Research note
Game theory for traders, explained
Game theory for traders: Nash equilibrium, classic games, bluffs, reputation and auctions, and how market makers, informed traders and crowds interact.
Short answer: a market is a game, not a weather system. Your best order depends on what other traders do, and they are reasoning about you at the same time. Game theory gives that interaction a vocabulary: equilibrium, information, commitment, reputation and auction rules. It explains why spreads exist, why crowded trades unwind violently and why closing prices form the way they do. It does not hand anyone a trading edge.
This post follows Adriano's 26-page reference, "Game Theory: A Comprehensive Reference", from the basics to the applications. After each idea it adds the trading version, with outside sources for every claim the reference doesn't make itself.
What is game theory?
Game theory is the mathematical study of strategic interaction: situations where your outcome depends on your own choices and on everyone else's. That is what separates it from decision theory. A farmer choosing between corn and wheat is optimizing against nature, and the weather doesn't react to the choice. Two gas stations setting prices are playing a game. Each one's best price depends on the other's, and both know it.
Every game breaks down into four building blocks:
- Players: the decision-makers.
- Strategies: complete plans of action that cover every situation that could come up, not just a single move.
- Payoffs: what each player gets for every combination of strategies.
- Information: what each player knows, and when.
The rules also fix the order of play. Players move at the same time, in sequence, or some mix of the two.
There are two branches. Non-cooperative game theory, the branch John Nash built, assumes players can't make binding agreements, so each one chooses alone and in their own interest. Cooperative game theory assumes binding coalitions are possible and asks how to split the value they create. Its best-known answer is the Shapley value. Most applied work, including almost everything a trader cares about, sits in the non-cooperative branch.
For traders, the farmer-versus-gas-station test is the useful one. A backtest treats the market like weather: the past repeats and doesn't react to you. A live order book reacts. Size, timing and visibility change other people's behavior, and theirs changes yours.
Where did game theory come from?
The roots go back further than most people expect. Antoine Cournot modeled two competing firms in 1838. Ernst Zermelo proved in 1913 that a game like chess has, in principle, a determined outcome you can find by working backward from every possible ending.
John von Neumann made it a field. He proved his minimax theorem for two-player zero-sum games in 1928. In 1944 he and Oskar Morgenstern published Theory of Games and Economic Behavior, which founded game theory and set up modern expected utility theory along the way. At the RAND Corporation, Merrill Flood and Melvin Dresher devised the scenario that Albert Tucker named the Prisoner's Dilemma.
The 1950s belonged to Nash, then a Princeton graduate student. His equilibrium concept extended the theory far beyond zero-sum games. Thomas Schelling's The Strategy of Conflict (1960) applied it to real conflict, above all Cold War nuclear deterrence, and introduced credible commitment and focal points. In the 1970s John Maynard Smith, working with George Price, showed that natural selection could stand in for the rational calculator, which started evolutionary game theory.
The reference highlights five Nobel prizes in economics for game-theoretic work:
- 1994, Nash, Harsanyi and Selten: equilibrium analysis in non-cooperative games.
- 2005, Aumann and Schelling: conflict and cooperation through game theory.
- 2007, Hurwicz, Maskin and Myerson: the foundations of mechanism design.
- 2012, Roth and Shapley: stable allocations and market design.
- 2020, Milgrom and Wilson: auction theory and new auction formats.
One of those names matters directly to traders. Paul Milgrom, a 2020 laureate, co-wrote one of the classic models of the bid-ask spread, covered below.
What are the core concepts?
Pure and mixed strategies
A pure strategy is a single fixed choice, such as always defect or always offer half. A mixed strategy assigns a probability to each option, so the player randomizes. Mixing isn't a curiosity. Some games have no stable outcome unless at least one player randomizes.
Trading version: a predictable pattern is a pure strategy, and other players can learn it. An algorithm that always buys at the same minute in the same size gives away information. Randomizing size and timing is the mixed-strategy answer.
Perfect, imperfect, complete and incomplete information
Under perfect information, every player sees all earlier moves, as in chess. Under imperfect information, some moves are hidden, as with the hole cards in poker. Separately, complete information means every player's payoffs and goals are common knowledge, and incomplete information means some of that is private. John Harsanyi's framework for incomplete-information games, now called Bayesian games, earned him a share of the 1994 prize.
Markets are games of imperfect and incomplete information. You don't see the hidden part of other traders' orders, and you don't know why they trade. That is why the information models later in this post matter so much.
Zero-sum and non-zero-sum games
In a zero-sum game, one player's gain is exactly another's loss: poker chips changing hands. In a non-zero-sum game, the total can grow or shrink with the players' choices. The reference warns that the instinct to treat competition as zero-sum is usually wrong.
Markets have both. Marked against later prices, one side's gain on a trade is the other side's loss, before costs. But trade also does something useful: a hedger who wants to shed risk and a liquidity provider who is paid to take it can both meet their own goals.
Dominant and dominated strategies
A dominant strategy is your best choice whatever anyone else does. A dominated strategy is one that some other strategy beats or matches against every possible opponent move. A rational player never plays a dominated strategy, and repeatedly deleting dominated strategies (iterated elimination) sometimes narrows a game to one outcome. The guessing game in the beauty-contest section below is solved exactly this way, and it shows how far real people are from doing it.
