Decision trees and expected monetary value

Decision trees and expected monetary value

Spec 7.1–2 · Edexcel A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • A decision tree has decision nodes (squares), chance nodes (circles) and end nodes showing pay-offs.
  • Only chance-node branches carry probabilities, and they sum to 1.
  • Expected monetary value =∑p×pay-off=\sum p\times\text{pay-off}. Fold back from right to left.
  • Costs are subtracted on the branch where they occur.
  • Utility captures attitude to risk; evaluate questions need a conclusion with a reason.

Decision trees

A decision tree is read left to right. Squares are choices, circles are chance events and end nodes show pay-offs.

A decision tree models a problem with choices and uncertain events, read from left to right.

At the end of each path is an end node showing the pay-off, usually the profit for that outcome.

Branch probabilities at a chance node add up to 1. At a decision node you choose, so its branches carry no probabilities.

  1. 1

    Decision node (square)

    you choose a branch; no probabilities

  2. 2

    Chance node (circle)

    branch probabilities add up to 1

  3. 3

    End node

    shows the pay-off for that outcome

Reading a decision tree from left to right.

Decision node

Drawn as:
Square
Meaning:
The decision maker chooses a branch

Chance node

Drawn as:
Circle
Meaning:
An outcome happens by chance; probabilities add to 1

End node

Drawn as:
End of a path
Meaning:
Shows the pay-off for that outcome

Which type of node has probabilities on its branches?

Constructing a tree

List events in time order, start with the first decision as a square, add chance circles and later decisions, and subtract costs on their branch.

Build the tree from the left.

Start with the first decision as a square. Each option leads either to a pay-off or to a chance node, which splits into the possible outcomes with their probabilities.

Where the outcome of one event leads to a further decision, add another square after it.

Any cost paid along a branch (such as the price of a survey) is subtracted from the pay-offs on that branch, or at the node where it is incurred.

  1. 1

    List events in time order

    every decision and every uncertain event

  2. 2

    Draw the first decision

    a square with a branch for each option

  3. 3

    Add chance nodes

    circles with probabilities on the branches

  4. 4

    Add later decisions

    another square where an outcome leads to a choice

  5. 5

    Write pay-offs and costs

    subtract costs on the branch where they occur

Building a decision tree

A survey costs £500 and is chosen at the first decision. Where is the cost accounted for?

Expected monetary value

At a chance node it is the sum of each pay-off times its probability. Fold back from right to left, choosing the best branch at each decision.

The expected monetary value (EMV) at a chance node is the sum of each pay-off multiplied by its probability:

EMV=∑p×pay-off.\text{EMV}=\sum p\times\text{pay-off}.

Work right to left, called folding back. At each chance node calculate the EMV. At each decision node choose the branch with the best value (the highest EMV for a profit) and write that value at the node.

marqueenonefine 0.7poor 0.3fine 0.7poor 0.3Hire?EMV 460EMV 540£700−£100£600£400
Pay-offs on the marquee branch already include its £300 cost.

Worked example

A café can hire a marquee for £300. In fine weather (probability 0.7) the profit is £1000 with the marquee and £600 without. In poor weather (0.3) it is £200 with and £400 without. Should it hire?

0100200300400500With marqueeWithout marqueeOptionEMV (£)
EMV of each option for the café

A gamble pays £50 profit with probability 0.4 and a £10 loss with probability 0.6. What is its EMV?

Multi-stage decisions

Evaluate later decisions first, weight them by the survey results, subtract the survey cost and compare with the best option without it.

In a multi-stage tree, later decisions are made first when folding back.

For a decision about commissioning a survey, find the best decision after each possible survey result. Use those values to find the EMV at the survey's chance node, subtract the survey cost, and compare with the best option without the survey.

The difference between the two values is how much the survey adds.

A negative EMV on a branch is a loss, so not proceeding (value 0) may be better.

  1. 1

    Best choice after each result

    take the best value at each later decision node

  2. 2

    EMV of the survey node

    weight those values by the survey-result probabilities

  3. 3

    Subtract the survey cost

    gives the value of surveying

  4. 4

    Compare

    with the best option that skips the survey

Folding back a survey tree

When folding back a tree with a survey, which decisions are evaluated first?

Utility

Utility measures personal value from 0 to 1. Choose the option with the highest expected utility; risk-averse people prefer a certain sum.

EMV treats £1 as equally valuable whatever the amount, and ignores risk.

Utility is a measure of the personal value of an outcome, usually scaled from 0 (worst) to 1 (best).

Give each pay-off a utility. The expected utility of a chance node is ∑p×utility\sum p\times\text{utility} and the option with the highest expected utility is chosen.

A risk-averse decision maker prefers a certain sum to a gamble with the same EMV. A risk-seeking decision maker does the opposite.

1020304050607080901000.20.40.60.81xy(25, 0.5)Risk-averse utilityEqual value per £
Illustrative utility curve: a risk-averse person values a certain £25 (utility 0.5) more than the straight-line value of £25 (0.25).

Risk-averse

Prefers:
a certain sum
Because:
the gamble's expected utility is lower

Risk-seeking

Prefers:
the gamble
Because:
the certain sum has lower utility

A risk-averse person is offered £100 for certain, or a gamble with an EMV of £100. Which do they prefer?

Evaluating courses of action

State the decision from each method, say whether they agree, and justify a recommendation with a reason such as risk or sensitivity.

EMV is suitable for repeated decisions, because the long-run average matters. For a one-off decision it may mislead, because one outcome occurs, not the average.

Probabilities may be estimates, and different utility values can change the decision.

When asked to evaluate, state the decision from each method, say whether they agree, and justify a final recommendation with a reason such as risk or the sensitivity of the values.

  1. 1

    EMV decision

    choose the highest expected monetary value

  2. 2

    Utility decision

    choose the highest expected utility

  3. 3

    Compare

    do the two methods agree?

  4. 4

    Recommend

    justify with risk, one-off or repeated decision, estimated probabilities

Evaluating courses of action.

EMV

Suits:
Repeated decisions
Weakness:
Misleads for a one-off decision

Utility

Suits:
Taking risk into account
Weakness:
Depends on the decision maker's values

What should an evaluate question end with?

Try an exam question

A firm can invest £2000 in a project. It succeeds with probability 0.4 and earns revenue of £8000; otherwise it earns revenue of £1000. Calculate the expected monetary value of the project and state whether, on this basis, the firm should proceed.

[4 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Find the expected monetary value (EMV) of ordering a large stock.

1 mark, marked against the mark scheme

Revision notes