility in Omega-Regular Games

Romain Brenguier,Jean-François Raskin,Mathieu Sassolas

DOI: https://doi.org/10.48550/arXiv.1304.1682

2014-01-24

Abstract:Iterated admissibility is a well-known and important concept in classical game theory, e.g. to determine rational behaviors in multi-player matrix games. As recently shown by Berwanger, this concept can be soundly extended to infinite games played on graphs with omega-regular objectives. In this paper, we study the algorithmic properties of this concept for such games. We settle the exact complexity of natural decision problems on the set of strategies that survive iterated elimination of dominated strategies. As a byproduct of our construction, we obtain automata which recognize all the possible outcomes of such strategies.

Computer Science and Game Theory
Quantal Response Equilibrium and Rationalizability: Inside the Black Box

Shuige Liu,Fabio Maccheroni

2021-06-30

Abstract:This paper aims to connect epistemic and behavioral game theory by examining the epistemic foundations of quantal response equilibrium (QRE) in static games. We focus on how much information agents possess about the probability distributions of idiosyncratic payoff shocks, in addition to the standard assumptions of rationality and common belief in rationality. When these distributions are transparent, we obtain a solution concept called $\Delta^p$-rationalizability, which includes action distributions derived from QRE; we also give a condition under which this relationship holds true in reverse. When agents only have common belief in the monotonicity of these distributions (for example, extreme value distributions), we obtain another solution concept called $\Delta^M$-rationalizability, which includes action distributions derived from rank-dependent choice equilibrium, a parameter-free variant of QRE. Our solution concepts also provide insights for interpreting experimental and empirical data.

Theoretical Economics
Conditional dominance in games with unawareness

Martin Meier,Burkhard C. Schipper

2023-04-27

Abstract:Heifetz, Meier, and Schipper (2013) introduced dynamic game with unawareness consisting of a partially ordered set of games in extensive form. Here, we study the normal form of dynamic games with unawareness. The generalized normal form associated with a dynamic game with unawareness consists of a partially ordered set of games in norm form. We use the generalized normal form to characterize extensive-form rationalizability (resp., prudent rationalizability) in dynamic games with unawareness by iterated conditional strict (resp., weak) dominance in the associated generalized normal form. We also show that the analogue to iterated admissibility for dynamic games with unawareness depends on extensive-form structure. This is because under unawareness, a player's information set not only determines which nodes she considers possible but also of which game tree(s) she is aware of.

Computer Science and Game Theory
Coherence without Rationality at the Zero Lower Bound

Guido Ascari,Sophocles Mavroeidis,Nigel McClung

2023-10-31

Abstract:Standard rational expectations models with an occasionally binding zero lower bound constraint either admit no solutions (incoherence) or multiple solutions (incompleteness). This paper shows that deviations from full-information rational expectations mitigate concerns about incoherence and incompleteness. Models with no rational expectations equilibria admit self-confirming equilibria involving the use of simple mis-specified forecasting models. Completeness and coherence is restored if expectations are adaptive or if agents are less forward-looking due to some information or behavioral friction. In the case of incompleteness, the E-stability criterion selects an equilibrium.

General Economics
Logic and Interactive RAtionality Yearbook 2012

Z. Christoff,P. Galeazzi,Nina Gierasimczuk,A. Marcoci,S. Smets,J. Benthem,Chanjuan Liu,Fenrong Liu,J. Eijck,A. Baltag,J. Bergfeld,K. Kishida,J. Sack,S. Smets,S. Zhong,Branden Fitelson,R. Parikh,M. Kaneko,V. Hendricks,Peter Hawke,Tai-Wei Hu

2014-01-01

Abstract:The theories of Nash noncooperative solutions and of rationalizability intend to describe the same target problem of ex ante individual decision making, but they are distinctively different. We consider what their essential difference is by giving a unified approach and parallel derivations of their resulting outcomes. Our results show that the only difference lies in the use of quantifiers for each player’s predictions about the other’s possible decisions; the universal quantifier for the former and the existential quantifier for the latter. Based on this unified approach, we discuss the statuses of those theories from the three points of views: Johansen’s postulates, the free-will postulate vs. complete determinism, and prediction/decision criteria. One conclusion we reach is that the Nash theory is coherent with the free-will postulate, but we would meet various difficulties with the rationalizability theory.
Set-rationalizable choice and self-stability

