Uncertainty and Information

Dr Bryony Reich

Topic 2: An Introduction to Uncertainty

To avoid distracting others, please turn off or silence your

phone and put it in your bag. Laptops/tablets should be in
airplane mode and not distract others in the class.
Optional Textbook Reading

• These notes draw from Steve Tadelis “Game Theory: An Introduction” Chapter 2
Agenda:

• Introducing Uncertainty
• Lotteries
• Payoffs
• Rational Choice

• Example: Investing in an MBA

• Probability Recap

• Example: Continuous Outcomes

• Example: The Value of Information

Introducing the Idea of Uncertainty
• The decision problem we set up in Topic 1 was very “minimal.”

• Most decision problems are complex.

• To make predictions about how people behave and how the world works, we probably want to add more
realistic elements of decision problems…but we want to do so in a way that is tractable.
Introducing the Idea of Uncertainty
• Take our example from Topic 1 of a manager of a firm choosing to embark on major R&D or not.

• What is the outcome if he invests in major R&D?

➢ R&D is costly, so the firm will need to pay the cost of that investment.
➢ R&D may result in an innovative valuable new product that makes high profits for the firm.
➢ R&D may produce nothing useful.

• There is uncertainty over the outcome of R&D. This uncertainty is a fundamental feature of any R&D
investment.

• How do we think the manager will approach this uncertainty?

• It’s not very useful if he just thinks “well this could happen or that could happen”. He needs a framework for
thinking about his problem.

• We want to add uncertainty into our decision problem framework…but in a manageable way.

• To do this, we will think of these outcomes of R&D as random events and put probabilities on those events.
Introducing the Idea of Uncertainty
• Put uncertainty into the decision problem.

• Player: Manager.

• Action set: A = {𝑟, 𝑠}, where 𝑟 is invest in R&D and 𝑠 is do not invest.

• Suppose that there are two possible outcomes: the product is successful (makes a profit of 10) or becomes
outdated (makes a profit of 0) so 𝑋 = {0,10}.

• If the manager does not invest in R&D then the product stays the same and is successful with probability 0.5
and not with probability 0.5

• If he invests in R&D then the product may improve and be successful with probability 0.75 or not with
probability 0.25.

• There is no longer a one-to-one correspondence between actions and outcomes.

*We are ignoring the cost of the R&D for now.

Lotteries
• In choosing between 𝑟, invest in R&D, and 𝑠, do not invest, we think of the manager as choosing between two
lotteries.
➢ The player can choose “lottery 𝑟” which pays 0 with probability 0.25 and pays 10 with probability 0.75.

➢ Or the player can choose “lottery 𝑠“ which pays 0 with probability 0.5 and pays 10 with probability 0.5.

• We can think of “Nature” as another player who chooses among the random outcomes and can we describe
the player’s decision problem using a decision tree:

0.75 10
Nature

r
0.25 0
Player
0.5 10
s

Nature 0.5 0
Lotteries
• Suppose we have a decision problem with 𝑛 different outcomes which are denoted 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 }.

• A lottery over those outcomes gives a probability of each outcome occurring.

• The probability of any given outcome occurring cannot be negative.

• The sum of probabilities over all outcomes must sum to 1 (that is, one of them has to occur).

Definition.
A simple lottery over outcomes 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 } is a probability distribution 𝑝 = (𝑝 𝑥1 , 𝑝 𝑥2 , … , 𝑝(𝑥𝑛 )),
where 𝑝(𝑥𝑘 ) denotes the probability that outcome 𝑥𝑘 occurs. The probability that 𝑥𝑘 occurs is 𝑝 𝑥𝑘 ≥ 0 for each
𝑘 ∈ {1, … , 𝑛}, and σ𝑛𝑘=1 𝑝(𝑥𝑘 ) = 1.

• We can use our framework then to say that when a player chooses an action 𝑎 ∈ 𝐴, he is choosing a lottery
over the outcomes.
Lotteries
• In the R&D example the outcomes are 𝑋 ∈ {0,10}.

• If the manager chooses action 𝑟 he faces a lottery over the outcomes where 𝑝 0 = 0.25, 𝑝 10 = 0.75.

• If the manager chooses action 𝑠 he faces a lottery over the outcomes where 𝑝 0 = 0.5, 𝑝 10 = 0.5.

