Version 2 of 2
Introduction
Generated Aksbel book section. · Working · Nov 08, 2026 12:01 · saved by @mujirin
Introduction
A strategic situation is a situation in which what one person should do depends on what others do, and what others do may depend on what they expect that first person to do. This is the starting point of game theory.
Consider a simple example. Two firms sell similar products. If one firm lowers its price while the other keeps its price high, the low-price firm may attract more customers. But if both firms lower prices, both may earn less than if both had kept prices high. Each firm’s best choice depends on the other firm’s choice. This is not merely an optimization problem for one isolated decision maker. It is a problem of strategic interdependence: the outcome depends on several decision makers whose choices interact.
Game theory studies such situations by building formal models of decision-making among multiple agents. The modern mathematical development of the subject is often traced to von Neumann and Morgenstern’s Theory of Games and Economic Behavior, which presented games as formal objects for studying conflict and cooperation in economic settings (von Neumann & Morgenstern, 1944). Later work by Nash gave central form to what is now called Nash equilibrium, a condition in which no player can gain by changing strategy alone when the other players’ strategies are held fixed (Nash, 1951). These ideas now appear throughout economics, political science, biology, computer science, philosophy, law, and many other fields.
This book begins from first principles. We will not start by assuming that game theory is a collection of famous examples or clever puzzles. Instead, we will build the subject carefully: What is a player? What is an action? What is a strategy? What is a payoff? What does it mean to solve a game? What assumptions are needed before a solution concept becomes meaningful?
The purpose of this introduction is to give you a first orientation.
Why strategic reasoning is different
In many ordinary decision problems, a person chooses an action and then receives a result determined by nature, technology, or chance. For example, if you decide how many hours to study for an exam, your outcome depends partly on your effort, the difficulty of the exam, and perhaps some uncertainty. Other people may matter indirectly, but the basic reasoning can often be framed as: choose the best action given the environment.
In game theory, the environment includes other decision makers.
Suppose two students must decide independently whether to contribute to a joint project. If both contribute, the project is excellent. If one contributes and the other does not, the contributor bears most of the cost while both may still receive some credit. If neither contributes, the project is poor. Each student’s choice changes the consequences of the other student’s choice. The problem is not simply “How much effort is best?” but “What should I do, given what the other person may do, and given what the other person thinks I may do?”
This recursive structure—my reasoning about your reasoning about my reasoning—is one reason game theory is intellectually powerful and sometimes difficult. Thomas Schelling emphasized that strategic behavior often turns on expectations, commitments, threats, promises, and focal points: people do not only choose actions; they try to influence what others expect and choose (Schelling, 1960).
A first lesson follows:
In game theory, an action is not evaluated only by its direct effect. It is evaluated in relation to the possible actions, beliefs, incentives, and information of other players.
This does not mean that game theory assumes people are perfectly wise or morally selfish in every real situation. A model must state its assumptions. Some models assume fully rational payoff maximization. Other models study bounded rationality, learning, fairness, error, or evolutionary adaptation. The discipline is broad, but the starting point is always the same: outcomes arise from interacting choices.
Players, choices, and consequences
The most basic ingredients of a game are players, choices, and consequences.
A player is a decision-making unit in the model. A player may be a person, a firm, a government, a committee, a biological organism, or even a software agent. The word “player” does not imply playfulness; it is a technical term for an entity whose choice matters.
An action is a move available to a player at a particular decision point. If two candidates decide whether to run a positive campaign or a negative campaign, each candidate’s possible actions might be “positive” and “negative.”
A strategy is a complete plan for a player. In a one-shot simultaneous game, an action and a strategy may look the same. But in a dynamic game, a strategy must specify what the player would do after every relevant history, including histories that may not actually occur. For example, in a negotiation, a strategy might say: “Make a high initial offer; if rejected, make a moderate offer; if the other side threatens to walk away, accept only if the offer is above a specified threshold.”
An outcome is the result produced by the players’ strategies. An outcome may include prices, profits, policies, allocations, agreements, conflicts, or any other result the model is designed to track.
A payoff is a numerical representation of a player’s preferences over outcomes. If player \(i\) receives payoff \(u_i(x)\) from outcome \(x\), a higher number means that player \(i\) prefers that outcome more, according to the model. Payoffs do not have to be money. They may represent utility, profit, votes, survival, status, fairness concerns, or any modeled objective. What matters is that the payoff numbers correctly represent the preference comparisons needed for the analysis.
