Economics 318 Class note: Basic Game Theory Concepts

Compiled by Melt van Schoor ([email protected]), March 2022

Copyright © Department of Economics, Stellenbosch University

1.Strategic Interaction

Game theory is the science of rational behaviour in strategic situations. This means that the outcome
(“payoff”) for an agent depends on the actions of other agents. Further, there must be mutual
awareness that each agent’s actions influence the others. Players care about what other players might
do and if they are rational they will attempt to work out what others will do based on their knowledge
of the other player’s behaviour (e.g. that they are rational). Given that players can affect each others’
outcomes, they can also potentially affect their behaviour.

2. Strategy

A strategy is a complete plan of action for one player that specifies what move(s) to make in a particular
game. Once a strategy is chosen, no other choices need to be made: You could give the strategy to
another person or a computer to carry out on your behalf and no further instructions would be required
from you, no matter what happens. A strategy must explain what to do for every possible contingency,
i.e. for every possible state that a game might be in where a player is required to make a choice,
whether or not those choices will actually need to be made or not.

In every game we model, each player will have a strategy set / strategy space. This is the collection of
possible strategies that that player can choose from. A strategy set might have a small number of
elements in it (choose A or B) or it might be infinite (choose a price).

3. Strategy profile (also called “joint strategy”)

When a game is played, each player chooses exactly one strategy. If each player in a game has chosen
one strategy each, and we collect the chosen strategies from different players, we call this combination
a strategy profile. For example, if person 1 chooses strategy G and person 2 chooses strategy H, then
the chosen strategy profile is (G, H).

3. Payoff

A payoff is a number that gives a measure of how much a given outcome is preferred by a player. It is
the game theory term for utility. It is a player’s objective to maximize her payoff. A rational player will
actually maximize her payoff. Payoffs numbers, like utility numbers, should in principle be deduced
from a person’s actual choices, e.g. if I observe a player choosing A when he had a choice of A or B, I
know that 𝑢(𝑎) ≥ 𝑢(𝐵). The numbers are in an important sense arbirarily scaled: If A is strictly
preferred to B, I can say assign 𝑢(𝐴) = 10, 𝑢(𝐵) = 1 or 𝑢(𝐴) = 10000, 𝑢(𝐵) = 9999 and both sets of
payoff numbers will reflect exactly the same information. The size of a utility difference does not matter
and has no meaning. (Later on, when we look at expected utility, size differences between utility
numbers will have meaning.)

Note that maximizing her payoff is the only thing a player cares about and the only thing that will drive
her behaviour. It is conceptually wrong to think of a player trading off payoffs for some other objective.
For example, in the real world, some people might be willing to accept a lower amount of wealth in
order to alleviate another person’s suffering, i.e. a person might be an altruist. But in game theory, we
would not say that a person accepts a lower payoff/utility because he is an altruist. If someone has
altruistic preferences, they should be reflected by the person’s payoffs, such that maximising payoffs
will then also entail to some degree ensuring the welfare of other people, thus reflecting the
appropriate preferences that the person has regarding outcomes for other people. (E.g. if a certain
action causes a colleague to die, an altruist might get a very low payoff from that outcome, while a
psychopath might not care).

Given that a player cares only about her payoff, it is wrong to think that a player will ever deliberately
behave to increase or decrease another player’s payoffs – they may affect another’s payoffs, but only
incidentally as they seek to maximise their own. In particular, there is no general sense in which people
try to “win” games, i.e. try to get a higher payoff than his opponent. In fact, the payoff numbers of two
players are generally not comparable, so a statement such as “Monty’s payoff is higher than Wawa’s”
has no general significance / meaning.

In some games we use simple concrete proxies for payoff numbers, such as money amounts, or the
number of ice creams that a person gets to consume. It is important to note that we are making
assumptions about people’s preferences when we use such numbers as payoff numbers. We will also
learn later in the course when we cover expected utility that we are making assumptions about
people’s risk preferences when we use simple money amounts as payoff numbers.

4. Rules of the game

The rules of the game are a precise description of a strategic interaction, and includes who the players
are, what actions they can take at various points in time and for various contingencies (i.e. given
everything that has happened up to the point where a decision is to be made). It also includes how
payoffs are assigned for each possible strategy profile. It also includes what all players know. The
knowledge is often implicit, but you can always assume that players have a full understanding of the
game as it is represented, i.e. players will fully understand the rules of the game.

5. Rational behaviour

Rational behaviour means that a person acts to obtain the best possible outcome for herself given the
circumstances. In strategic interaction situations, these circumstances include the behaviour of other
players, and potentially their responses to her behaviour. Rationality means that a person has an
infinite amount of computational ability, which is sometimes, depending on the context, highly
unrealistic.

