Game Theory Models

𝐀𝐩𝐩𝐥𝐢𝐞𝐝 𝐆𝐚𝐦𝐞 𝐓𝐡𝐞𝐨𝐫𝐲 𝐑𝐞𝐯𝐞𝐚𝐥𝐬 𝐌𝐚𝐫𝐤𝐞𝐭 𝐌𝐢𝐬𝐩𝐫𝐢𝐜𝐢𝐧𝐠 𝐢𝐧 𝐈𝐧𝐝𝐢𝐚-𝐏𝐚𝐤𝐢𝐬𝐭𝐚𝐧 𝐓𝐞𝐧𝐬𝐢𝐨𝐧𝐬 Just published my deep dive on the India-Pakistan situation using rigorous game theory models. Markets routinely misread geopolitical events - I'm seeing it happen again right now. When most analysts rely on historical pattern-matching and vague analogies, they miss the underlying strategic logic driving predictable outcomes. My payoff matrices don't lie: contained conflict (75% probability) emerges as the dominant Nash equilibrium due to Pakistan's economic impossibility of sustained warfare colliding with India's nuclear-ceiling constraints. I've run 10,000 Monte Carlo simulations testing this mathematical reality against real-world constraints. The results confirm what game theory predicted - the structural incentives create a remarkably stable strategic equilibrium around limited military action. The investment edge? Most market participants are flying blind. They're pricing in the expected 7.2% correction correctly but completely missing the fat tail risk in the escalation scenario. Classic mistake. What fascinates me most is watching foreign institutional investors and retail create their own coordination game in real-time. FIIs hunting stag while retail hunts rabbit - an unstable equilibrium begging to resolve. The central bank response function adds another dimension. My models show intervention probability has dropped from historical 85% to current 67% given inflation constraints. Nobody's factoring this into their positioning. For business strategists, the principles at work here extend far beyond this specific conflict: - Payoff structures trump personalities - Economic constraints create hard boundaries on strategic options - Narrative fragility creates exploitable market inefficiencies - The most dangerous scenarios are often systematically underpriced After 15+ years applying these models to corporate warfare and investment strategies, I'm still amazed how few decision-makers understand the mathematical reality underlying strategic interactions. Look beyond headlines. When you understand the game theory, you see market participants consistently misreading the strategic chessboard. That's your edge. Thoughts? Challenges to my analysis welcome. #GameTheory #MarketStrategy #GeopoliticalRisk #InvestmentStrategy

LLMs struggle with rationality in complex game theory situations, which are very common in the real world. However integrating structured game theory workflows into LLMs enables them to compute and execute optimal strategies such as Nash Equilibria. This will be vital for bringing AI into real-world situations, especially with the rise of agentic AI. The paper "Game-theoretic LLM: Agent Workflow for Negotiation Games" (link in comments) examines the performance of LLMs in strategic games and how to improve them. Highlights from the paper: 💡 Strategic Limitations of LLMs in Game Theory: LLMs struggle with rationality in complex game scenarios, particularly as game complexity increases. Despite their ability to process large amounts of data, LLMs often deviate from Nash Equilibria in games with larger payoff matrices or sequential decision trees. This limitation suggests a need for structured guidance to improve their strategic reasoning capabilities. 🔄 Workflow-Driven Rationality Improvements: Integrating game-theoretic workflows significantly enhances the performance of LLMs in strategic games. By guiding decision-making with principles like Nash Equilibria, Pareto optimality, and backward induction, LLMs showed improved ability to identify optimal strategies and robust rationality even in negotiation scenarios. 🤝 Negotiation as a Double-Edged Sword: Negotiations improved outcomes in coordination games but sometimes led LLMs away from Nash Equilibria in scenarios where these equilibria were not Pareto optimal. This reflects a tendency for LLMs to prioritize fairness or trust over strict game-theoretic rationality when engaging in dialogue with other agents. 🌐 Challenges with Incomplete Information: In incomplete-information games, LLMs demonstrated difficulty handling private valuations and uncertainty. Novel workflows incorporating Bayesian belief updating allowed agents to reason under uncertainty and propose envy-free, Pareto-optimal allocations. However, these scenarios highlighted the need for more nuanced algorithms to account for real-world negotiation dynamics. 📊 Model Variance in Performance: Different LLM models displayed varying levels of rationality and susceptibility to negotiation-induced deviations. For instance, model o1 consistently adhered more closely to Nash Equilibria compared to others, underscoring the importance of model-specific optimization for strategic tasks. 🚀 Practical Implications: The findings suggest LLMs can be optimized for strategic applications like automated negotiation, economic modeling, and collaborative problem-solving. However, careful design of workflows and prompts is essential to mitigate their inherent biases and enhance their utility in high-stakes, interactive environments.