Felix Brandt,Paul Harrenstein

DOI: https://doi.org/10.1016/j.jet.2011.03.006

IF: 1.79

2011-07-01

Journal of Economic Theory

Abstract:Rationalizability and similar notions of consistency have proved to be highly problematic in the context of social choice, as witnessed by a range of impossibility results, among which Arrowʼs is the most prominent. We propose to rationalize choice functions by preference relations over sets of alternatives (set-rationalizability) and introduce two consistency conditions, αˆ and γˆ, which are defined in analogy to Senʼs α and γ. We find that a choice function satisfies αˆ if and only if it is set-rationalizable and that it satisfies αˆ and γˆ if and only if it is self-stable, a new concept based on earlier work by Dutta. The class of self-stable social choice functions contains a number of appealing Condorcet extensions.

economics
Self-confirming Games: Unawareness, Discovery, and Equilibrium

Burkhard C. Schipper

DOI: https://doi.org/10.4204/EPTCS.251.35

2017-07-27

Abstract:Equilibrium notions for games with unawareness in the literature cannot be interpreted as steady-states of a learning process because players may discover novel actions during play. In this sense, many games with unawareness are "self-destroying" as a player's representation of the game must change after playing it once. We define discovery processes where at each state there is an extensive-form game with unawareness that together with the players' play determines the transition to possibly another extensive-form games with unawareness in which players are now aware of actions that they have previously discovered. A discovery process is rationalizable if players play extensive-form rationalizable strategies in each game with unawareness. We show that for any game with unawareness there is a rationalizable discovery process that leads to a self-confirming game that possesses an extensive-form rationalizable self-confirming equilibrium. This notion of equilibrium can be interpreted as steady-state of a learning and discovery process.

Computer Science and Game Theory,Multiagent Systems
An equilibrium analysis of the Arad-Rubinstein game

Christian Ewerhart,Stanisław Kaźmierowski

DOI: https://doi.org/10.1016/j.jebo.2024.106696

IF: 2

2024-08-21

Journal of Economic Behavior & Organization

Abstract:Colonel Blotto games with discrete strategy spaces nicely illustrate the intricate nature of multidimensional strategic reasoning. This paper studies the equilibrium set of such games where, in line with prior experimental work, the tie-breaking rule is allowed to be flexible. We begin by pointing out that equilibrium constructions known from the literature extend to our class of games. However, we also note that, irrespective of the tie-breaking rule, the equilibrium set is excessively large. Specifically, any pure strategy that allocates at most twice the fair share to each battlefield is used with positive probability in some equilibrium. Furthermore, refinements based on the elimination of weakly dominated strategies prove ineffective.

economics
Erratum: "rational dynamics and epistemic logic in games"

JOHAN VAN BENTHEM

DOI: https://doi.org/10.1142/s0219198907001485

2007-01-01

International Game Theory Review

Abstract:Game-theoretic solution concepts describe sets of strategy profiles that are optimal for all players in some plausible sense. Such sets are often found by recursive algorithms like iterated removal of strictly dominated strategies in strategic games, or backward induction in extensive games. Standard logical analyses of solution sets use assumptions about players in fixed epistemic models for a given game, such as mutual knowledge of rationality. In this paper, we propose a different perspective, analyzing solution algorithms as processes of learning which change game models. Thus, strategic equilibrium gets linked to fixed-points of operations of repeated announcement of suitable epistemic statements. This dynamic stance provides a new look at the current interface of games, logic, and computation.
Level-k thinking in the extensive form

Burkhard C. Schipper,Hang Zhou

DOI: https://doi.org/10.1007/s00199-024-01556-x

2024-03-07

Economic Theory

Abstract:Level- k thinking has been widely applied as a solution concept for games in normal form in behavioral and experimental game theory. We consider level-k thinking in games in extensive form. Player's may learn about levels of opponents' thinking during the play of the game because some information sets may be inconsistent with certain levels. In particular, for any information set reached, a level- k player attaches the maximum level- thinking for to her opponents consistent with the information set. We compare our notion of strong level- k thinking with other solution concepts such as level- k thinking in the associated normal form, strong rationalizability, -rationalizability, iterated admissibility, backward rationalizability, backward level- k thinking, and backward induction. We use strong level- k thinking to reanalyze data from some prior experiments in the literature.