Nash equilibrium and Pareto efficiency
A Nash equilibrium is a set of strategies, one per player, where no player can do better by switching alone, given what everyone else is doing. Nobody has to cooperate, talk or like the result. They just can't regret their own choice in isolation. Nash proved in 1950 that every finite game has at least one equilibrium once mixed strategies are allowed, which is what made the theory a general tool. A game can have one equilibrium or several, and the several can favor different players.
An outcome is Pareto efficient if nobody can be made better off without making someone worse off. A Nash equilibrium need not be Pareto efficient. That gap is one of the most important ideas in the field: individually rational choices can trap a group in an outcome everyone would rather avoid.
Correlated equilibrium
Robert Aumann's correlated equilibrium (1974) lets players condition their choices on a shared signal that none of them controls, such as a traffic light, a coin flip or a contract. The traffic light works because it coordinates two drivers into go and stop, which independent randomizing would manage far less reliably. Every Nash equilibrium is a correlated equilibrium, but not the other way round. Correlation can reach higher joint payoffs than independent mixing, and Aumann considered it the more natural concept for many real settings.
A market analogy, ours and not the reference's: scheduled public events like a data release or the closing auction act a bit like shared signals. Everyone can see them and condition on them, so activity clusters around them.
Worked example: solving Matching Pennies
Matching Pennies has no pure-strategy equilibrium, but you can solve for the mixed one. Say Player A plays Heads with probability p. Player B will only randomize if Heads and Tails pay B the same. B's expected payoff from Heads is −p + (1 − p) and from Tails is p − (1 − p). Setting them equal gives p = 1/2. The same logic applied to A shows B must also mix 50/50. That method, choosing your mix so the opponent is indifferent, solves any two-by-two mixed-strategy game.
The trading lesson is about being exploitable. If your behavior leaves an opponent with a clearly better response, they will find it. In equilibrium, you leave them nothing to exploit.
Which classic games show up in markets?
The reference uses eight small games. Each captures a structure that keeps reappearing.
The Prisoner's Dilemma
Two suspects are questioned separately. Each can stay silent or betray. Payoffs are listed as (A, B), higher is better:
- Both silent: 3, 3.
- A silent, B betrays: 0, 5.
- A betrays, B silent: 5, 0.
- Both betray: 1, 1.
Betraying is dominant for both, so mutual betrayal is the only equilibrium, even though mutual silence leaves both better off. The same structure drives price wars, arms races, overfishing and doping in sport.
In markets, it shows up when everyone has a reason to defect from a shared good. Each liquidity provider would like everyone else to keep quoting in a selloff while it steps back itself. Each fund in a crowded position does better if it sells first.
The Stag Hunt
Two hunters can work together for a stag that feeds both well, or each can snare a hare alone, which is smaller but certain:
- Both stag: 4, 4.
- A stag, B hare: 0, 3.
- A hare, B stag: 3, 0.
- Both hare: 3, 3.
Both stag is the best outcome (payoff-dominant). Both hare is the safe one (risk-dominant). Two stable equilibria differ only in how much trust they need. Co-founders, joint ventures and climate commitments face versions of this choice.
Trading version: liquidity. Market makers keep quoting tight spreads if they expect others to keep quoting too. If they start to doubt that, pulling back becomes the safe choice for each of them. The market can jump from the good equilibrium to the bad one without any change in fundamentals. More on this under crowded trades.
Chicken (Hawk-Dove)
Two drivers race at each other. Each can swerve or go straight:
- Both swerve: 3, 3.
- A swerves, B straight: 1, 5.
- A straight, B swerves: 5, 1.
- Both straight: 0, 0.
There are two equilibria, each favoring the driver who holds their nerve. Schelling's point was that visibly destroying your own ability to swerve, by ripping out the steering wheel where the other driver can see it, can be a winning move, because it makes your commitment credible.
Markets have Chicken standoffs whenever two large players want the other to blink first, for example over who absorbs a big imbalance or who keeps defending a level.
Battle of the Sexes
Two partners want to spend the evening together but prefer different events:
- Both at the concert: 2, 1.
- Both at the game: 1, 2.
- Apart: 0, 0.
Two equilibria, one favoring each partner. It is coordination with a conflict of interest on top, the shape of many negotiations where both sides want a deal but disagree on its terms. Conventions such as taking turns or flipping a coin matter a lot here.
Matching Pennies
Each player secretly picks heads or tails. A wins if the coins match, B wins if they don't:
- Match: 1, −1.
- No match: −1, 1.
It is zero-sum with no pure-strategy equilibrium. Whatever one player picks, the other wants to switch, so the only equilibrium is for both to randomize 50/50. Studies of professional soccer penalty kicks have found kickers and goalkeepers randomizing close to this prediction. Ignacio Palacios-Huerta's 2003 paper on penalty kicks is titled, fittingly, "Professionals Play Minimax."