Most negotiators lose deals because they divide value wrong.    In every deal, the big question is:    Who gets what?   Most people decide based on:    ❌ Who speaks the loudest  ❌ Who has more power  ❌ Who simply asks for more    But the best negotiators don’t guess.    They use Shapley Value:   A game theory concept that shows exactly how much each person should get based on their real contribution.    Here’s the problem:    Most negotiators assume their value is obvious.    It’s not.     Let’s say three companies form a partnership:    - One brings technology  - One brings customers - One brings funding   Who deserves the biggest share?    Instead of arguing, Shapley Value calculates each partner’s real impact.   ✅ What happens if one partner leaves?  ✅ How much does each person’s role increase the total success?  ✅ What’s their actual contribution in numbers?    This shifts the conversation from opinion to logic.    How to use this in negotiations:   (Step-by-Step)    🔹 Step 1: Identify all contributors   List out everyone involved in the deal:   - partners, - suppliers, - team members - anyone adding value.    🔹 Step 2: Define measurable contributions   Ask:   What does each person bring to the table?   Focus on revenue impact, risk reduction, efficiency, or access to key resources.    🔹 Step 3: Calculate impact if one party is removed   For each contributor, ask:    “If this person/company walked away, how much value would be lost?”   🔹 Step 4: Assign value based on actual impact   If one party is responsible for 40% of the success, they should get a 40% share.   Not just an equal split.    🔹 Step 5: Use this data to justify your position   Instead of saying, “I want 30%,”* say:    “Based on our contribution analysis, our role increases revenue by 30%, reduces risk by 20%, and improves efficiency by 25%. Our fair share should reflect that.”   This eliminates emotional arguments and forces negotiations to focus on real impact.    Bottom line:   Most people negotiate based on feelings.    The best negotiators prove their worth.   If you’re not using game theory in negotiations, you’re leaving money on the table.   P.S. How do you ensure fairness in your deals?    Drop your insights below. I’d love to hear your take.    ---------------------- Hi, I’m Scott Harrison and I help executive and leaders master negotiation & communication in high-pressure, high-stakes situations. - ICF Coach and EQ-i Practitioner - 24 yrs | 19 countries | 150+ clients  - Negotiation | Conflict resolution | Closing deals 📩 DM me or book a discovery call (link in the Featured section)

UX research gets much stronger when we stop looking at users as if they make decisions alone. In real products, people are constantly reacting to systems that are also shaping them. Users respond to defaults, incentives, AI suggestions, privacy friction, social pressure, platform rules, and competitor moves. That is exactly why game theory can be so useful in UX research. It gives us a way to study interaction as a system, not just as isolated behavior. This matters because many UX problems are not simply usability problems. They are strategic problems. Why do users keep clicking “Accept All” even when they care about privacy? Why do people overtrust AI after a few good experiences? Why do communities fail to cooperate even when cooperation would help everyone? Why do some product patterns persist even when they create poor experiences? Game theory helps us answer those questions by focusing on players, strategies, and payoffs. In UX terms, the players might be users, platforms, AI systems, companies, or regulators. The strategies are the choices available to each of them. The payoffs are what they gain or lose, such as convenience, time, trust, engagement, privacy, or revenue. Once you look at UX this way, many messy behaviors start to make more sense. A Nash equilibrium, for example, helps explain why unhealthy patterns can become stable. Repeated games help us think about trust over time, especially in human AI interaction. Evolutionary game theory helps explain how certain habits spread because they work well enough in practice. The prisoner’s dilemma helps us study cooperation and exploitation in social systems. Stackelberg models help us understand what happens when platforms move first by setting defaults, rules, or pricing and users respond afterward. Bargaining models help us examine whether value, control, or rewards are distributed fairly enough to keep people engaged. What I like most about game theory in UX is that it pushes us to ask better questions. Not just “Is this interface usable?” but “What behavior does this system reward?” Not just “Do users trust the AI?” but “When does trust become blind trust?” Not just “Does this feature increase engagement?” but “What kind of equilibrium is this product creating over time?”