For example, imagine two roommates choosing whether to clean the kitchen. Each prefers a clean kitchen to a dirty one, but each also prefers that the other person do the cleaning. A payoff table could assign numbers to the four possible outcomes: both clean, only the first cleans, only the second cleans, or neither cleans. The numbers are not the kitchen itself; they are a compact representation of how each roommate ranks the possible outcomes.
This distinction is important. Game theory does not begin with equations for their own sake. It begins by asking: What are the relevant decision makers? What can they choose? What do they care about? What do they know? When do they move? How do their choices combine to produce outcomes?
Models are maps, not mirrors
A game-theoretic model is a simplified representation of a strategic situation. A model is not valuable because it includes every detail. It is valuable if it isolates the incentives and information that matter for the question being asked.
A map of a subway system does not show every building, tree, and slope in the city. It shows the stations, lines, and connections needed for subway travel. In the same way, a game model should include the strategic structure needed for analysis, while leaving out details that would distract from the main mechanism.
Suppose we model price competition between two firms. We may ignore the personalities of the managers, the architecture of the stores, and the color of the advertisements. But we should not ignore whether prices are chosen simultaneously or sequentially, whether products are identical or differentiated, whether firms know each other’s costs, and whether the interaction happens once or repeatedly. Those features can change the strategic incentives.
Because models simplify, they can mislead if used carelessly. A model’s conclusion is always conditional on its assumptions. If a model assumes that players know each other’s payoffs, then its prediction may fail when players are uncertain about each other’s objectives. If a model assumes that players maximize only monetary payoffs, then it may miss behavior driven by fairness, identity, anger, or norms. Later in the book, especially in behavioral game theory and real-world modeling, we will return to these limitations.
The correct attitude is neither blind trust nor cynical rejection. A model is a disciplined way to think. It helps us see which assumptions produce which conclusions.
What it means to “solve” a game
In algebra, solving an equation usually means finding values that make the equation true. In game theory, solving a game means identifying behavior that satisfies a specified concept of strategic consistency. Different solution concepts answer different questions.
One important solution concept is dominance. A strategy is strictly dominated if another strategy gives a player a higher payoff no matter what the other players do. If a strategy is strictly dominated, a rational payoff-maximizing player should not use it. For example, if one route to campus is always slower than another route regardless of traffic, the slower route is strictly dominated for a student who only cares about travel time.
Another important concept is best response. A strategy is a best response to other players’ strategies if it gives the highest payoff among the available strategies, given what the others are doing. If your friend chooses the restaurant and you choose transportation, your best response may depend on the chosen restaurant’s location.
A Nash equilibrium is a profile of strategies—one strategy for each player—such that every player’s strategy is a best response to the strategies of the others. No player can improve by changing strategy alone. Nash equilibrium became a central concept because it captures mutual strategic consistency in noncooperative games, where each player chooses independently rather than through a binding collective agreement (Nash, 1951).
Here is a simple example. Two people want to meet but forgot to choose a location. Each can go to the library or the café. Both prefer meeting to missing each other. If both go to the library, they meet. If both go to the café, they also meet. If they choose different places, they miss each other. This game has two natural pure-strategy Nash equilibria: both go to the library, and both go to the café. In either case, no one wants to change location alone after seeing the other’s choice.
This example also teaches a warning: equilibrium does not always mean uniqueness. A game may have several equilibria. The solution concept may tell us which outcomes are stable in a particular sense, but not always which stable outcome will occur. To understand equilibrium selection, we may need additional ideas such as payoff dominance, risk dominance, history, communication, convention, or focal points. Schelling’s analysis of coordination emphasized that seemingly small contextual cues may guide people toward one equilibrium rather than another (Schelling, 1960).
Rationality and its role
Game theory often uses the word rationality, but this word must be handled carefully. In this book, a rational player is usually one whose choices are consistent with the player’s preferences and beliefs. If the player has a well-defined objective and beliefs about the situation, rational choice means choosing an available strategy that best serves that objective according to those beliefs.
This definition is narrower and more technical than ordinary uses of the word “rational.” It does not mean morally good, emotionally cold, socially admirable, or correct in hindsight. A person can rationally choose an action that leads to a bad result if the person’s information was limited. A person can rationally pursue altruistic goals if those goals are part of the person’s preferences. A person can rationally refuse a monetary gain if accepting it violates something the person values more.
For example, suppose a donor chooses between keeping money and giving it to charity. If the donor genuinely prefers the charitable outcome, then giving can be rational in the preference-based sense. Game theory does not require selfishness as a universal psychological law. It requires clarity about the preferences represented in the model.
