Skip to content

Economic Modelling · Rational choice theory and utility

Expected Utility Theory and the von Neumann-Morgenstern Axioms

Updated 11 October 2026 · Fact-checked

Expected utility theory says that if your preferences obey the von Neumann-Morgenstern axioms, you choose between risky options as if you maximise E[U(W)], the probability-weighted average of the utility of each outcome. To solve a question, compute each outcome's utility, weight by probability, sum, and pick the highest.

Understand Expected Utility Theory

Many choices involve risk. A gamble gives different wealth outcomes with different probabilities. A simple rule would be to pick the gamble with the highest expected money value. That rule fails in practice. Most people will not pay a very large sum for a gamble with a very large expected value, and many people buy insurance even though insurance has a negative expected money value for the buyer.

Expected utility theory fixes this by valuing outcomes through a utility function U(W). Utility measures the satisfaction you get from wealth W. You do not average the money. You average the utility. The expected utility of a gamble is E[U(W)] = Σ p(i) × U(W(i)). You prefer the gamble with the higher expected utility.

The von Neumann-Morgenstern axioms say when this works. They describe preferences over gambles (lotteries). The usual four are: completeness (you can compare any two gambles and prefer one or be indifferent), transitivity (if A is preferred to B and B to C, then A is preferred to C), continuity (if A is preferred to B and B to C, some mix of A and C with probabilities p and 1 − p is exactly as good as B), and independence (if you prefer A to B, you still prefer a mix of A with any C to the same mix of B with C). If your preferences obey these, a utility function exists whose expected value represents your choices.

The shape of U shows your attitude to risk. If U is increasing and concave (U'(W) > 0, U''(W) < 0), you are risk averse. By Jensen's inequality, E[U(W)] < U(E[W]) for a non-degenerate gamble, so you prefer the certain expected wealth to the gamble. A linear U means risk neutral. A convex U means risk seeking.

The St Petersburg paradox motivated the idea. A fair coin is tossed until the first head. If this happens on toss n, you win 2^n rupees. The expected money value is infinite, since each term contributes 1. Yet people pay little to play. Bernoulli resolved this with a concave utility such as U = ln W, where the expected utility is finite. Note that U is only defined up to a positive linear transformation a + bU with b > 0. It gives the ranking of gambles, not a measure of absolute satisfaction.

Key rules to remember

Expected utility
E[U(W)] = Σ p(i) × U(W(i))
Probabilities must sum to 1. Apply U to each outcome before weighting.
Expected wealth
E[W] = Σ p(i) × W(i)
Used for comparison with U(E[W]). Do not confuse it with E[U(W)].
Certainty equivalent
U(CE) = E[U(W)], so CE = U⁻¹(E[U(W)])
The certain wealth that gives the same utility as the gamble.
Risk premium
Risk premium = E[W] − CE
Positive for a risk-averse person. The most you would pay to avoid the risk.
Risk-aversion test
U'(W) > 0 and U''(W) < 0 means risk averse; U'' = 0 risk neutral; U'' > 0 risk seeking
Jensen's inequality: for concave U, E[U(W)] ≤ U(E[W]).
Invariance of utility
V(W) = a + b U(W), b > 0, gives the same choices
Utility is ordinal over gambles up to a positive linear transformation.
von Neumann-Morgenstern axioms
Completeness, transitivity, continuity, independence
Together they imply preferences can be represented by expected utility.

How to solve Expected Utility Theory questions

Use this method for any question that asks you to compare gambles, find a certainty equivalent or explain the theory.

  1. 1Write down the utility function U(W) and the wealth outcomes. Check whether the outcomes are final wealth or just gains. Add initial wealth if the question gives it.
  2. 2List the probability of each outcome and check that they sum to 1.
  3. 3Compute U for every outcome separately.
  4. 4Multiply each utility by its probability and add to get E[U(W)].
  5. 5Compare E[U(W)] across the options. The highest is preferred. For a sure option, compute U of the sure amount.
  6. 6If asked for the certainty equivalent, set U(CE) = E[U(W)] and invert U. Then risk premium = E[W] − CE.
  7. 7If asked about attitude to risk, check the sign of U'' or compare E[U(W)] with U(E[W]). State your conclusion in words.
  8. 8For theory questions, name the axiom used and state what it means in plain words.

Quickest way: Compare utilities, not money

When to use it: Use in multiple-choice questions and short calculations where U is simple, such as ln W, √W or a quadratic.

  1. Compute E[U] for each option first. Skip the expected money value unless asked.
  2. For U = √W, work with square roots of the outcomes. Pick perfect squares to keep arithmetic clean.
  3. For the certainty equivalent, invert at the end only: if U = √W, CE = (E[U])². If U = ln W, CE = exp(E[U]).
  4. Sanity check: for a risk-averse person, CE must be below E[W]. If not, you made an error.

Common mistakes in Expected Utility Theory

  • Averaging the wealth first and then applying U, getting U(E[W]) instead of E[U(W)].

    Expected value feels natural, and the order of the two operations is easy to swap.

    Fix: Always apply U to each outcome first, then weight by probability. Write the Σ line explicitly.