0.75 10
Nature

r
0.25 0
Player
0.5 10
s

Nature 0.5 0
Lotteries
• Suppose an action results in one outcome only not probabilities of different outcomes. How can we
write that in terms of a lottery?

➢ That outcome has prob 1 and the rest have prob 0.

• Can add an example from section 2.1.2 of compound lotteries.

• Example of lotteries over continuous outcomes 2.1.3

Payoffs
• In the R&D example the outcomes are 𝑋 ∈ {0,10}.
➢ If the manager chooses action 𝑟 he faces a lottery 𝑝 0 = 0.25, 𝑝 10 = 0.75.
➢ If the manager chooses action 𝑠 he faces a lottery 𝑝 0 = 0.5, 𝑝 10 = 0.5.

• Question: Which action will the manager choose?

0.75 10
Nature

r
0.25 0
Player
0.5 10
s

Nature 0.5 0
Payoffs
• Add some costs of R&D into the example. Suppose the costs of R&D are 1.
➢ Now if the manager chooses action 𝑟 the profit from success is 9 and the profit if not successful is -1.
➢ If the manager chooses action 𝑠 the profit from success is still 10 and the profit if not successful is still 0.

• How can you write this as a choice over lotteries?

➢ Outcome set 𝑋 = {−1,0,9,10}.
➢ Lottery from choosing action 𝑟 is (𝑝 −1 = 0.25, 𝑝 9 = 0.75, 𝑝 0 = 0, 𝑝 10 = 0).
➢ Lottery from choosing action 𝑟 is (𝑝 −1 = 0, 𝑝 9 = 0, 𝑝 0 = 0.5, 𝑝 10 = 0.5).

Next question: Which action will the manager choose 𝑟 or 𝑠? How will he evaluate a choice over different
lotteries? 0.75 9
Nature

r
0.25 -1
Player
0.5 10
s

Nature 0.5 0
Payoffs
• How should a player evaluate and compare different lotteries?

• The framework we use, called “expected utility theory”, thinks of players as valuing an “average” over the different
possible outcomes of the lottery.

• Let 𝑢(𝑥) be the player’s payoff from an outcome in 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 }.

• Let 𝑝 = (p1 , p2 … , pn ) be a lottery over those outcomes where 𝑝𝑘 denotes the probability of outcome 𝑥𝑘
occurring, pk = 𝑝(𝑥 = 𝑥𝑘 ).

Definition.
The player’s expected payoff from the lottery 𝒑 over outcomes X is
𝑛

𝐸 𝑢 𝑥 𝑝 = 𝑝1 𝑢 𝑥1 + 𝑝2 𝑢 𝑥2 + ⋯ + 𝑝𝑛 𝑢 𝑥𝑛 = ෍ 𝑝𝑘 𝑢 𝑥𝑘 .
𝑘=1

• The player’s expected payoff from a lottery 𝐸 𝑢 𝑥 𝑝 is just the sum of the payoff from each outcome 𝑢(𝑥𝑘 )
weighted by its probability of occurring 𝑝𝑘 .
Payoffs - Notation
• Let 𝑢(𝑥) be the player’s payoff from an outcome in 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 }.
• Let 𝑝 = (p1 , p2 … , pn ) be a lottery over those outcomes where 𝑝𝑘 denotes the probability of outcome 𝑥𝑘
occurring, pk = 𝑝(𝑥 = 𝑥𝑘 ).

Definition.
The player’s expected payoff from the lottery 𝒑 over outcomes X is
𝑛

𝐸 𝑢 𝑥 𝑝 = 𝑝1 𝑢 𝑥1 + 𝑝2 𝑢 𝑥2 + ⋯ + 𝑝𝑛 𝑢 𝑥𝑛 = ෍ 𝑝𝑘 𝑢 𝑥𝑘 .
𝑘=1

• Notice there are two payoff functions floating around. Be aware of the difference.
• A payoff function 𝑢(𝑥) over outcomes in 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 }.
• An expected payoff 𝐸 𝑢 𝑥 𝑝 from the lottery 𝑝 over outcomes in 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 }.

