Game theory - Advanced Topics and Tools

Advanced Topics in Game Theory This section covers sophisticated applications and extensions of game theory that go beyond the foundations. These topics address real-world complications like incomplete information, the design of institutions, and how actual human behavior deviates from theoretical predictions. Incomplete and Asymmetric Information: Signaling Games In many real situations, one player knows something important that others don't. A job applicant knows her actual ability, but employers don't. A company knows its financial health, but investors don't. Signaling games are models that analyze how informed players can communicate private information through their actions. How Signaling Works In a signaling game, an informed player (the "sender") chooses an action that other players (the "receivers") observe. The sender's goal is to convince receivers of something favorable about herself. However, receivers understand that the sender has an incentive to mislead them, so they interpret signals skeptically. The key insight is that a signal is credible only if it's expensive to fake. For example, getting a college degree is expensive in time and money. A high-ability person might be willing to bear this cost to prove herself, while a low-ability person might not find the cost worthwhile. This difference in costs can make the signal "separate" the two types. Consider a simple example: suppose workers have either high ability or low ability (known only to themselves), and firms can observe education but not ability. If firms believe "educated workers are high-ability," then only high-ability workers will invest in education (since they can recoup the cost with higher wages), while low-ability workers won't. This self-fulfilling belief creates a separating equilibrium where the signal perfectly reveals type. The tricky part: not all information can be signaled this way. If the cost difference between types isn't large enough, both types will choose the same action, and no separation occurs. Mechanism Design and Auction Theory Instead of analyzing games as they exist, mechanism design asks: how should we design institutions, rules, and incentive structures to achieve desired outcomes? The Revelation Principle The revelation principle is perhaps the most important theoretical result in mechanism design. It states that any outcome achievable by any mechanism can be achieved by a truthful direct-revelation mechanism, where players simply report their private information honestly. This might seem surprising: why would players ever tell the truth if they have incentives to lie? The answer is that the mechanism can be designed to make truthfulness optimal. Specifically, if you can design payoffs cleverly, you can make it a dominant strategy for each player to truthfully report their information, regardless of what others do. The revelation principle's power lies in simplification. Instead of designing complex, indirect mechanisms, mechanism designers can focus on direct-revelation mechanisms where players are incentivized to be honest. Any outcome you could achieve with a complicated mechanism, you can achieve this way. Vickrey Auctions and the Dominant Strategy for Truth The Vickrey auction (also called a second-price sealed-bid auction) is a concrete example where truthful bidding is a dominant strategy. Here's how it works: All bidders simultaneously submit sealed bids (no one sees others' bids) The highest bidder wins the object The winner pays the second-highest bid, not their own bid This last rule is crucial. Suppose the object is worth $100 to you. Should you bid truthfully ($100)? If you bid $100 and win, you pay the second-highest bid. If that bid is $80, you pay $80 and gain $20. You're happy to have bid honestly. Could you do better by bidding lower, say $90? If the second-highest bid is $80, you still win and pay $80—same outcome. But if the second-highest bid is $95, you lose the auction with a bid of $90 when you would have won with $100. So you lose your surplus. Could you do better by bidding higher, say $110? If you win, you still pay the second-highest bid. If the second-highest is $80, you pay $80 regardless of whether you bid $100 or $110. If the second-highest is $105, you pay $105 with either bid. But if it's $102, bidding $110 means you win and pay $102 (losing $2), while bidding $100 means you lose and gain $0. Again, honesty is better. The key insight: because you pay the second-highest bid, not your own, your bid can't directly affect your payment. This severs the connection that normally incentivizes bidding strategically. You might as well bid your true valuation. Behavioral and Experimental Game Theory Laboratory experiments reveal that real people often behave differently from what strict rationality predicts. These deviations are systematic and important to understand. Deviations from Pure Rationality Players show fairness preferences and inequality aversion—they care not only about their own payoffs but also about whether outcomes are fair and whether others are treated well. For example, in the ultimatum game, a proposer offers to split money with a responder. A purely self-interested proposer offers the minimum, and a rational responder accepts any positive amount. Yet in practice, proposers offer roughly equal splits, and responders frequently reject unfair offers—even though rejection makes both worse off. People are willing to punish unfairness at a cost to themselves. Similarly, people exhibit inequality aversion: they dislike situations where they or others are treated unequally. This drives behavior away from inequitable Nash equilibria toward more balanced outcomes. Bounded Rationality Players don't actually engage in unlimited strategic reasoning. Bounded rationality models recognize that people have cognitive constraints. Rather than solving the entire game tree, players use heuristics, engage in limited lookhead, or use simple rules of thumb. This matters because it means: Players may not find Nash equilibrium even when it exists The depth of strategic thinking is limited (I think about what you'll do, but maybe not about what you think I think you'll do, and so on) Simpler strategies often outperform complex ones in real settings Initial choices and learning effects matter, unlike in perfect rationality models Social Choice and Voting When a group must make a collective decision, voting rules are supposed to aggregate individual preferences into a group choice. But there are fundamental impossibilities. Arrow's Impossibility Theorem Arrow's impossibility theorem proves that no rank-order voting system can simultaneously satisfy four seemingly reasonable fairness criteria: Unrestricted domain: The voting rule works for any possible preference profile Non-dictatorship: No single voter determines the outcome Pareto efficiency: If everyone prefers option A to option B, the group outcome ranks A above B Independence of irrelevant alternatives (IIA): The group's ranking of A versus B depends only on voters' rankings of A versus B, not on their preferences over other options The theorem says you cannot have all four. Some voting system must fail at least one criterion. Why this matters: Many voting rules seem natural but must fail one of these properties. Majority voting often fails IIA (the presence of a third candidate can flip the majority choice between two others). Weighted voting systems might have a dictator. There's no escape: group decision-making inevitably has uncomfortable properties. The Gibbard-Satterthwaite Theorem The Gibbard-Satterthwaite theorem addresses strategic voting. It shows that every non-trivial voting rule (one where at least three outcomes are possible) is susceptible to strategic manipulation. This means: for any voting rule except a dictatorship, there exist preference profiles where some voter can submit a false preference ranking and improve the group outcome (in their view) compared to voting honestly. The intuition: If a voting rule responds to preferences in a systematic way, and your honest preference might lose, you might have an incentive to vote strategically. The theorem proves this possibility always exists for reasonable voting rules. <extrainfo> Evidence on Forecast Accuracy Game theory models often outperform non-game-theoretic approaches at predicting outcomes in controlled laboratory and field settings, though accuracy depends heavily on the context and assumptions. This empirical support justifies the continued use of game-theoretic models in economics, political science, and other fields, even when deviations from rationality occur. Game Theory Software Tools Gambit is a comprehensive software package for analyzing and solving games in normal form, extensive form, and stochastic form. While such tools are useful for practitioners, they are rarely the focus of exam questions in a theory course. </extrainfo>