arXiv subjects
Publications and source records attributed to Eilon Solan.
We study approachability theory in the presence of constraints. Given a repeated game with vector payoffs, we characterize the pairs of sets (A,D) in the payoff space such that Player 1 can guarantee that the long-run average payoff converges to the set A, while the average payoff always remains in D.
We prove that every multiplayer quitting game admits a sunspot $\varepsilon$-equilibrium for every $\varepsilon > 0$, that is, an $\varepsilon$-equilibrium in an extended game in which the players observe a public signal at every stage. We also prove that if a certain matrix that is derived from the payoffs in the game is a $Q$-matrix in the sense of linear complementarity problems, then the game admits a Nash $\varepsilon$-equilibrium for every $\varepsilon > 0$.
We present a new tool for the study of multiplayer stochastic games, namely the modified game, which is a normal-form game that depends on the discount factor, the initial state, and for every player a partition of the set of states and a vector that assigns a real number to each element of the partition. We study properties of the modified game, like its equilibria, min-max value, and max-min value. We then show how this tool can be used to prove the existence of a uniform equilibrium in a certain class of multiplayer stochastic games.
A set $A$ in a finite dimensional Euclidean space is \emph{monovex} if for every two points $x,y \in A$ there is a continuous path within the set that connects $x$ and $y$ and is monotone (nonincreasing or nondecreasing) in each coordinate. We prove that every open monovex set as well as every closed monovex set is contractible, and provide an example of a nonopen and nonclosed monovex set that is not contractible. Our proofs reveal additional properties of monovex sets.
This paper presents a new solution concept for multiplayer stochastic games, namely, acceptable strategy profiles. For each player $i$ and state $s$ in a stochastic game, let $w_i(s)$ be a real number. A strategy profile is \emph{$w$-acceptable}, where $w=(w_i(s))$, if the discounted payoff to each player $i$ at every initial state $s$ is at least $w_i(s)$, provided the discount factor of the players is sufficiently close to 1. Our goal is to provide simple strategy profiles that are $w$-acceptable for payoff vectors $w$ in which all coordinates are high.
We provide a full characterization of the set of value functions of Markov decision processes.
In repeated games, cooperation is possible in equilibrium only if players are sufficiently patient, and long-term gains from cooperation outweigh short-term gains from deviation. What happens if the players have incomplete information regarding each other's discount factors? In this paper we look at repeated games in which each player has incomplete information regarding the other player's discount factor, and ask when full cooperation can arise in equilibrium. We provide necessary and sufficient conditions that allow full cooperation in equilibrium that is composed of grim trigger strategies, and characterize the states of the world in which full cooperation occurs. We then ask whether these "cooperation events" are close to those in the complete information case, when the information on the other player's discount factor is "almost" complete.
Bandit problems model the trade-off between exploration and exploitation in various decision problems. We study two-armed bandit problems in continuous time, where the risky arm can have two types: High or Low; both types yield stochastic payoffs generated by a Levy process. We show that the optimal strategy is a cut-off strategy and we provide an explicit expression for the cut-off and for the optimal payoff.
We study a dynamic model of information provision. A state of nature evolves according to a Markov chain. An informed advisor decides how much information to provide to an uninformed decision maker, so as to influence his short-term decisions. We deal with a stylized class of situations, in which the decision maker has a risky action and a safe action, and the payoff to the advisor only depends on the action chosen by the decision maker. The greedy disclosure policy is the policy which, at each round, minimizes the amount of information being disclosed in that round, under the constraint that it maximizes the current payoff of the advisor. We prove that the greedy policy is optimal in many cases -- but not always.
Two concepts of random stopping times in continuous time have been defined in the literature, mixed stopping times and randomized stopping times. We show that under weak conditions these two concepts are equivalent, and, in fact, that all types of random stopping times are equivalent. We exhibit the significance of the equivalence relation between stopping times using stopping problems and stopping games. As a by-product we extend Kuhn's Theorem to stopping games in continuous time.
We introduce the concept of attainable sets of payoffs in two-player repeated games with vector payoffs. A set of payoff vectors is called {\em attainable} if player 1 can ensure that there is a finite horizon $T$ such that after time $T$ the distance between the set and the cumulative payoff is arbitrarily small, regardless of what strategy player 2 is using. This paper focuses on the case where the attainable set consists of one payoff vector. In this case the vector is called an attainable vector. We study properties of the set of attainable vectors, and characterize when a specific vector is attainable and when every vector is attainable.
Three notions of random stopping times exist in the literature. We introduce two concepts of equivalence of random stopping times, motivated by optimal stopping problems and stopping games respectively. We prove that these two concepts coincide and that the three notions of random stopping times are equivalent.
Typically, models with a heterogeneous property are considerably harder to analyze than the corresponding homogeneous models, in which the heterogeneous property is replaced with its average value. In this study we show that any outcome of a heterogeneous model that satisfies the two properties of \emph{differentiability} and \emph{interchangibility}, is $O(ε^2)$ equivalent to the outcome of the corresponding homogeneous model, where $ε$ is the level of heterogeneity. We then use this \emph{averaging principle} to obtain new results in queueing theory, game theory (auctions), and social networks (marketing).
Typically, models with a heterogeneous property are considerably harder to analyze than the corresponding homogeneous models, in which the heterogeneous property is replaced with its average value. In this study we show that any outcome of a heterogeneous model that satisfies the two properties of differentiability and interchangibility is O(ε^2) equivalent to the outcome of the corresponding homogeneous model, where εis the level of heterogeneity. We then use this averaging principle to obtain new results in queueing theory, game theory (auctions), and social networks (marketing).
We consider a dynamic version of sender-receiver games, where the sequence of states follows an irreducible Markov chain observed by the sender. Under mild assumptions, we provide a simple characterization of the limit set of equilibrium payoffs, as players become very patient. Under these assumptions, the limit set depends on the Markov chain only through its invariant measure. The (limit) equilibrium payoffs are the feasible payoffs that satisfy an individual rationality condition for the receiver, and an incentive compatibility condition for the sender.
We prove that every two-player non-zero-sum Dynkin game in continuous time admits an epsilon-equilibrium in randomized stopping times. We provide a condition that ensures the existence of an epsilon-equilibrium in non-randomized stopping times.
We study repeated games played by players with bounded computational power, where, in contrast to Abreu and Rubisntein (1988), the memory is costly. We prove a folk theorem: the limit set of equilibrium payoffs in mixed strategies, as the cost of memory goes to 0, includes the set of feasible and individually rational payoffs. This result stands in sharp contrast to Abreu and Rubisntein (1988), who proved that when memory is free, the set of equilibrium payoffs in repeated games played by players with bounded computational power is a strict subset of the set of feasible and individually rational payoffs. Our result emphasizes the role of memory cost and of mixing when players have bounded computational power.
We study a class of two-player repeated games with incomplete information and informational externalities. In these games, two states are chosen at the outset, and players get private information on the pair, before engaging in repeated play. The payoff of each player only depends on his `own' state and on his own action. We study to what extent, and how, information can be exchanged in equilibrium. We prove that provided the private information of each player is valuable for the other player, the set of sequential equilibrium payoffs converges to the set of feasible and individually rational payoffs as players become patient.