economics
Conservative Rationalizability and the Second-Knowledge Mechanism

Jing Chen,Silvio Micali

2010-01-01

Abstract:In mechanism design, the traditional way of modeling the players’ incomplete information about their opponents is “assuming a Bayesian.” This assumption, however, is very strong and does not hold in many real applications. Accordingly, we put forward (1) a set-theoretic way to model the knowledge that a player might have about his opponents, and (2) a new class of mechanisms capable of leveraging such more conservative knowledge in a robust way. In auctions of a single good, we show that such a new mechanism can perfectly guarantee a revenue benchmark (always lying in between the second highest and the highest valuation) that no classical mechanism can even approximate in any robust way. ∗This work is funded by the Office of Naval Research under award number N00014091059. (Any opinions, findings, and conclusions or recommendations expressed in this email are those of the author (s) and do not necessarily reflect the views of the Office of Naval Research.)
A Theory of Bounded Inductive Rationality

Caspar Oesterheld,Abram Demski,Vincent Conitzer

DOI: https://doi.org/10.4204/EPTCS.379.33

2023-07-11

Abstract:The dominant theories of rational choice assume logical omniscience. That is, they assume that when facing a decision problem, an agent can perform all relevant computations and determine the truth value of all relevant logical/mathematical claims. This assumption is unrealistic when, for example, we offer bets on remote digits of pi or when an agent faces a computationally intractable planning problem. Furthermore, the assumption of logical omniscience creates contradictions in cases where the environment can contain descriptions of the agent itself. Importantly, strategic interactions as studied in game theory are decision problems in which a rational agent is predicted by its environment (the other players). In this paper, we develop a theory of rational decision making that does not assume logical omniscience. We consider agents who repeatedly face decision problems (including ones like betting on digits of pi or games against other agents). The main contribution of this paper is to provide a sensible theory of rationality for such agents. Roughly, we require that a boundedly rational inductive agent tests each efficiently computable hypothesis infinitely often and follows those hypotheses that keep their promises of high rewards. We then prove that agents that are rational in this sense have other desirable properties. For example, they learn to value random and pseudo-random lotteries at their expected reward. Finally, we consider strategic interactions between different agents and prove a folk theorem for what strategies bounded rational inductive agents can converge to.

Artificial Intelligence,Computer Science and Game Theory,Machine Learning
Equilibrium Behaviors in Repeated Games

Yingkai Li,Harry Pei

DOI: https://doi.org/10.48550/arXiv.2007.14002

2021-02-11

Abstract:We examine a patient player's behavior when he can build reputations in front of a sequence of myopic opponents. With positive probability, the patient player is a commitment type who plays his Stackelberg action in every period. We characterize the patient player's action frequencies in equilibrium. Our results clarify the extent to which reputations can refine the patient player's behavior and provide new insights to entry deterrence, business transactions, and capital taxation. Our proof makes a methodological contribution by establishing a new concentration inequality.

Theoretical Economics
Discovery and Equilibrium in Games with Unawareness

Burkhard Schipper

DOI: https://doi.org/10.48550/arXiv.2109.05145

2021-09-11

Abstract:Equilibrium notions for games with unawareness in the literature cannot be interpreted as steady-states of a learning process because players may discover novel actions during play. In this sense, many games with unawareness are "self-destroying" as a player's representation of the game may change after playing it once. We define discovery processes where at each state there is an extensive-form game with unawareness that together with the players' play determines the transition to possibly another extensive-form game with unawareness in which players are now aware of actions that they have discovered. A discovery process is rationalizable if players play extensive-form rationalizable strategies in each game with unawareness. We show that for any game with unawareness there is a rationalizable discovery process that leads to a self-confirming game that possesses a self-confirming equilibrium in extensive-form rationalizable conjectures. This notion of equilibrium can be interpreted as steady-state of both a discovery and learning process.

Computer Science and Game Theory

Iterated Admissibility does not refine Extensive-Form Rationalizability

The Robustness of Extensive-Form Rationalizability

Forcing Iterated Admissibility in Strategic Belief Models

The Order Independence of Iterated Dominance in Extensive Games

Rationalizing Irrational Beliefs

Uncoupled Bandit Learning towards Rationalizability: Benchmarks, Barriers, and Algorithms

The Complexity of Admissibility in Omega-Regular Games

Quantal Response Equilibrium and Rationalizability: Inside the Black Box

Conditional dominance in games with unawareness

Coherence without Rationality at the Zero Lower Bound

Logic and Interactive RAtionality Yearbook 2012

Set-rationalizable choice and self-stability

Self-confirming Games: Unawareness, Discovery, and Equilibrium

An equilibrium analysis of the Arad-Rubinstein game

Erratum: "rational dynamics and epistemic logic in games"

Level-k thinking in the extensive form

Conservative Rationalizability and the Second-Knowledge Mechanism

A Theory of Bounded Inductive Rationality

Equilibrium Behaviors in Repeated Games

Discovery and Equilibrium in Games with Unawareness