• We will sometimes write a player’s expected payoff as 𝐸[𝑢(𝑥)] when it is clear what lottery/probabilities we are
referring to.
• Sometimes 𝑢(x) is called the Bernoulli utility function and 𝐸[𝑢(𝑥)] the von Neumann-Morgenstern utility function.
• Sometimes 𝐸[𝑢(𝑥)] is represented by E𝑈 or 𝑈.
Payoffs
• What is the manager’s expected payoff from each action (and resulting lottery) in the R&D example?

• Suppose his payoff is equal to his profit so 𝑢 𝑥 = 𝑥 .

• His expected payoff from action 𝑟 is 𝐸 𝑢 𝑥 𝑟 = 0.75𝑢(9) + 0.25𝑢 −1 = 0.75 × 9 + 0.25 × −1 = 6.5.

• His expected payoff from action s is 𝐸 𝑢 𝑥 𝑠 = 0.5𝑢(10) + 0.5𝑢(0) = 0.5 × 10 + 0.5 × 0 = 5.

• The manager will choose to invest in R&D.

0.75 9
Nature

r
0.25 -1
Player
0.5 10
s

Nature 0.5 0
Payoffs – Cardinality Matters
• When we talked about simple decision problems in Topic 1, we said that we can represent the preference
relation with a payoff function and we just require that the payoff function assigns a higher number to more
preferred outcomes.

Definition.
A payoff function 𝑢: 𝑋 → ℝ represents the preference relation ≽ if for any pair 𝑥, 𝑦 ∈ 𝑋, 𝑢 𝑥 ≥ 𝑢(𝑦) if any only if
𝑥 ≽ 𝑦.

• The value of the payoff function is irrelevant since it simply represents an ordinal ranking.

• When we move into the world of expected payoffs, this is no longer true. Cardinality matters.

• In expected payoffs, the intensity of preferences matters.

• See why that is the case using our R&D example…

Payoffs – Cardinality Matters
• Example. In the R&D example we said let’s suppose payoffs are equal to profits so 𝑢 10 = 10 > 𝑢 9 = 9 >
𝑢 0 = 0 > 𝑢 −1 = −1.

• Suppose we keep the ordering the same, but instead payoffs are 𝑢 10 = 10 > 𝑢 9 = 9 > 𝑢 0 = 0 >
𝑢 −1 = −8, so the manager receives a more negative payoff if profits are negative.

• His expected payoff from action 𝑟 is now 𝐸 𝑢 𝑥 𝑟 = 0.75𝑢 9 + 0.25𝑢 −1 = 0.75 × 9 + 0.25 ×
−8 =4.75.

• His expected payoff from action s is the same 𝐸 𝑢 𝑥 𝑠 = 0.5𝑢 10 + 0.5𝑢 0 = 0.5 × 10 + 0.5 × 0 = 5.

• He prefers to take action 𝑠.

• What is happening here? Expected payoffs weight the payoffs from the different outcomes and so the
value that the player puts on those outcomes matters.

• Now choice of payoff function 𝑢() really matters!

Back to the Rational Choice Paradigm
• We assume decision-makers are “rational” in the sense that they 1) grasp the entirety of the decision-
problem and 2) choose the action that maximizes their well-being.

• Embedded in the rational choice paradigm is the assumption that a player completely understands their
decision problem:
➢ knows the set of actions 𝐴 they can choose from;
➢ Knows the set of outcomes 𝑋;
➢ knows how the actions translate into outcomes;
➢ and has rational preferences over those outcomes.

• In a world with uncertainty, the assumption that the player “knows how the actions translate into outcomes”
implies that he knows for each action he could take the probability of each outcome occurring.

• That is, if he takes action a ∈ 𝐴 he knows the probabilities p1 , p2 … , pn of each outcome 𝑋 =

{𝑥1 , 𝑥2 , … , 𝑥𝑛 } occurring, where 𝑝𝑘 denotes the probability of outcome 𝑥𝑘 occurring.

• Is this a strong assumption? Is this a strange way to think about uncertainty?

Back to the Rational Choice Paradigm
• In a world with uncertainty, the assumption that the player “knows how the actions translate into outcomes”
implies that he knows for each action he could take the probability of each outcome occurring.

• A rational player chooses the action which maximizes his expected payoff.

• To put it in terms of lotteries, an action a ∈ 𝐴 results in a lottery with probabilities p1 , p2 … , pn of each

outcome 𝑋 = {𝑥1 , 𝑥2 , … , 𝑥𝑛 } occurring.