That said, many classical models assume that players maximize their own payoffs and that these payoffs represent self-interested objectives such as profit or utility. This assumption can be useful, especially in markets and auctions, but it must not be confused with a universal description of human motivation. Later chapters will study models that incorporate fairness, reciprocity, bounded reasoning, and learning.
Information and beliefs
Strategic decisions depend not only on preferences and actions, but also on information. A player may know the rules of the game but not another player’s payoff, type, cost, valuation, or intention.
A belief is a player’s assessment of uncertainty. For example, a bidder in an auction may be uncertain about how much other bidders value the item. A firm considering entry into a market may be uncertain about whether the incumbent firm has low costs or high costs. A government in a crisis may be uncertain about another government’s willingness to fight.
A type is a formal description of a player’s private information. If a seller privately knows whether a used car is high quality or low quality, the seller’s type may be “high-quality seller” or “low-quality seller.” Other players may hold beliefs about which type the seller is.
Games with incomplete information were given a general formal treatment by Harsanyi, who showed how uncertainty about players’ characteristics can be represented by types and beliefs within a game model (Harsanyi, 1967–1968). This approach is central to modern auction theory, signaling, screening, and mechanism design.
For example, in a job-market signaling model, a worker may know their own productivity, while an employer does not. Education may then serve not only to build skill but also to signal information, depending on the model’s assumptions. The employer forms beliefs from observed choices, and the worker chooses while anticipating how those beliefs will affect wages.
The key idea is simple but deep:
When information is asymmetric, actions may communicate information, and beliefs become part of strategic analysis.
Time, credibility, and dynamic strategy
Many games unfold over time. One player moves, another observes something, then someone moves again. In such games, strategy must specify contingent plans: what to do after different possible histories.
A history is a sequence of earlier actions or events in a dynamic game. A contingent plan is a plan that says what a player will do depending on which history occurs. For example, a parent, child, firm, union, army, or regulator may say, “If you do this, I will respond in that way.” But not every announced response is credible.
A threat is credible if the player would actually want to carry it out when the time comes. Suppose an incumbent firm says it will start a costly price war if a new firm enters the market. If entry occurs and fighting would hurt the incumbent more than accommodating entry, then the threat may not be credible. A Nash equilibrium may sometimes rely on such noncredible threats. This is why dynamic games require refinements such as subgame-perfect equilibrium, which demands optimal behavior after every relevant history, not only along the path that actually occurs.
The idea of credibility is one of game theory’s most practical contributions. In law, diplomacy, business, and personal life, promises and threats matter only when others believe they will be carried out. Strategic analysis asks what makes such belief reasonable.
Cooperation without central control
A common misunderstanding is that game theory is only about conflict. In fact, game theory studies both conflict and cooperation. Often the most interesting cases involve both.
The Prisoner’s Dilemma is a famous example. Each player has an individual incentive to defect, even though both would be better off if both cooperated. In a one-shot version, noncooperation may be the equilibrium outcome. But if the same players interact repeatedly, cooperation can sometimes be sustained by future rewards and punishments. The possibility of future interaction changes present incentives.
This does not mean repetition automatically produces cooperation. Cooperation depends on patience, observability, the severity and credibility of punishments, the value of future interaction, and the availability of alternative partners. Repeated-game theory studies these conditions precisely.
Game theory therefore helps explain why cooperation may emerge even among self-interested agents, and why it may collapse when incentives change. A long-term supplier may maintain quality to preserve a relationship. Countries may follow treaties when future retaliation is likely. Neighbors may help each other because today’s help affects tomorrow’s support. In each case, the future casts a strategic shadow over the present.
Designing rules, not only predicting behavior
Much of game theory asks: given the rules, what will players do? Mechanism design asks a reverse question: what rules should be chosen so that strategic behavior leads to desirable outcomes?
A mechanism is a rule system that takes players’ actions or messages and produces an outcome. An auction is a mechanism: bidders submit bids, and the rules determine who gets the object and what price is paid. A voting system is a mechanism: voters submit ballots, and the rules determine the winner. A school-choice system, organ-exchange procedure, or public-project funding rule can also be studied as a mechanism.
Mechanism design became a central part of economic theory because it treats incentives as design constraints rather than afterthoughts. Myerson’s textbook presents mechanism design as part of the broader analysis of conflict and cooperation under strategic incentives (Myerson, 1991). The basic lesson is that if people have private information and their own objectives, a rule must be designed with their likely strategic responses in mind.