In modern economic theory, rationality places constraints on the preferences a person may have. For
example a person who, given choices of outcomes, chooses B over A, then C over B and then A over C
is not rational according to the modern view, because it is not possible to model such a person’s
behaviour as maximising some consistent objective. There are some other rules of consistency; we do
not have to worry about the detail here.

As game theoreticians, we often choose to model people as being rational and having consistent
preferences. This is not because we believe that real people are necessarily always fully rational –
sometimes they are very irrational! – but because it is possible to get some answers by modeling using
this approach. If we understand what we are doing, these answers can guide us to understanding real
people’s behaviour as well, even if they are not rational and behave differently to what the model’s
solution suggests. We can understand people’s imperfectly rational behaviour much more easily once
we have studied rational behaviour. Also, people may not be perfectly rational, but usually, given some
time to learn and improve their behaviour in response to real incentives, their behaviour will tend
towards rationality, even if they never quite get there it is still useful to understand the theoretical
endpoint of such a learning and adaptation process.
6. Common knowledge

An assumption often made is that all the players in a game have common knowledge of the rules of the
game and of each other’s rationality. To explain this, consider an example game with two players,
Palesa and Abe. If the Palesa and Abe have common knowledge, then Palesa is rational, and knows the
rules of the game. In addition, Abe knows that she is rational and knows the rules of the game. In
addition, Palesa knows that Abe knows that she is rational and knows the rules of the game. In addition,
Abe knows that Palesa knows that Abe knows that she is rational and knows the rules of the game. This
goes on for an infinite “levels” of knowledge. It is also true of Abe’s rationality and knowledge of the
rules of the game.

5. Equilibrium

An equilibrium is a situation where some forces relevant to a particular setting are in some sense “in
balance”. In physics there are many simple and complicated examples of equilibrium: If you sit on a
chair, the force of gravity pulling you downwards is exactly equal to the upwards force the chair exerts
on your seating area, so that you do not move up or down.

In economics models, the forces are the incentives acting to change people’s economic behaviour. So
if there are unexploited gains from trade in a competitive demand and supply model, the situation is
not an equilibrium because some people can do better for themselves by changing their behaviour.
Note that there is typically interaction between behaviour of people and the optimality of choices of
other people. If consumers decide not to buy any ice creams, it may be optimal for an ice cream seller
to stay at home (rather than spend money to travel to the beach with an ice cream cart). Equilibrium
is when people’s choices are optimal simultaneously given their context, including the behaviours of
other people. In game theory, an equilibrium is when all players are choosing optimal strategies at the
same time, where an optimal strategy is the strategy that will give the relevant player the highest
possible payoff in a game given the strategies of the other players.

Getting to an equilibrium may require a process of adjustment, readjustment, yet more readjustment,
etc., until finally, an equilibrium may be reached (or sometimes not). We can study the process of
getting to an equilibrium, but we often just study the equilibriums themselves, without explicitly
addressing the question of how the equilibrium is reached. It is useful to keep this question in the back
of your mind when studying various kinds of equilibriums in various games. Near the end of the course
we study the process of getting to an equilibrium in more detail when we consider evolutionary game
theory, where we do not assume that people behave rationally, but that they slowly adjust their
behaviour in response to incentives, while other players do the same.

6. Solution concept

The meaning of rationality in game theory can have subtly different meanings, amongst others because
we can make subtly different assumptions about the knowledge that players have and the context in
which they make rational decisions. There are therefore various different ways of thinking about what
“the solution” of a game is, meaning what strategies we think rational players in that strategic
interaction would choose when they play the game. We will study different solution concepts in this
course, including backwards induction solutions, dominant strategy solutions, iterated elimination of
strictly dominated strategies solution and the Nash equilibrium.

7. Conflict and cooperation

In some games, especially sports and board games, the rules are set up so that the only way in which a
player can get a better outcome (often simply, “win the game”) is to make the other player do worse
(and maybe “lose the game”). These are called zero sum games and they feature completely conflicting
preferences between the players. For example, if there are three outcomes, A, B and C, and player one
prefers A over B over C, then in a zero sum game, player 2 prefers C over B over A, i.e. a directly opposite
ranking of the three outcomes. Most interesting games are not like this though. On the other extreme,
there could be a perfect correspondence between preferences of two players, so that they have
identical rankings of outcomes in their preferences – in these games there is typically much scope for
cooperation between the players as there are immediate incentives to behave in a wat that will benefit
the other player. A lot of games also have both the potential for conflict and potential for cooperation.
This occurs when there is partial correspondence of preference rankings. For example, player 1 might
prefer A over B over C and player 2 might prefer A over C over B. In this case, they might be able to
cooperate to obtain A or there might be conflict over whether the outcome is B or C.