• The player chooses the action which results in the lottery which maximizes his expected payoff.

Definition.
A player who faces a decision problem with payoff function 𝑢(⋅) over outcomes is rational if he chooses an action 𝑎∗ ∈ 𝐴
that maximizes his expected payoff, 𝐸 𝑢 𝑥 𝑎∗ ≥ 𝐸[𝑢(𝑥)|𝑎] for all 𝑎 ∈ 𝐴.
Example: Investing in an MBA
• You have been working for a few years after your undergraduate and now you are thinking about going to do a
prestigious Masters or MBA which will propel your career.

• The cost of doing the MBA is 10. This cost takes into account your opportunity costs of doing the MBA, so the
money you would have earnt by working instead, etc.

• The benefit of doing the MBA is your value of your future stream of income which depends on market
conditions when you re-enter the labour market.
➢ If the labour market is strong then your income without an MBA is 32. Without an MBA is 12.
➢ If the labour market is average then your income 16. Without an MBA is 8.
➢ If the labour market is weak then your income is 12. Without an MBA is 4.

• The labour market is strong with probability 0.25, average with probability 0.5, and weak with probability 0.25.

• Your payoff 𝑢() is equal to your income minus the costs, that is 𝑢 𝑥 = 𝑥.
Example: Investing in an MBA
• You have a choice of two actions: do the MBA or not.

• Write down the lotteries that result from these actions.

• The cost of doing the MBA is 10.

• If the labour market is strong then your income without an MBA is 32. Without an MBA is 12.
• If the labour market is average then your income 16. Without an MBA is 8.
• If the labour market is weak then your income is 12. Without an MBA is 4.
• The labour market is strong with probability 0.25, average with probability 0.5, and weak with probability 0.25.

• If you do the MBA you face a lottery with: income 32-10 with probability 0.25, income 16-10 with probability
0.5, income 12-10 with probability 0.25.

• If you don’t do the MBA you face a lottery with: income 12 with prob 0.25, income 8 with prob 0.5, income 4
with prob 0.25.
Example: Investing in an MBA
Strong labour market

0.25 32-10 Average labour market

Nature
0.5
16-10
Get MBA Weak labour market
0.25 12-10
Player
0.25 12
Don’t get
MBA 0.5 8
Nature
0.25 4
Example: Investing in an MBA
• You have a choice of two actions: do the MBA or not.

• If you do the MBA you face a lottery with: income 32-10 with probability 0.25, income 16-10 with probability
0.5, income 12-10 with probability 0.25.

• If you don’t do the MBA you face a lottery with: income 12 with prob 0.25, income 8 with prob 0.5, income 4
with prob 0.25.

• Calculate the expected payoff of each action.

• Expected payoff from an MBA 𝐸 𝑢 𝑥 𝑀𝐵𝐴 = 0.25 × 32 − 10 + 0.5 × 16 − 10 + 0.25 × 12 − 10 = 9

• Expected payoff from not: 𝐸 𝑢 𝑥 𝑛𝑜 𝑀𝐵𝐴 = 0.25 × 12 + 0.5 × 8 + 0.25 × 4 =8

• You choose the action that maximizes expected payoff: you do the MBA.
Probability Recap
• A random variable is a variable that can equal one of many different values, each with some probability. For
example:

1 1
• A coin toss, can be either heads or tails, each with probability 2, Pr ℎ𝑒𝑎𝑑 = Pr 𝑡𝑎𝑖𝑙 = 2.
1
• A fair die roll, will take a value 1,2,3,4,5 or 6, each with probability 6.
• The temperature in Cambridge tomorrow will take a real number somewhere between -5 Celsius and 15
Celsius, with some values more likely than others.

• The coin toss or die roll is an example of a discrete random variable – it will result in one of a discrete set of values.

• Temperature tomorrow is a continuous random variable – it will result in an outcome from an interval or continuum
of values. E.g. −5,15 .

• An event is an outcome or set of outcomes that can occur. E.g. Event A is “the die shows an odd number.” Then
1 1
Pr 𝐴 = 2 . Event B is “the die shows 1 or 2” with Pr 𝐴 = 3.
Probability Recap
• The probability distribution function for a discrete random variable is a function that describes the relative
frequency of each value that can be realised by the random variable.

• That is, the probability distribution function gives the probability Pr 𝑥 ∈ [0,1] of each outcome 𝑥 ∈ 𝑋.