  • Reporting the certainty equivalent as E[U(W)].

    Students stop one step early. E[U] is in utility units, not rupees.

    Fix: Invert U to convert back to wealth. Check the answer lies between the lowest and highest outcomes.

  • Using only the gain as W, ignoring initial wealth.

    The question lists prizes and the starting wealth is easy to overlook.

    Fix: Define W as final wealth. Add initial wealth to each outcome before applying U, especially for ln W.

  • Saying a gamble with higher expected money value is always preferred.

    This confuses expected value maximisation with expected utility maximisation.

    Fix: State that a risk-averse person may prefer a lower but safer expected wealth. Justify using concave U.

  • Treating utility values as cardinal measures, such as saying utility 10 is twice as good as utility 5.

    Numbers on utility look like measurements.

    Fix: Say utility is only unique up to a positive linear transformation a + bU. Only the ranking of expected utilities matters.

  • Mixing up the axioms, for example describing independence as transitivity.

    The names are abstract and similar in sound.

    Fix: Learn one line each: complete (compare anything), transitive (no cycles), continuous (some mix equals the middle option), independent (mixing with a common third gamble does not reverse preference).

Worked examples

Example 1

An individual has U(W) = √W. She chooses between (A) a certain wealth of ₹40,000 and (B) a gamble giving ₹10,000 with probability 0.5 and ₹90,000 with probability 0.5. Which does she prefer? Find the certainty equivalent of B and the risk premium.

Show the solution
  1. U(A) = √40,000 = 200.
  2. For B: U(10,000) = 100 and U(90,000) = 300.
  3. E[U(B)] = 0.5 × 100 + 0.5 × 300 = 200.
  4. So E[U(A)] = E[U(B)] = 200. She is indifferent.
  5. Certainty equivalent of B: √CE = 200, so CE = 40,000.
  6. E[W] for B = 0.5 × 10,000 + 0.5 × 90,000 = ₹50,000.
  7. Risk premium = 50,000 − 40,000 = ₹10,000.

Answer: She is indifferent between A and B. The certainty equivalent of B is ₹40,000 and the risk premium is ₹10,000.

Example 2

An investor has U(W) = ln W and current wealth ₹1,00,000. She is offered a bet: with probability 0.5 she gains ₹80,000 and with probability 0.5 she loses ₹20,000. Does she accept? Also state which concept the answer illustrates.

Show the solution
  1. Final wealth if she wins: 1,00,000 + 80,000 = ₹1,80,000. If she loses: 1,00,000 − 20,000 = ₹80,000.
  2. If she refuses, utility = ln 1,00,000 = 11.5129.
  3. ln 1,80,000 = ln 1.8 + ln 1,00,000 = 0.5878 + 11.5129 = 12.1007.
  4. ln 80,000 = ln 0.8 + ln 1,00,000 = −0.2231 + 11.5129 = 11.2898.
  5. E[U] = 0.5 × 12.1007 + 0.5 × 11.2898 = 11.6953.
  6. 11.6953 > 11.5129, so she accepts.
  7. Check by the geometric mean: CE = √(1,80,000 × 80,000) = √(1.44 × 10¹⁰) = ₹1,20,000, which is above 1,00,000.
  8. Note E[W] = ₹1,30,000. CE is below E[W], consistent with risk aversion.

Answer: She accepts, because expected utility 11.6953 exceeds 11.5129. Her certainty equivalent is ₹1,20,000, below the expected wealth of ₹1,30,000, so she is risk averse but the bet is still favourable enough.

Exam tips

  • Define W as final wealth and write U on each outcome before you weight. Examiners award marks for clear working.
  • For written questions on the axioms, state each axiom in one sentence and say what goes wrong if it fails. Do not just list the names.
  • Link concavity to risk aversion using Jensen's inequality, and link the shape of U to the risk premium.
  • When asked about the St Petersburg paradox, give the infinite expected value, then show a concave utility such as ln W gives a finite expected utility.
  • In Paper B or any computer-based question, show the formula, define the inputs, and state the result in rupees with a one-line interpretation.

Practice questions from Rational choice theory and utility

Expected Utility Theory in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Expected Utility Theory: frequently asked questions

What are the von Neumann-Morgenstern axioms?

They are rules on preferences over gambles: completeness, transitivity, continuity and independence. If your preferences satisfy them, your choices can be represented by maximising expected utility. Some texts split or word them slightly differently, so learn the meaning of each.

How do I calculate the expected utility of a gamble?

Find the utility of each possible final wealth, multiply each by its probability, and add. For example, with U = √W and outcomes 100 and 400 with equal probability, E[U] = 0.5 × 10 + 0.5 × 20 = 15.

Why does the St Petersburg paradox matter?

It shows that maximising expected money value does not describe real choices, since the game has infinite expected value but people pay little. Bernoulli's answer was diminishing marginal utility, which led to expected utility theory.

Is the certainty equivalent the same as expected utility?

No. Expected utility is measured in utility units. The certainty equivalent is the sure amount of wealth whose utility equals that expected utility. You find it by inverting U.