8. Once-off and repeated games

When we build a model of a game, we usually assume that players’ payoffs are totally determined in
that interaction, and nothing that happens afterwards could possible matter. This requires us to also
assume that the choices made in a game can have no impact on any future interactions that any of the
players might care about. The simplest way to conceptualise this is to suppose that the two players
playing the game will only play the game once and afterwards they will never meet or interact in any
way ever again. In such settings, we call the game a one-shot game. But is this realistic for real-world
scenarios of interest? Not usually, because most strategic interactions are with people that we interact
with not just once, but with some regularity. It is possible to model such repeated interactions in game
theoretic models, which we will do in this course for the Prisoner’s dilemma. It is important to note
that, due to strategic interaction over the different rounds over which the game is played by the same
two players, the solution to such repeated games is typically very different to the corresponding one-
shot game. For example, we can get a solution to the repeated prisoner’s dilemma game where two
entirely rational and selfish players actually cooperate (i.e. not snitch on their partner-in-crime).

9. Sequential-move games and simultaneous-move games

If a game has any notion of timing of moves, e.g. player 1 moves first, then player 2 (think Chess), then
the game has sequential moves. If a game features all players making a move at the same time, the
game has simultaneous moves (think Rock Paper Scissors). However, it is not vital that both players
literally move at the same instant in time. We consider games in which players make their choice
without knowing the choice of the other players as simultaneous-move games. In that case, both moves
are independent of one another in the sense that their moves cannot be conditional on the other
player’s moves, and we can model it as a simultaneous move game.

10. Perfect information

If all players have observed all moves of other players prior to (or happening at the same time as) their
own moves, the game has perfect information. No moves that are relevant to a player for deciding on
the optimal choice at a given point in the game are “hidden” from him. If something happens that is
payoff-relevant for some player, but the player cannot observe it when making her choice, the game
has imperfect information. This can either be because there is some kind of external uncertainty in
the game or because of strategic uncertainty. External uncertainty is some random event, for example
a coin flip or tomorrow’s weather, that affects the payoffs that the players will get from a given choice
of strategies. Strategic uncertainty is if the other player’s strategy is not observed when a player makes
her choice. This happens in simultaneous-move games. Importantly, a player may have good
knowledge about what strategy the other player is playing, but because he cannot physically observe
the actual choice, he cannot make his own strategy conditional on the other player’s strategy, e.g.
decide that if the other player plays Left, I will play Up, and if the other player plays Right, I will play
Down. He simply has to choose Up or Down.

11. Incomplete information (also called asymmetric information)

When one player has some knowledge about himself that the other does not possess, we have
incomplete information. This “secret” knowledge is usually modelled so that the other player does not
know what payoffs the first will get from a given strategy profile. This then means that the other player
may not know how the player might respond to various incentives. Players still understand the rules of
the game. We will only deal with games of complete information in this course, so you can always
assume that all players know everything about another player that the other player knows herself,
particularly the payoffs for each possible outcome.

12. Cooperative and noncooperative game theory

Players may behave differently in many situations if they are able to make binding agreements, i.e. sign
a contract and be sure that it will be respected by all parties. In the prisoner’s dilemma game, for
example, both players prefer the outcome that would happen if both players cooperate, rather than
the usual solution’s outcome, which is that each player tries to maximise only his own payoff, leading
both players to the bad uncooperative outcome. But binding agreements need an enforcement
mechanism, e.g. law courts and prisons for when people break contracts. In typical situations, such
binding contracts are not available, and players have to make choices knowing that all players will be
free to make any choice they wish to when they choose their strategies, regardless of any promises
they made in the past.

If players are unable to make binding agreements, we have a non-cooperative game, which we study
using non-cooperative game theory. We will only study non-cooperative game theory in this course,
for two reasons: 1) It is the more basic theory that will feel more familiar to students that have studied
“standard” economic theory, i.e. individual agents acting rationally on their own, and 2) it is possible
to model the enforcement mechanisms that might make binding agreements possible using non-
cooperative game theory, so the latter is in a sense the more “general” theory. Cooperative game
theory is a large and interesting field on its own, however, for those interested in further study.

13. Model building versus model solving

Model building is the skill and art of representing strategic interactions from the real world as precisely
specified, simplified game theoretic models. It involves gathering information, making simplifying
assumptions, specifying players, possible strategies and outcomes, and – importantly – assigning payoff
numbers (based on data and/or a good understanding of the involved persons preferences) for each
player. A model is always only a partial representation of a real-world situation, and is typically designed
to answer specific questions or illuminate specific relations or aspects of a situation.

Once a model is fully specified, solving it is a matter of applying rigorous logic and mathematical
techniques (which we will study in this course) to find a solution conforming to the requirements of the
solution concept of interest. Solving a model does not involve judgement or any knowledge of the real
world apart from that which is captured in the model. After solving a model, we could go back to the
first step and build a new, improved model, understanding that we are creating a new model which will
likely have a new solution. This iterative process can help us understand the link between model design
choices and results, a particular benefit of the modeling approach because it potentially helps to focus
our attention on and understand how essential features of the real world lead to certain outcomes.