For example, suppose a city wants to learn how much residents value a new public park. If residents are simply asked, “How much would you pay?” they may exaggerate or understate their values depending on how the answer will be used. A good mechanism must ask whether truthful reporting is in each participant’s interest. This leads to ideas such as incentive compatibility, participation constraints, and the revelation principle, which we will study later.
What this book will build
The book is organized as a path from basic strategic reasoning to advanced undergraduate topics.
We begin with formal models: players, actions, strategies, preferences, payoffs, outcomes, and rationality assumptions. We then study dominance, best responses, rationalizability, and Nash equilibrium in pure and mixed strategies. These chapters provide the core language of noncooperative game theory.
Next we study how equilibria change when the environment changes, and how to reason when multiple equilibria are possible. We then move to extensive-form games, backward induction, subgame perfection, and repeated games, where time and credibility become central.
After that, we introduce incomplete information: types, beliefs, Bayesian games, signaling, screening, and perfect Bayesian equilibrium. These tools are essential for understanding auctions, bargaining, reputation, market entry, and strategic communication.
The later chapters broaden the perspective. Bargaining theory studies how surplus may be divided. Auction theory studies competitive bidding. Mechanism design studies how institutions can be built to align incentives. Cooperative game theory studies coalitions and allocation rules. Evolutionary game theory and learning in games ask how strategic patterns may emerge without full rational calculation. Network and matching models examine how the structure of interaction shapes incentives. Behavioral game theory studies systematic deviations from the standard rational-agent model.
Throughout, we will return to three guiding questions:
-
What is the strategic structure?
Who are the players, what can they do, what do they know, and what do they care about? -
What solution concept is appropriate?
Are we studying dominance, Nash equilibrium, backward induction, Bayesian equilibrium, stability of coalitions, evolutionary stability, or learning dynamics? -
What does the conclusion really depend on?
Which assumptions are essential, and how would the result change if those assumptions changed?
These questions are more important than memorizing famous games. A famous game is useful only when it teaches a transferable structure.
How to think while reading
When you meet a model, slow down and ask what each symbol represents. If \(u_i\) is a payoff function, ask: payoff for whom? Over which outcomes? If \(s_i\) is a strategy, ask: what complete plan does it describe? If a profile \(s = (s_1,\dots,s_n)\) is called an equilibrium, ask: according to which solution concept, and under what assumptions?
When you see a theorem, separate the statement from the intuition. The intuition helps you remember why the theorem matters. The formal statement tells you exactly when it applies. Do not use a theorem outside its hypotheses. A result for finite games may not automatically hold for infinite games. A result for complete information may not automatically hold when types are private. A result for simultaneous moves may not automatically hold when moves are sequential.
When you see an example, change it. What happens if one payoff changes? What if players move in sequence? What if one player observes a signal? What if the game is repeated? Comparative thinking is one of the best ways to learn game theory.
Most importantly, remember that equilibrium is not the same as goodness. An equilibrium can be inefficient, unfair, fragile, or undesirable. A Nash equilibrium says that no player has a profitable unilateral deviation under the modeled assumptions. It does not say that society should approve of the outcome. Part of mature game-theoretic reasoning is learning to distinguish prediction, explanation, evaluation, and design.
The promise and discipline of game theory
Game theory offers a disciplined language for thinking about strategic life. It can explain why individually reasonable choices may produce collectively poor outcomes, why promises need credibility, why information can be valuable or dangerous, why institutions matter, and why cooperation sometimes requires carefully structured incentives.
But the discipline also demands humility. A game-theoretic conclusion is not stronger than the model that supports it. If the model leaves out an important motive, information constraint, institutional rule, or dynamic feature, the conclusion may be incomplete. The goal of this book is therefore not to make every situation look like the same simple game. The goal is to teach you how to build, solve, interpret, and critique strategic models responsibly.
By the end of the book, you should be able to look at a strategic situation and ask clear questions: Who moves? What do they know? What do they want? What can they commit to? What do they believe? What happens if the rules change? These questions are the beginning of game theory from first principles.
References
Harsanyi, J. C. (1967–1968). Games with incomplete information played by “Bayesian” players, I–III. Management Science, 14(3), 159–182; 14(5), 320–334; 14(7), 486–502.
Myerson, R. B. (1991). Game Theory: Analysis of Conflict. Harvard University Press.
Nash, J. (1951). Non-cooperative games. Annals of Mathematics, 54(2), 286–295.
Schelling, T. C. (1960). The Strategy of Conflict. Harvard University Press.
von Neumann, J., & Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton University Press.