Trading version: an informed trader and the people trying to detect them play something like Matching Pennies. If the informed side is predictable, the other side adjusts.
The Public Goods Game
Players privately decide how much to put into a shared pot. The pot is multiplied by a factor above one and split equally, whatever each person put in.
The reference's worked example: four players can each put in $10, and the pot is doubled. If all four contribute, $40 becomes $80 and each gets $20 back, a $10 profit each. But your own $10 becomes $20 in the pot and returns only $5 to you once it is split four ways. Paying $10 to get $5 back is a loss, even though it hands $5 to each of the other three. So free-riding is individually rational at every level, and lasting public goods usually need reputation, repeated play, social pressure or an enforced rule.
Markets have a public good: informative prices. Grossman and Stiglitz (1980) showed why prices can't be perfectly efficient. If prices already revealed everything, nobody would pay to gather information, so nobody would gather it. Some inefficiency has to remain to reward the people who do the research.
Colonel Blotto
Two commanders split a fixed budget of troops across several fronts. Whoever sends more to a front wins it, and whoever wins more fronts wins the game. Émile Borel formalized it in the 1920s. Blotto games generally have no pure-strategy equilibrium, since any predictable allocation can be beaten by concentrating on the fronts it neglects. The same structure covers campaign spending across districts, military deployment and marketing budgets.
A trading parallel, ours: a fixed risk budget spread across many positions, against opponents who can see which fronts you are thin on.
The Ultimatum Game
One player proposes how to split a sum of money. The other accepts or rejects, and a rejection leaves both with nothing. Backward induction says to offer the minimum and accept it, since anything beats nothing. Real people don't play that way. Since Werner Güth and colleagues introduced the game in 1982, experiments across many cultures have found that low offers are often rejected, as responders pay to punish unfairness, and that proposers offer well above the minimum. It helped launch behavioral economics.
For traders, the lesson is that real counterparties are not the perfectly rational players of textbook models. Fairness, pride and spite move real decisions.
How do sequential games, bluffs and signals work?
Backward induction and subgame perfection
When players move in turn, the game is drawn as a tree (extensive form) instead of a matrix (normal form). With perfect information you solve it backward. Find the best choice at each final decision, fold it back one step, and repeat until you reach the start. The result is rational at every decision point, which is called sequential rationality.
Reinhard Selten's subgame perfect equilibrium requires the strategies to be an equilibrium in every subgame, not just the game as a whole. In practice it rules out empty threats. An incumbent's threat of a ruinous price war only counts if fighting would still be the best move after the entrant has already come in.
The reference's worked example is a market-entry game. A challenger chooses to Enter or Stay Out. If it stays out, the incumbent earns 5 and the challenger 0. If it enters, the incumbent can Fight (incumbent 1, challenger −1) or Accommodate (3 each). Start at the end. After entry, the incumbent compares 1 against 3 and accommodates. Folding back, the challenger expects accommodation, compares 3 against 0 and enters. So the threat to fight any entrant is not credible, and a rational challenger calls the bluff. Selten called a version of this the chain-store paradox in his 1978 paper.
Trading version: a large player saying "I'll defend this level" is a threat. If defending would clearly cost them more than letting go once the level is tested, other traders have reason to treat it as cheap talk.
Stackelberg competition
In the Stackelberg duopoly, one firm commits to its output first and the other responds after seeing it. In Cournot, both choose at once. Moving first is an advantage: the leader commits to more output, the follower's best response is to produce less, and the leader takes the larger share. Commitment and timing are strategic assets, not just details.
Chess shows the limit of all this. It is solvable in principle by Zermelo's backward induction, but its tree is far too large to search fully. Engines use search heuristics and learned evaluation instead. A game being solved in theory and solvable in practice are very different things, and markets are much bigger than chess.
Signaling and screening
Sequential games get interesting once one side holds private information. In Michael Spence's 1973 job-market model, workers know their productivity and employers don't. A credential can signal ability even if it teaches nothing useful, as long as it is cheaper to earn for high-ability workers. That cost difference alone can separate the two types in equilibrium. Spence shared the 2001 Nobel with George Akerlof and Joseph Stiglitz for work on markets with asymmetric information, including Akerlof's used-car "market for lemons", where sellers know a car's quality and buyers don't.
The lemons problem is the bridge to market microstructure. If you suspect the person trading with you knows more, you protect yourself with your price. The next section shows how that turns into a bid-ask spread.
How does game theory explain market makers and takers?
The reference puts it plainly: market making is a repeated game against a mix of informed and uninformed counterparties, and the spread has to be wide enough to cover the risk of trading against someone who knows more. It calls this a close cousin of the winner's curse. Two 1985 papers made that precise.
Glosten-Milgrom: the spread as a lemons discount
Glosten and Milgrom (1985) model a specialist who is risk-neutral and makes zero expected profit, facing a mix of informed and uninformed traders. Their result is that the mere presence of traders with better information produces a positive bid-ask spread, even with no inventory cost and no profit motive. Each trade carries information, so the quote moves after it. In their words, observed returns are approximately realizable returns "plus what the uninformed anticipate losing to the insiders."