• The sum of all the probabilities of the outcomes must add up to 1.

1
• E.g. a roll of a die 𝑋 = {1,2,3,4,5,6} with Pr 𝑥 = 6 for each 𝑥 ∈ {1,2,3,4,5,6}.

• The cumulative distribution function 𝐹(𝑥0 ) for a random variable describes the probability that the realised
value 𝑥 is less than or equal to 𝑥0 . That is 𝐹 𝑥0 = Pr{𝑥 ≤ 𝑥0 }.

1
• E.g. if the random variable is generated by the roll of a die then 𝐹 2 = 3, the probability that the outcome is
1 1
less than or equal to 2 is 3. Similarly 𝐹 2.9 = 3, the probability that the outcome is less than or equal to 2.9 is
1
3
.
Probability Recap
• For a discrete random variable, the cumulative and probability distribution functions are related according to
𝐹 𝑥0 = σ𝑥≤𝑥0 𝑝(𝑥).

• Any cumulative distribution function must satisfy

• If 𝑥2 > 𝑥1 then 𝐹 𝑥2 ≥ 𝐹 𝑥1
• 𝐹 −∞ = 0, 𝐹 ∞ = 1
• 1 − 𝐹 𝑥0 = Pr 𝑥 > 𝑥0

• The expected value/mean of a discrete random variable is given by σ𝑥∈𝑋 𝑥𝑝(𝑥).

• The expected value of a function u(·) of a discrete random variable is given by σ𝑥∈𝑋 u(𝑥)𝑝(𝑥).
Probability Recap
• A continuous random variable has Pr 𝑥 = 0. That is each outcome has a probability 0 of occurring.

• E.g. the temperature in Cambridge tomorrow which will be in the interval −5,15 , this can take one of an
uncountable number of values and so the probability of any one of them occurring is 0.

• E.g. a random variable that is uniformly distributed over the interval [0,6]. Every number in this interval is
equally likely to be realised. The probability of any given number occurring is 0.

• Because, for a continuous random variable, each outcome has a probability 0 of occurring, we cannot
represent this with a probability distribution.

• We instead use a cumulative distribution function.

• The cumulative distribution function 𝐹(𝑥0 ) for a random variable describes the probability that the realised
value 𝑥 is less than or equal to 𝑥0 . That is 𝐹 𝑥0 = Pr{𝑥 ≤ 𝑥0 }.
Probability Recap
• If the cdf F is differentiable then it has a derivative 𝑓 which we call the probability density function and
𝑓 𝑥 >0
𝑥2
න 𝑓 𝑥 𝑑𝑥 = 𝐹 𝑥2 − 𝐹(𝑥1 )
𝑥1
∞
න 𝑓 𝑥 𝑑𝑥 = 1
−∞
• The cumulative distribution function for a continuous random variable is given by
𝑥0
𝐹 𝑥0 = Pr 𝑥 ≤ 𝑥0 = න 𝑓 𝑥 𝑑𝑥
−∞

• The mean of a continuous random variable is given by ‫𝑥𝑑 𝑥 𝑓𝑥 𝑥׬‬

• The expected value of a function 𝑢() of a continuous random variable is given by ‫𝑥𝑑 𝑥 𝑓)𝑥(𝑢 𝑥׬‬
Probability Recap
• Take a random variable that is uniformly distributed over the interval [𝑎, 𝑏]. Every value in this interval is
equally likely to be realised.

1
• The probability density function for the uniform distribution is given by 𝑓 𝑥 = 𝑏−𝑎 for all 𝑏 ≥ 𝑥 ≥ 𝑎 and
𝑓 𝑥 = 0 otherwise.

• The cumulative density function for the uniform distribution is given by

𝐹 𝑥 = 0 𝑓𝑜𝑟 x < a
𝑥−𝑎
𝐹 𝑥 = 𝑓𝑜𝑟 𝑥 ∈ 𝑎, 𝑏
𝑏−𝑎
𝐹 𝑥 = 1 𝑓𝑜𝑟 𝑥 > 𝑏
What about Continuous Outcomes? An Example.

• Player: Manager.
• Action set: A = {𝑟, 𝑠}, where 𝑟 is invest in R&D and 𝑠 is do not invest.
• Two outcomes: the product is successful (makes a profit of 10) or becomes outdated (makes a profit of 0) so
𝑋 = {0,10}.