In 1950, a twenty two year old Princeton doctoral student named John Nash proved something game theorists had not managed to establish for competitive situations generally. He showed that in any game with a finite number of players and strategies, a stable point exists where no player benefits from changing strategy alone. This built on earlier work by John von Neumann and Oskar Morgenstern, who had formalized game theory for cooperative and zero sum contests. Nash extended the framework to noncooperative games, where players act independently without binding agreements, a shift that broadened game theory's reach into economics, biology, and eventually computing. The connection to multi-agent artificial intelligence is real but requires care. In generative adversarial networks, a generator and discriminator train against each other, and researchers often describe the ideal training outcome using equilibrium language borrowed from Nash's framework. Self-play systems in reinforcement learning invoke similar reasoning. However, reaching a true Nash equilibrium in these settings is not automatic. Goodfellow and colleagues noted early on that GAN training can cycle or destabilize rather than converge cleanly. Nash equilibrium functions more as a theoretical target and diagnostic lens than a guarantee that competing AI agents will settle down. Nash's own life added a sobering dimension to this legacy. He spent decades navigating schizophrenia before receiving the 1994 Nobel Memorial Prize in Economic Sciences for the equilibrium concept, a reminder that foundational ideas often emerge from difficult and nonlinear paths. Citations: Nash, J. (1950). Equilibrium points in n-person games. Proceedings of the National Academy of Sciences. Nash, J. (1951). Non-cooperative games. Annals of Mathematics. Von Neumann, J., Morgenstern, O. (1944). Theory of Games and Economic Behavior. Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. (2014). Generative Adversarial Networks. Advances in Neural Information Processing Systems. #NashEquilibrium #GameTheory #MultiAgentAI #HistoryOfScience #PhilosophyOfScience

Through collaboration with researchers of economics, we evaluated ChatGPT, DeepSeek and other LLMs for their strategic reasoning capabilities. When AI agents are deployed in real-world applications, they often encounter decision-making scenarios requiring cooperation or competition with other entities —  a process known as strategic reasoning. Consider an AI agent participating in a high-stakes negotiation, such as allocating resources in a disaster relief effort among multiple parties. The agent must evaluate the needs and strategies of other agents, analyze available resources, and understand the broader context to make a single, impactful decision — hopefully it is a good decision!   Drawing on the Truncated Quantal Response Equilibrium (TQRE) from behavioral game theory (TQRE is more realistic compared to Nash Equilibrium based game-theoretic settings thus it can model human behaviors, or LLM behaviors in this study, more accurately), we tested 10 state-of-the-art LLMs on 13 abstracted real-world games, such as Prisoner’s Dilemma, Stag Hunt, Bayesian Coordination Games, and Signaling Games, etc. Below is the partial table just showing the results for the top two winners: DeepSeek-R1 and GPT-o1.   GPT-o1 consistently ranks among the top models in competitive (e.g., zero-sum games) and incomplete-information games (games simulating environments in which players must infer unknown elements, such as their opponent’s type or payoff structure), while DeepSeek-R1 demonstrates stronger performance in cooperative and mixed-motive games (e.g., Stag Hunt or Prisoner’s Dilemma), where decision-making involves balancing trade-offs rather than pure optimization. This suggests that GPT-o1 is optimized for rational, goal-oriented decision-making and adversarial reasoning. DeepSeek-R1 excels in cooperation and mixed-motive games, likely due to its enhanced ability to model social interactions and optimize mutual benefits under reinforcement learning rewards.   Two important conclusions based on this study: (1) Chain-of-Thought (CoT) promotes reasoning when a model’s capability is limited. However, it can introduce distractions for LLMs with higher reasoning levels, leading to suboptimal choices. (2) Superior reasoning ability does not necessarily yield desirable or ethical outcomes, highlighting the need for a balanced approach and calibration in future LLM development.

The trading floor has evolved, but the underlying mathematical game remains the same. Today’s electronic limit order books are the ultimate proving ground for applied game theory. When executing a large institutional block, the fundamental challenge is a variation of the classic "Prisoner's Dilemma" mapped onto liquidity: if you reveal your full size, the market will move against you (adversarial impact). If you hide too much, you risk non-execution and adverse selection (timing risk). This is where the math of algorithmic execution takes over, most notably formalized by frameworks like Almgren-Chriss. The execution problem becomes a stochastic optimization task, balancing market impact against price volatility. We can define the optimal execution trajectory x by minimizing the objective function: U(x) = E[x] + lambda * V[x] Here, E[x] represents the expected execution cost driven by temporary and permanent market impact, V[x$ is the variance of the cost representing timing risk, and lambda is the trader's risk aversion penalty. In a vacuum, this calculus is straightforward. But the modern market is an adversarial, multi-agent environment. Your execution algorithm is constantly interacting with other participants, including statistical arbitrageurs and market makers, who are solving their own optimization problems. They are attempting to reverse-engineer your trajectory x through subtle cues like order book imbalances, cancellation rates, and trade flow toxicity. Every parent order sliced into smaller child orders and routed to lit exchanges or dark pools is a strategic move in a continuous game of incomplete information. To survive, modern execution models cannot just be static schedules (like standard VWAP or TWAP). They must be deeply adaptive, utilizing stochastic control to dynamically adjust to the Nash equilibrium of the moment. How do you see the intersection of game theory and execution evolving next? Are we reaching the theoretical limits of what traditional mean-variance optimization can achieve in adversarial limit order books? #AlgorithmicTrading #GameTheory #QuantitativeFinance #CapitalMarkets #USMarketHistory #Mathematics #MarketMicrostructure