So the spread is partly a fee charged to everyone to cover losses to the few who know more. A market maker who can't tell informed flow from uninformed flow has to price the average.
Kyle: how an informed trader hides
Kyle (1985) models a single risk-neutral insider, random noise traders and competitive market makers who see only total order flow. Noise trading provides camouflage. The insider makes money by spreading trades over time instead of trading all at once, and in the continuous-time limit all of the private information is in the price by the end of trading.
That is a game: the insider chooses how fast to trade, knowing the market makers are inferring information from order flow. Trade too fast and the price moves against you. Trade too slowly and you leave money on the table.
Makers, takers and the modern book
In today's US equity market, the SEC's 2010 Concept Release on Equity Market Structure describes the "maker-taker" model. Resting orders that add liquidity can earn rebates, and orders that take liquidity pay access fees. It describes proprietary firms doing passive market making, earning the spread and rebates by posting large numbers of orders and often cancelling a very high percentage of them. It also describes reserve orders, which show part of their size and hide the rest, and broker algorithms that split a large parent order into many child orders.
Read through game theory, the maker posts a quote and accepts the risk of being picked off by better-informed takers. The taker pays the spread for immediacy and may be the informed one. Iceberg orders and slicing algorithms are, as the reference says, applied game theory: attempts to leak less information to traders who might act on it.
The same release discusses two directional strategies it says may pose problems for long-term investors, "order anticipation" and "momentum ignition." Both are attempts to read or provoke other players' behavior, which is the strategic part of trading.
Why do crowded trades unwind so violently?
Crowding is where the Prisoner's Dilemma and the Stag Hunt meet.
Coordination and multiple equilibria
Diamond and Dybvig (1983) showed that bank deposit contracts have multiple equilibria, one of which is a bank run. If everyone expects others to keep their money in, staying is rational. If everyone expects a run, running is rational. Nothing about the bank has to change for the outcome to flip. This is the Stag Hunt at the scale of an economy.
Crowded trades work the same way. When many funds hold the same position, each one's safety depends on the others not selling. Once enough of them expect others to sell, selling first becomes the dominant move.
Predatory trading
Brunnermeier and Pedersen (2005) studied what happens when one trader needs to sell. Other traders also sell, then buy back later. The price overshoots, the distressed trader gets less, and the market is illiquid exactly when liquidity is most needed. A trader can even profit from triggering someone else's crisis, and the crisis can spread across traders and markets.
The August 2007 quant meltdown
Khandani and Lo (2011) studied the "Quant Meltdown" of August 2007 using simulated long/short equity portfolios. They found it began in July and lasted until the end of 2007. A simulated high-frequency market-making strategy lost heavily in the week of August 6, 2007, but was profitable before and after. They read the dislocation as market-wide deleveraging plus a sudden withdrawal of market-making risk capital starting August 8.
That is the Stag Hunt and the Prisoner's Dilemma together. Many funds held similar positions, liquidity providers pulled back, and the safe individual choice made the collective outcome worse.
Where MRT tracks crowding and sentiment
MRT shows several crowding and sentiment readings. None of them is a signal to act on, and we have tested at least one ourselves.
/fear-greedblends six market-internal proxies into a 0 to 100 sentiment gauge. MRT's own research note tested it as an overlay on our Stovall sector-rotation strategy and found that it does not improve that strategy. It is kept as backtest evidence only, not as a live overlay./social-iqis a social digest with a market board on what the crowd is saying and a sentiment field in each ticker digest./newsis a news board./cryptohas a BTC top-trader long/short panel that flags when BTC top-trader positioning is more crowded on the long side than across total crypto./pairstracks spread z-scores for 30 correlated pairs. The page's own backtest verdict is that it is not a return source, which is a reminder that crowded relationships can break apart.
What is the Keynesian beauty contest?
In chapter 12 of The General Theory (1936), Keynes compared professional investment to a newspaper contest in which readers pick the six prettiest faces from a hundred photographs. The prize goes to the reader whose choice comes closest to the average choice of all readers. So each reader picks the faces they think others will pick, not the ones they like best. Keynes wrote: "We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be."
Rosemarie Nagel turned this into an experiment in 1995. Many players each pick a number from 0 to 100. The winner is the one closest to p times the average of all numbers, where p is known to everyone. For p below 1 the only Nash equilibrium is for everyone to pick zero, which you reach by iterated elimination of dominated strategies.
People didn't play zero. Nagel found that first-period behavior deviated strongly from the game-theoretic solution. Most choices were near 50p or 50p², consistent with one or two steps of reasoning from a starting point of 50. Over the four rounds, choices moved toward the equilibrium.
For traders, the beauty contest explains why price can move on what people expect others to think, not just on value. It also says the edge, if any, is in being about one step ahead of the crowd, not infinitely ahead. Being "right" at step ten loses to people reasoning at step two.
How do repeated games and reputation shape market behavior?
The folk theorem
Every game above was played once. Repetition changes the incentives, because a strategy can reward cooperation or punish betrayal next round. The folk theorem, developed by several game theorists through the 1970s and 1980s, shows that in a game repeated indefinitely by patient enough players, a wide range of outcomes can be an equilibrium. That includes full cooperation held up only by the threat of future punishment, even in a Prisoner's Dilemma whose one-shot version predicts none.