• You are growing cherry tomatoes and you are deciding how much to water them before you go on holiday for
a week. You can water them any amount from 0 up to 10 litres. You want to grow as many kilos of juicy cherry
tomatoes as you can, which depends on how much you water and how much the sun shines while you are
away. The most you could possibly grow from your vine is 20 kilos.

• Player: You.
•
•
Action set: A = [0,10] litres of water.
Outcome set: 𝑋 = [0,20] kilos of tomatoes.
+ =
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Question: Suppose you water some amount 𝑎 ∈ 𝐴 before you leave. You don’t know how much the sun will
shine and there is uncertainty over how many kilos you will get. How do we express this uncertainty?

Definition.
A simple lottery over an interval 𝑋 = [പ𝑥 , 𝑥]ҧ is given by a cumulative distribution function 𝐹: 𝑋 → [0,1] where
𝐹 𝑥 ′ = Pr[𝑥 ≤ 𝑥′] is the probability that the outcome is less than or equal to 𝑥 ′ .

• In our example, the outcome set is 𝑋 = [0,20] kilos of tomatoes. 𝐹 10 = Pr[𝑥 ≤ 10] is the probability that we
get less than or equal to 10 kilos of tomatoes.
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

Definition.
A simple lottery over an interval 𝑋 = [പ𝑥 , 𝑥]ҧ is given by a cumulative distribution function 𝐹: 𝑋 → [0,1] where
𝐹 𝑥 ′ = Pr[𝑥 ≤ 𝑥′] is the probability that the outcome is less than or equal to 𝑥 ′ .

• We want to express the lottery if you choose action 𝑎. That is, the probability distribution over the outcomes 𝑋 if
you choose action 𝑎. Write the cumulative distribution function given you chose action 𝑎 as 𝐹(⋅ |𝑎), where 𝐹 𝑥 𝑎
is the probability that the outcome is less than or equal to 𝑥 given you took action 𝑎.

• If 𝑎 = 0, you water 0 litres. Then 𝐹 2 0 is the probability of getting less than 2 kilos of tomatoes given you did not
water.
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• E.g. if you use 5 litres of water then your output will be uniformly distributed over the interval [0,10] kilos.

Probability density function

0 Kilos of tomatoes 10
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• E.g. if you use 5 litres of water then your output will be uniformly distributed over the interval [0,10] kilos.

Probability density function

• E.g. if you use 7 litres of water then your output will be
uniformly distributed over the interval [0,14] kilos.

0 Kilos of tomatoes 10 14
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• E.g. if you use 5 litres of water then your output will be uniformly distributed over the interval [0,10] kilos.

• The cumulative distribution function 𝐹(𝑥|𝑎 = 5) describes the probability that you produce less than 𝑥 kilos of
tomatoes given you used 5 litres of water.
𝑥
• 𝐹 𝑥 𝑎 = 5 = 10 for 0 ≤ 𝑥 ≤ 10. 𝐹 𝑥 𝑎 = 5 = 1 for 𝑥 > 10.

2
• The probability you produce less than 2 kilos of tomatoes is 𝐹 𝑥 = 2 𝑎 = 5 = 10 = 0.2.
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• E.g. if you use 7 litres of water then your output will be uniformly distributed over the interval [0,14] kilos.

• The cumulative distribution function 𝐹(𝑥|𝑎 = 7) describes the probability that you produce less than 𝑥 kilos of
tomatoes given you used 7 litres of water.
𝑥
• 𝐹 𝑥 𝑎 = 7 = 14 for 0 ≤ 𝑥 ≤ 14. 𝐹 𝑥 𝑎 = 7 = 1 for 𝑥 > 14.

7
• The probability you produce less than 7 kilos of tomatoes is 𝐹 𝑥 = 2 𝑎 = 7 = 14 = 0.5.
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• More generally, the cumulative distribution function for the uniform distribution over [0,2𝑎] is
𝑥
𝐹 𝑥 𝑎 = 2𝑎 for 0 ≤ 𝑥 ≤ 2a and 𝐹 𝑥 𝑎 = 1 for 𝑥 > 2𝑎.

