Game Theory Practice Problems

Working Through Game Theory Practice Problems

Most people approach game theory practice problems the same way they approach calculus homework. They stare at the payoff matrix, try to plug numbers into whatever formula their professor mentioned in passing, and hope something reasonable comes out. This doesn't work. Game theory problems aren't about computation, they're about reasoning through strategic interaction. You need to figure out what each player would do given what they think the others will do. The math is secondary. The structure is everything. I spent years working with these kinds of problems in operations research, and I can tell you the most common mistake I see isn't computational, it's conceptual. People miss the iteration in iterated elimination of strictly dominated strategies. They eliminate one strategy, then stop, when the removal of that strategy may create new dominance relationships for the remaining players. I once had a colleague working on a procurement auction model where the suppliers had three bidding strategies each. He eliminated the low bid as dominated, then got stuck because he didn't realize that after removing the low bid, the medium bid became dominated relative to the high bid in certain opponent configurations. Took him three days to spot. I spotted it in twenty minutes because I'd made the same mistake on a transportation game back in 2014 and learned not to stop after one pass. When you're working with Game Theory Practice Problems, start by identifying the game type. Is it simultaneous or sequential? Complete or incomplete information? One shot or repeated? These classifications determine which solution concepts apply. A prisoner's dilemma structure demands a different analytical approach than a battle of the sexes coordination game, even though both are 2x2 normal-form games. Mixing up the solution frameworks is another common error I see regularly. Nash equilibrium, dominant strategy equilibrium, and Pareto efficiency are not interchangeable. They answer different questions.

Step-by-step approach to solving these problems

Write out the full game representation before doing anything else. For normal-form games, draw the payoff matrix with clear labels. For extensive-form games, draw the game tree. This seems obvious but people skip it and jump straight to reasoning, which introduces errors. I keep a spreadsheet template that I fill in with player names, strategy sets, and payoff functions before I ever start analyzing. It takes about five minutes and has saved me from multiple incorrect conclusions over the years. Next, check for dominant strategies. A dominant strategy exists when one action yields a strictly higher payoff than every other action, regardless of what the other players do. Mark these clearly. If a player has a dominant strategy, they'll play it. That's your first anchor point. After identifying dominant strategies, look for weakly dominated strategies. This is where most students and practitioners lose accuracy. Weak dominance is trickier because eliminating weakly dominated strategies can remove Nash equilibria that are still valid. Use strict dominance for elimination when possible. Save weak dominance for when you have no other way forward and be transparent about which equilibria might disappear. For Nash equilibrium identification, use the best response method rather than brute-forcing every cell. For each player, circle the best response to every possible strategy profile of the other players. Cells where both players are playing a best response to each other are Nash equilibria. This is faster and less error-prone than checking the formal definition for every outcome. In a 4x4 game, this cuts your analysis time from roughly ten minutes to about three.

Mixed strategy equilibria are where things get messy. The standard approach is to make the opponent indifferent between their pure strategies. Set up the expected payoff equations and solve. The trap here is assuming a mixed strategy equilibrium always exists or that it's always unique. In many real-world games with continuous strategy spaces or asymmetric information, mixed equilibria can be non-unique or impossible to express in closed form. I encountered this in a routing game where two networks were choosing bandwidth allocations under congestion. The standard indifference calculation produced two mathematically valid mixed equilibria, but only one was stable under small perturbations. The other was a saddle point in the strategy space. I resolved it by introducing a stability criterion based on best-response dynamics convergence, which filtered out the unstable equilibrium.

Common pitfalls that waste time

One major pitfall is confusing correlated equilibria with Nash equilibria. A correlated equilibrium allows players to condition their strategies on a shared random signal, which expands the set of possible outcomes beyond what Nash equilibria permit. If a practice problem involves a correlation device, treating it as a standard Nash problem will give you the wrong answer. Check whether the problem statement mentions a mediator, a signal, or correlated recommendations. If it does, you need the correlated equilibrium framework, not the Nash framework. Another pitfall is assuming that every game has a pure-strategy Nash equilibrium. Finite games always have at least one Nash equilibrium in mixed strategies, but not all have one in pure strategies. The matching pennies game is the textbook example. When you encounter a 2x2 game with no pure-strategy equilibrium, immediately set up the mixed strategy calculation rather than searching indefinitely for an equilibrium that doesn't exist. Backward induction in sequential games is straightforward in theory and tedious in practice. The problem is that people make arithmetic errors when working through long trees, and they often misidentify the information sets. When a player cannot distinguish between two nodes in the same information set, you must treat those nodes as belonging to a single decision point. I've seen students apply backward induction as if perfect information existed when the problem actually specified imperfect information. This changes the solution completely. Double-check the information set notation before beginning your backward induction procedure.

When standard methods break down

Game theory practice problems become significantly harder when you move beyond two-player, finite, zero-sum or general-sum games. Three-player games with continuous strategies, games with incomplete information that require Harsanyi transformation, and games with more than two stages of commitment can push standard textbook methods past their breaking point. In these cases, computational approaches become necessary. Software like Gambit or MATLAB toolboxes can solve for equilibria in games that are analytically intractable. I typically use Gambit for games up to about 5x5 normal form, beyond which the computation time starts to climb and I switch to custom scripts. For Bayesian games with continuous type distributions, I've found that discretizing the type space into approximately twelve bins gives reasonable accuracy while keeping the problem solvable by standard equilibrium-finding algorithms. Going finer than twelve bins usually doesn't improve results meaningfully but does increase computation time substantially. The biggest limitation of practicing with textbook problems is that they're almost always cleaner than real strategic situations. Real games have unmodeled players, ambiguous information structures, and strategies that aren't neatly enumerable. Working through practice problems builds your mechanical skills, but it won't prepare you for the ambiguity of applied work. I'd recommend supplementing your practice with case studies from industrial organization, political science, or mechanism design literature where the game structure isn't fully specified and you have to make modeling choices yourself. This gap between clean exercises and messy reality is something that most learning resources don't address adequately.

Resources for structured practice

The most reliable sources for Game Theory Practice Problems are your course textbook's problem sets, supplemented by materials from MIT OpenCourseWare and the Game Theory.net problem archive. Avoid random websites that pump out unverified problems with answer keys that contain errors. I've lost count of how many times I've corrected an answer key from a popular problem set that had the wrong equilibrium due to a sign error in the payoff matrix. If you're self-studying, the Osborne and Rubinstein textbook problems are among the most rigorously checked available, though they're still quite dense and will take considerable time to work through properly. For sequential games specifically, working through examples with actual game trees drawn on paper rather than digitally seems to produce better intuition. There's something about the physical act of drawing branches and information sets that reinforces the structure in a way that digital tools don't replicate for most people. I spent about six weeks doing nothing but drawing and solving extensive-form games by hand, and it directly improved my ability to recognize structural patterns in much more complex problems later on. The practical timeline for becoming competent with standard game theory problems—Nash equilibrium in 2x2 and 3x3 games, backward induction in multi-stage games, basic mixed strategy calculations—is somewhere between four and six weeks of focused practice at about an hour per day. More complex problems involving Bayesian games or correlated equilibria will take additional time and typically require a second pass through the material once you have the foundations solid. Rushing through problems without checking your work against a reliable solution set will embed incorrect reasoning patterns that are harder to unlearn than to prevent in the first place.