Axelrod's tournaments
In the early 1980s Robert Axelrod ran computer tournaments of the repeated Prisoner's Dilemma. The winner, submitted by Anatol Rapoport, was Tit-for-Tat: cooperate first, then copy the opponent's last move. It won again in a larger rerun. Axelrod credited four properties. It is nice (never defects first), retaliatory (punishes defection at once), forgiving (returns to cooperation when the opponent does) and clear (easy to read). When moves can be misread, strict Tit-for-Tat can spiral into needless retaliation, and more forgiving variants or "win-stay, lose-shift" do better. The grim trigger cooperates until the first defection and then defects forever.
How patient is patient enough?
Under a grim trigger, defecting today earns the one-time Temptation payoff T and then the Punishment payoff P forever, instead of the Reward R from cooperating. With δ (delta) as the discount factor, how much a payoff one round later is worth today, from 0 to 1, cooperation holds when:
δ > (T − R) / (T − P)
With the Prisoner's Dilemma numbers above (T = 5, R = 3, P = 1), that threshold is (5 − 3) / (5 − 1) = 0.5. The more tempting defection is, or the milder the punishment, the more patient players need to be. The reference calls this one inequality the mathematical heart of the folk theorem.
Reputation when the game ends
Repetition has a problem when the game has a known last round: backward induction unravels cooperation from the end. Kreps, Milgrom, Roberts and Wilson (1982) showed that a little uncertainty about whether the other player might be committed to Tit-for-Tat can sustain cooperation in a finitely repeated Prisoner's Dilemma. In the same journal issue, Kreps and Wilson studied reputation under imperfect information in the chain-store setting. The general lesson is that being seen as tough, or as reliable, can be worth more than it costs in any single round.
Markets as repeated games
The reference's own conclusion is that reputation systems, long-term supplier relationships and marketplace ratings all work by turning a one-shot temptation to cheat into a repeated game.
Market makers play that repeated game every day. Christie and Schultz (1994) found that odd-eighth quotes were virtually absent for 70 of 100 actively traded NASDAQ stocks. That implied inside spreads of at least $0.25 for many of them, and they raised the question of whether NASDAQ dealers were implicitly colluding to keep spreads wide. If dealers were coordinating, it would have been the tacit kind that repeated play among the same small group makes possible, with no written agreement.
On the other side, a broker or liquidity provider with a reputation for not leaking client orders gets more flow. A desk that defects once can lose that for a long time.
How does auction theory explain opening and closing auctions?
The four standard formats
- English: open, rising bids. The highest bidder wins and pays their final bid.
- Dutch: the price starts high and falls on a clock. The first bidder to accept wins at that price.
- First-price sealed-bid: one hidden bid each. The highest bid wins and pays that amount.
- Second-price sealed-bid (Vickrey): one hidden bid each. The highest bid wins but pays the second-highest bid.
Revenue equivalence and the Vickrey insight
Under fairly restrictive assumptions (risk-neutral bidders, independent private values, no collusion), the Revenue Equivalence Theorem says all four formats give the seller the same expected revenue, because bidders adjust to each format's rules. In a Vickrey auction your bid decides only whether you win, not what you pay, so bidding your true value is a dominant strategy. William Vickrey set out the idea in a 1961 paper. The insight that rules can be designed to make honesty the rational choice became the founding idea of mechanism design, recognized in 2007.
Worked example from the reference: three bidders value a painting at $900, $650 and $400 under Vickrey rules. Each bids their true value. The $900 bidder wins and pays $650. Nobody can do better by bidding above or below their value. That only risks winning at a worse price or losing a purchase that would have paid off.
The winner's curse
In a common-value auction, where the item is worth about the same to everyone but nobody knows exactly how much (an oil lease, for example), the winner tends to be whoever had the most optimistic estimate. Unless bidders shade their bids down to correct for that, winning is on average evidence of overpaying. The idea was popularized in a 1971 paper by Capen, Clapp and Campbell on bidding for oil leases. It also shows up in M&A, real estate bidding wars and IPO allocation.
For traders, a filled limit order is a small auction win. If your order fills right before price moves through it, you may have been the most optimistic bidder in the room. That is adverse selection, the same thing Glosten and Milgrom put inside the spread.
Modern applications
The Vickrey-Clarke-Groves mechanism extends truthful bidding to many related items and underlies systems like search advertising auctions. Governments have used simultaneous ascending auctions, developed largely by Milgrom, to sell wireless spectrum licenses worth tens of billions of dollars.
Opening and closing auctions
Stock exchanges run auctions every day. Nasdaq's Closing Cross at 4:00 p.m. ET matches on-close orders (MOC, LOC and imbalance-only) together with continuous-book orders and executes at a single price that maximizes the number of shares matched. From 3:50 p.m. ET Nasdaq publishes a Net Order Imbalance Indicator with paired shares, the imbalance side and size, and a reference price. From 3:55 p.m. ET it adds near and far indicative clearing prices and updates every second. Its Opening Cross runs at 9:30 a.m. ET, with imbalance data from 9:25 a.m. NYSE publishes opening-auction imbalances from 8:00 a.m. ET and closing imbalances from 3:50 p.m. ET, every second until each auction completes.