𝑥
• 𝐹 𝑥 𝑎 = 2𝑎 describes the probability that you produce less than 𝑥 kilos of tomatoes given you used 𝑎 litres of water.
What about Continuous Outcomes? An Example.
• You want to grow as many kilos of juicy cherry tomatoes as you can, which depends on how much you water
and how much the sun shines.
• Player: You.
• Action set: A = [0,10] litres of water.
+ =
• Outcome set: 𝑋 = [0,20] kilos of tomatoes.

• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes is
uniformly distributed over [0,2𝑎].

• More generally, the cumulative distribution function for the uniform distribution over [0,2𝑎] is
𝑥
𝐹 𝑥 𝑎 = 2𝑎 for 0 ≤ 𝑥 ≤ 2a and 𝐹 𝑥 𝑎 = 1 for 𝑥 > 2𝑎.

𝑥
• 𝐹 𝑥 𝑎 = 2𝑎 describes the probability that you produce less than 𝑥 kilos of tomatoes given you used 𝑎 litres of water.
Probability density function
1
• The probability that you produce less than 𝑥 kilos of
𝑥 2𝑎
tomatoes is 2𝑎, given by the shaded area in the pdf.
0 𝑥 2𝑎
What about Continuous Outcomes?
• What is the player’s expected payoff when the outcome can be any one of a continuum of values distributed
on an interval 𝑋 = 𝑥പ , 𝑥ҧ ?

• Let 𝑢(𝑥) be the player’s payoff over outcomes in the interval 𝑋 = 𝑥പ , 𝑥ҧ .

Definition.
The player’s expected payoff from the lottery over X given by the cumulative distribution function
𝑥ҧ
𝑭(𝒙), with density 𝑓(𝑥), is 𝐸 𝑢 𝑥 = ‫𝑥׬‬ഫ 𝑢 𝑥 𝑓 𝑥 𝑑𝑥.

• When F(𝑥) is not differentiable expected utility is given by 𝐸 𝑢 𝑥 = ‫)𝑥(𝐹𝑑 𝑥 𝑢 𝑋∈𝑥׬‬.

What about Continuous Outcomes? An Example.

+ =
• Add detail to the example. Suppose that given you water by amount 𝑎 ∈ 𝐴, the resulting weight of tomatoes
is uniformly distributed over [0,2𝑎].
• Question: What is your expected payoff from choosing to water amount a ∈ A?
𝑥ҧ
• Expected payoff from a lottery with density 𝑓 𝑥|𝑎 is 𝐸 𝑢 𝑥 |𝑎 = ‫𝑥׬‬ഫ 𝑢 𝑥 𝑓 𝑥|𝑎 𝑑𝑥.

𝑥 1
• The cdf is 𝐹 𝑥 𝑎 = 2𝑎 for 0 ≤ 𝑥 ≤ 2a. What is the pdf 𝒇 𝒙|𝒂 ? 𝑓 𝑥|𝑎 = 2𝑎

• Plug this into expected utility

𝑥ҧ 2𝑎
1 2𝑎
𝐸 𝑢 𝑥 |𝑎 = න 𝑢 𝑥 𝑓 𝑥|𝑎 𝑑𝑥 = න 𝑢 𝑥 𝑓 𝑥|𝑎 𝑑𝑥 = න 𝑢 𝑥 𝑑𝑥
𝑥
ഫ 0 2𝑎 0
What about Continuous Outcomes?
+ =

2𝑎 1 2𝑎
• Expected payoff is 𝐸 𝑢 𝑥 |𝑎 = ‫׬‬0 𝑢 𝑥 𝑓 𝑥|𝑎 𝑑𝑥 = ‫𝑢 ׬‬ 𝑥 𝑑𝑥
2𝑎 0

• Add detail to the example: Suppose your payoff function is 𝑢 𝑥 = 12√𝑥

1 2𝑎
• Plug this into the expected utility: 𝐸 𝑢 𝑥 |𝑎 = 2𝑎 ‫׬‬0 𝑢 𝑥 𝑑𝑥
1 2𝑎 6 2𝑎
= න 12√𝑥𝑑𝑥 = න √𝑥𝑑𝑥
2𝑎 0 𝑎 0
2𝑎
6 2 3
= 𝑥2
𝑎 3 0

4 3 4 1
= 2a 2 = (2𝑎) 2a 2 = 8√2𝑎
a a
What about Continuous Outcomes?
+ =

• When 𝑢 𝑥 = 12√𝑥 we derived the expected utility from watering amount 𝑎 as 𝐸 𝑢 𝑥 |𝑎 = 8√2𝑎.