Tariffs are being levied–and rescinded. Equity markets are volatile. Bond markets are jittery. Sentiment is darkening. Rumors are swirling. This isn’t just political economy—it’s a multi-move, multi-level game. Over the past few days, the U.S. and China have played a fast-moving game with serious economic consequences. Here’s how a game theory lens can help you decipher what’s really happening—and what might happen next. 🎭 The Madman Theory 📌 Definition: A strategy where a player deliberately behaves in an irrational, unpredictable, or even dangerous way. Why it Matters: Trump claims China pays the tariffs. He treats trade deficits as losses. He swings from threats to calling Xi a friend. It’s unclear if he’s bluffing, posturing—or genuinely insane. The Upshot: Madman behavior breaks standard equilibria. It raises the cost of misreading intent, which pushes opponents toward caution and concessions, giving the madman an advantage. 🧠 Bayesian Game 📌 Definition: A strategic game with incomplete information about the other player’s “type” (aggressive, cautious, bluffing, etc.). Poker is a Bayesian game—you don’t know your opponent’s hand or strategy. Chess is not—everything is visible and symmetric. Why it Matters: U.S. and Chinese leaders are playing a Bayesian game. Neither side knows the other’s full preferences, constraints, or level of resolve. The Upshot: Each move sends a signal. Over time, each side updates beliefs, narrows uncertainty, and adjusts its strategy accordingly. 🧩 Multiple Audience Problem 📌 Definition: When a player must send different (sometimes conflicting) messages to different audiences at once. Why it Matters: Trump talks tough to foreign rivals, talks nostalgia to American voters, talks deals to investors, and talks vision to the media. The Upshot: Flip-flopping becomes a feature, not a bug. One moment hawkish, the next dovish—this duality lets Trump message-shift without fully committing, satisfying multiple constituencies at once. 🧱 Belief Inertia 📌 Definition: The tendency of players or audiences to resist updating their beliefs—even when presented with contradictory evidence. Why it Matters: Despite lawsuits, bankruptcies, falsehoods, or tariff U-turns, many Trump supporters continue to view him as decisive, patriotic, and successful. The Upshot: Most leaders would fear looking weak after backing down. But when belief inertia is strong, a player can be aggressive, passive, or inconsistent without penalty. That lack of accountability expands optionality—an underappreciated strategic edge. What looks erratic or illogical from a classical perspective may in fact be an edge in a game of ambiguity, noise, and belief distortion. In a world of Bayesian uncertainty and dissonant signaling, the rules of engagement shift. 👉 There’s more to unpack. Part 2 coming soon.

Everyone thinks game theory is about games. It's actually the foundation of every AI system that has to make decisions with incomplete information. Game theory entered AI research in the 1950s. Here’s the core idea: how does a rational agent choose the best move when the outcome depends on what others do? Chess computers use it. Autonomous vehicles use it. Negotiation agents, pricing algorithms, and multi-agent AI systems use it every day. The concept that matters: Minimax Minimax is a decision-tree algorithm. It assumes: You will always choose the move that maximizes your outcome. Your opponent will always choose the move that minimizes yours. The AI looks ahead (building a tree of possible futures) and works backward to find the move that's best even in the worst case. IBM's Deep Blue beat Garry Kasparov in 1997 using a version of this. Here’s why this matters beyond chess Modern AI agents operating in competitive or uncertain environments (financial trading, resource allocation, autonomous negotiation) are running variations of the same logic. The question isn't "what's the best outcome?" It's "what's the best move given that the environment will push back?" As AI agents operate more autonomously in 2026, the organizations that understand adversarial decision-making will design better guardrails, and better strategies. *** How are you thinking about AI systems that have to make decisions in environments that push back?