These are designed games. Everyone who trades in the cross gets the same price. Published imbalances tell other players which side needs liquidity, and Nasdaq's imbalance-only orders exist to offset on-close orders. The US Treasury runs its debt auctions on uniform terms: all successful bidders get the same rate, yield or discount margin as the highest accepted bid.
Our reading, not a sourced finding: published imbalance data turns the close into a small beauty contest, where each player is also guessing how others will react to the same numbers.
How do bargaining and focal points apply to trading?
The Nash bargaining solution
Nash's 1950 bargaining solution picks the split that maximizes the product of each side's gain over their disagreement point, which is what each gets if talks fail. It rests on a few requirements: efficiency, symmetry, independence of irrelevant alternatives, and independence from how utility is measured. The practical takeaway is that bargaining power comes mostly from your walk-away option.
Worked example from the reference: a founder and an investor split $1,000,000 from a sale. If talks fail, the founder gets $100,000 and the investor $200,000. The surplus above those points is $700,000, split equally. The founder gets $450,000 and the investor $550,000. Improve either side's outside option and the split moves with it.
Negotiating a large block trade works the same way. The side with better alternatives, such as other buyers, other venues or time to wait, gets the better price.
Alternating offers
In Ariel Rubinstein's 1982 model, two players take turns proposing splits of a pie that shrinks with every round of delay. The unique subgame-perfect solution depends on patience. The side with the lower cost of waiting gets more. Whoever needs the deal less usually gets better terms, which is why a forced seller tends to get a bad price.
Focal points
Schelling noticed that without any communication, people often converge on the same answer because it is salient: a round number, a precedent, a landmark. In one of his experiments, strangers told to meet in a large city on a given day, with no way to coordinate, mostly picked the same place and time. Focal points explain why deals settle around round numbers.
Prices do this too. Harris (1991) found that stock prices cluster on round fractions, and that clustering rises with price level and volatility and falls with market cap and trading frequency. He explained it as traders using coarser price sets to simplify negotiation. Christie and Schultz's NASDAQ quotes, above, were a case of clustering that raised a collusion question.
Where else does game theory apply?
Business competition
Two models cover most oligopoly analysis. In Cournot competition firms choose quantities at the same time. In Bertrand competition they choose prices for an identical product. Bertrand's result is stark: with just two firms selling an identical good, price falls to cost. Real markets escape through differentiation, capacity limits or repeated play. Incumbents also use entry deterrence and limit pricing to discourage competitors.
The reference's Cournot worked example shows why cartels are unstable. Demand is Price = 100 − total quantity, and production is free. Each firm's best response is to supply a third of the maximum market, about 33.3 units, so the price settles near $33 and each firm earns about $1,111. A monopolist would produce 50 units at $50 and earn $2,500, more than the roughly $2,222 the two rivals make together. Each firm's rational output undercuts their joint profit. That is the Prisoner's Dilemma in a market, and why cartels such as OPEC quotas are a standing temptation to cheat.
Bertrand competition is a rough picture of competing market makers. If quotes were truly identical and costs equal, competition would push spreads down to cost. Adverse selection, inventory risk and repeated play are the reasons they don't always get there.
Financial markets
The reference's summary is that price discovery can be modeled as a continuous game among participants with different information and time horizons, and that strategic execution, including iceberg orders and algorithms built to leak less information, is applied game theory running in production. The market makers and takers section above is that idea worked out.
Evolutionary biology
John Maynard Smith used game theory to explain animal behavior without assuming anyone calculates anything. An Evolutionarily Stable Strategy is one that, once common, can't be invaded by a rare alternative. Selection plays the role of rational choice. The Hawk-Dove game, the biological relative of Chicken, explains recurring patterns in animal fights over territory and mates.
Markets have an evolutionary side too. Andrew Lo's Adaptive Markets Hypothesis (2004) applies competition, adaptation and natural selection to financial markets. Strategies that work attract capital until they stop working. MRT's Hidden Markov Model page (/hmm) fits a model to SPY daily returns and 21-day realized volatility and estimates the probability that the market is in each of several regimes, which is one way to see that the environment strategies compete in keeps changing.
Politics and international relations
Arms races map onto the Prisoner's Dilemma, and nuclear standoffs onto Chicken. Both informed Schelling's work on commitment and deterrence. Coalitions in legislatures and agreements like climate accords are large cooperative-game problems, where the hard part is making the deal stay in each signatory's interest after signing.
The 1962 Cuban Missile Crisis is the standard example of Chicken at the highest stakes. Both sides preferred backing down to nuclear war, and each preferred that the other back down first. Schelling's writing, published two years earlier, had given analysts a shared vocabulary of brinkmanship, credible commitment and the value of visibly narrowing your own options. The naval blockade is often read as that kind of move: public, hard to reverse, signaling resolve while leaving both sides room to swerve.