• Add detail to the example: Suppose the cost of a litre of water is 2 and you want to choose the amount of water
to maximize your expected net payoff, which is expected payoff from the tomatoes minus the cost of watering
max 8 2𝑎 − 2𝑎
a∈[0,10]

8
• Differentiating gives the first order condition: − 2 = 0.
2𝑎

• Solving: 𝑎 = 8 maximizes expected net payoff.

Example: Investing in an MBA
• You have been working for a few years after your undergraduate and now you are thinking about going to do a
prestigious Masters or MBA which will propel your career.

• The cost of doing the MBA is 10.

• If the labour market is strong then your income without an MBA is 32. Without an MBA is 12.
• If the labour market is average then your income 16. Without an MBA is 8.
• If the labour market is weak then your income is 12. Without an MBA is 4.
• The labour market is strong with probability 0.25, average with probability 0.5, and weak with probability 0.25.

• You have a choice of two actions: do the MBA or not.

• Expected payoff from an MBA 𝐸 𝑢 𝑥 𝑀𝐵𝐴 = 0.25 × 32 − 10 + 0.5 × 16 − 10 + 0.25 × 12 − 10 = 9

• Expected payoff from not: 𝐸 𝑢 𝑥 𝑀𝐵𝐴 = 0.25 × 12 + 0.5 × 8 + 0.25 × 4 =8

• You choose to do the MBA since your expected payoff is higher.

Example: Investing in an MBA
Strong labour market

0.25 32-10 Average labour market

Nature
0.5
16-10
Get MBA Weak labour market
0.25 12-10
Player
0.25 12
Don’t get
MBA 0.5 8
Nature
0.25 4
The Value of Information
• Recall, the expected payoff of getting an MBA is 9 and the expected payoff of not getting an MBA is 8.

• You get an MBA.

• Suppose an oracle comes to you right before you resign and will tell you the future labour market conditions
in exchange for a price.

• Are you willing to pay for this information and, if so, how much?
0.25 32-10=22
Nature
0.5
• The information is valuable if it will change your decision. 16-10=6
Get MBA
0.25 12-10=2
• Will it change your decision? Player
➢ If you learn the labour market is strong you will get an 0.25 12
MBA. Don’t get
➢ If you learn it is average or weak you will not. MBA 0.5 8
Nature
0.25 4
The Value of Information
• How valuable is this information?

Value of information = Expected payoff with information – expected payoff without information

• Without the information, you will get an MBA and your expected payoff is 9.

• Calculate your expected payoff with the information.

The Value of Information
Strong labour market
• Calculate your expected payoff with the information.
Get MBA 22
• Suppose the oracle tells you the labour market is strong. What will
you do? Get MBA payoff 22.
Don’t 12
• Suppose the oracle tells you the labour market is average. What will 0.25
you do? Not get MBA. Payoff 8. Get MBA 6
Nature 0.5
• Suppose the oracle tells you the labour market is weak.
What will you do? Not get MBA. Payoff 4.
Don’t 8
0.25
Get MBA 2

• Labour market is strong with prob 0.25, you will get MBA, payoff 22. Player
• Labour market is average with prob 0.5, you don’t get MBA, payoff 8.
• Labour market is weak with prob 0.25, you don’t get MBA, payoff 4. Don’t 4

𝐸 𝑢 = 0.25 × 22 + 0.5 × 8 + 0.25 × 4 = 10.5 Weak labour market

The Value of Information
• How valuable is this information?

Value of information = Expected payoff with information – expected payoff without information

• Without the information, you will get an MBA and your expected payoff is 9.

• Expected payoff with the information is 10.5.

• The value of the information is 1.5.

Notation for a lottery

• Oddly enough there is no consistent notation to denote a lottery.

• A lottery is made up of a set of outcomes 𝑋 and a probability distribution over those outcomes, which in the
discrete case we denoted by 𝑝 and in the continuous case by the cdf 𝐹.

• Thus, a sensible way we might denote a lottery is by its two components, e.g. 𝑋, 𝑝 or (𝑋, 𝐹).

• We will also use 𝓁 to denote a lottery.