Computer science, AI and mechanism design
Algorithmic game theory designs systems such as auctions, resource allocation and network protocols that work well even when everyone acts strategically. Multi-agent reinforcement learning borrows Nash and correlated equilibrium to train AI systems that face other strategic agents rather than a fixed environment. Braess's Paradox, where adding a road can make congestion worse for everyone once drivers route selfishly, is a classic case of individual and collective rationality pulling apart.
The trading parallel is that a market full of learning algorithms is itself a multi-agent game. A strategy tested against fixed historical data isn't tested against opponents who adapt.
Who are the key game theorists?
The reference's selective list:
- John von Neumann (1903–1957): minimax theorem, co-founder of the field with the 1944 book.
- Oskar Morgenstern (1902–1977): co-author of the 1944 book, expected utility theory.
- John Nash (1928–2015): Nash equilibrium and its existence proof, Nobel 1994.
- Thomas Schelling (1921–2016): focal points, credible commitment, strategic conflict, Nobel 2005.
- Reinhard Selten (1930–2016): subgame perfect equilibrium, bounded rationality, Nobel 1994.
- John Harsanyi (1920–2000): games of incomplete information, Nobel 1994.
- Robert Aumann (born 1930): correlated equilibrium, common knowledge, Nobel 2005.
- Lloyd Shapley (1923–2016): Shapley value, matching theory, Nobel 2012.
- Roger Myerson (born 1951): mechanism design, Nobel 2007.
- Alvin Roth (born 1951): market design for kidney exchange and school choice, Nobel 2012.
Two stories stand out. Game theory was something of a side project for von Neumann. He was also central to the Manhattan Project's implosion-lens calculations, helped design the architecture behind most modern computers and pioneered numerical weather simulation. Nash developed paranoid schizophrenia soon after his early work and spent years outside mathematics. A remission beginning in the 1980s let him return to research and accept the 1994 Nobel. Sylvia Nasar's biography and its film adaptation, both called A Beautiful Mind, made his story widely known.
What are the key takeaways for traders?
The reference's takeaways, each with its trading reading:
- Game theory formalizes strategic interdependence. Your best order depends on what others do, and they know it too.
- A Nash equilibrium is a stable mutual best response, not necessarily a good or fair outcome. Wide spreads or crowded positions can be equilibria.
- Individually rational choices can trap a group in an outcome everyone wanted to avoid. Liquidity withdrawal in a selloff is one example.
- Repetition changes everything. Reputation keeps dealers, brokers and counterparties honest, and sometimes keeps spreads wide.
- Not every game has a stable pure strategy, and being deliberately unpredictable can be rational. Predictable order flow gets read.
- A threat or promise only matters if it would be rational to carry out. Treat "I'll defend this level" with care.
- Bargaining power is mostly your outside option. Forced sellers get bad prices.
- Rules can be designed to produce a desired outcome. Opening and closing auctions are designed games.
- A few structures (Prisoner's Dilemma, Stag Hunt, Chicken, Battle of the Sexes) keep recurring, including in markets.
What game theory does not do is tell you which trade to make. It explains why the game looks the way it does. Testing whether any reading of it holds up is a separate job. MRT's Strategy Lab (/strategy-lab) compares rules against buy-and-hold and a seeded shuffle-null, and labels in-sample results as in-sample, and the newsletter at /newsletter audits viral trading claims.
FAQ
What is game theory in trading?
It is the study of trading as strategic interaction: your best action depends on what other traders do and what they know. It explains things like the bid-ask spread, the effect of order size on price and the way auctions set prices. It describes how markets work rather than giving trading signals.
Why do market makers charge a spread?
Partly to cover the risk of trading with someone better informed. Glosten and Milgrom (1985) showed that informed traders alone produce a positive spread, even for a risk-neutral market maker that makes zero expected profit. Inventory risk and costs add to it.
What is the Keynesian beauty contest in investing?
Keynes compared investing to a contest where you win by picking what others will pick, not what you like best. In markets, prices can move on expectations about other people's expectations. Nagel's 1995 experiment found that most people reason only one or two steps ahead.
Is the stock market a zero-sum game?
Not exactly. Measured against later prices, short-term trading gains and losses roughly offset before costs and are negative after them. But trade also lets people transfer risk and get liquidity, so both sides can meet their own goals.
How do opening and closing auctions work?
Exchanges collect orders before the open or close and execute them at a single price that maximizes the shares matched. Nasdaq and NYSE publish imbalance information beforehand so other traders can offset it. Everyone who trades in the auction gets the same price.
Further reading from the source reference
Accessible introductions:
- Avinash Dixit and Barry Nalebuff, Thinking Strategically (1991).
- Avinash Dixit, Susan Skeath and David Reiley, Games of Strategy, a rigorous but teachable textbook.
- Thomas Schelling, The Strategy of Conflict (1960), dense in places but foundational and unusually readable.
Behavioral and applied:
- Robert Axelrod, The Evolution of Cooperation (1984).
- William Poundstone, Prisoner's Dilemma (1992), history and biography woven through the theory.
Rigorous and graduate-level:
- Martin Osborne, An Introduction to Game Theory (2003).
- Drew Fudenberg and Jean Tirole, Game Theory (1991), the standard graduate reference.
- Roger Myerson, Game Theory: Analysis of Conflict (1991).
The original 1944 von Neumann and Morgenstern text is essential history, but it is best read after one of the introductions.
Sources
- Adriano Manzerra, "Game Theory: A Comprehensive Reference" (26-page reference document, unpublished, not publicly linked), 2026
- Glosten and Milgrom, "Bid, ask and transaction prices in a specialist market with heterogeneously informed traders", Journal of Financial Economics 14(1), March 1985: https://doi.org/10.1016/0304-405X(85)90044-3
- Kyle, "Continuous Auctions and Insider Trading", Econometrica 53(6), November 1985: https://www.econometricsociety.org/publications/econometrica/1985/11/01/continuous-auctions-and-insider-trading
- US Securities and Exchange Commission, "Concept Release on Equity Market Structure" (Release No. 34-61358), January 2010: https://www.sec.gov/rules/concept/2010/34-61358.pdf
- Grossman and Stiglitz, "On the Impossibility of Informationally Efficient Markets", American Economic Review 70(3), June 1980: https://ideas.repec.org/a/aea/aecrev/v70y1980i3p393-408.html
- Diamond and Dybvig, "Bank Runs, Deposit Insurance, and Liquidity", Journal of Political Economy 91(3), 1983: https://doi.org/10.1086/261155
- Brunnermeier and Pedersen, "Predatory Trading", Journal of Finance 60(4), 2005: https://doi.org/10.1111/j.1540-6261.2005.00781.x
- Khandani and Lo, "What happened to the quants in August 2007? Evidence from factors and transactions data", Journal of Financial Markets 14(1), 2011: https://doi.org/10.1016/j.finmar.2010.07.005
- Keynes, The General Theory of Employment, Interest and Money, chapter 12, 1936 (online text at Marxists Internet Archive): https://www.marxists.org/reference/subject/economics/keynes/general-theory/ch12.htm
- Nagel, "Unraveling in Guessing Games: An Experimental Study", American Economic Review 85(5), December 1995: https://www.cs.princeton.edu/courses/archive/spr07/cos444/papers/nagel95.pdf
- Kreps, Milgrom, Roberts and Wilson, "Rational cooperation in the finitely repeated prisoners' dilemma", Journal of Economic Theory 27(2), 1982: https://doi.org/10.1016/0022-0531(82)90029-1
- Kreps and Wilson, "Reputation and imperfect information", Journal of Economic Theory 27(2), 1982: https://doi.org/10.1016/0022-0531(82)90030-8
- Selten, "The chain store paradox", Theory and Decision 9(2), 1978: https://doi.org/10.1007/BF00131770
- Christie and Schultz, "Why do NASDAQ Market Makers Avoid Odd-Eighth Quotes?", Journal of Finance 49(5), 1994: https://doi.org/10.1111/j.1540-6261.1994.tb04782.x
- Harris, "Stock Price Clustering and Discreteness", Review of Financial Studies 4(3), 1991: https://doi.org/10.1093/rfs/4.3.389
- Vickrey, "Counterspeculation, Auctions, and Competitive Sealed Tenders", Journal of Finance 16(1), 1961: https://doi.org/10.1111/j.1540-6261.1961.tb02789.x
- Capen, Clapp and Campbell, "Competitive Bidding in High-Risk Situations", Journal of Petroleum Technology 23(6), 1971: https://doi.org/10.2118/2993-PA
- Palacios-Huerta, "Professionals Play Minimax", Review of Economic Studies 70(2), 2003: https://doi.org/10.1111/1467-937X.00249
- Spence, "Job Market Signaling", Quarterly Journal of Economics 87(3), 1973: https://doi.org/10.2307/1882010
- Akerlof, "The Market for Lemons: Quality Uncertainty and the Market Mechanism", Quarterly Journal of Economics 84(3), 1970: https://doi.org/10.2307/1879431
- Lo, "The Adaptive Markets Hypothesis", Journal of Portfolio Management 30(5), 2004: https://doi.org/10.3905/jpm.2004.442611
- Nasdaq, "Nasdaq Closing Cross" FAQ, accessed 2026-10-17: https://www.nasdaqtrader.com/content/productsservices/Trading/ClosingCrossfaq.pdf
- Nasdaq, "The Nasdaq Opening and Closing Crosses", accessed 2026-10-17: https://www.nasdaqtrader.com/Trader.aspx?id=OpenClose
- NYSE, "Auctions", accessed 2026-10-17: https://www.nyse.com/trade/auctions
- TreasuryDirect, "How Auctions Work", accessed 2026-10-17: https://www.treasurydirect.gov/auctions/how-auctions-work/
- MacroMomentum Terminal, Fear and Greed methodology and backtest note (internal research, not publicly linked), accessed 2026-10-17
- MacroMomentum Terminal, Pairs Divergence Monitor backtest verdict, accessed 2026-10-17: /pairs
Educational only, not financial advice. Past and backtested results do not guarantee future returns. Do your own research or speak with a licensed professional before making investment decisions.