Expected Value and Decisions

Expected value is the closest thing decision-making has to a universal yardstick: it lets you put a single number on an uncertain choice and set it beside another. Insurers, investors, and poker players all lean on it. But it carries a sharp limitation that ruins people who forget it — expected value describes the average outcome over many repetitions, and some decisions you get to make only once, or cannot survive losing. Knowing both what expected value tells you and where it stops being the right guide is the difference between using probability and being used by it.

What expected value is

Expected value is each possible outcome multiplied by its probability, all added up. A bet that pays $10 with probability 0.5 and loses $4 with probability 0.5 has expected value (0.5 × $10) + (0.5 × −$4) = $3. Positive expected value means that, averaged over many identical plays, you come out ahead; negative means you bleed. It is the long-run average value of one decision.

Insurance and everyday choices

Insurance has negative expected value for you — the premium exceeds your average payout, because the insurer must profit. You buy it anyway, and rationally, because it converts a small certain cost into protection against a rare ruinous one. The same arithmetic runs the other way for the insurer, who holds thousands of such policies and so genuinely gets to rely on the average. Expected value also quietly guides ordinary calls: whether to leave early for a train, whether a warranty is worth its price, whether to double-check the work.

Why positive expected value can still be a bad bet

Here is the trap. Offered a bet that stakes your entire net worth on a coin flip to triple it, the expected value is strongly positive — yet taking it is foolish, because a loss ends the game. You cannot average across a game you will not survive to replay. The average is a promise about the long run, and ruin cancels your subscription to the long run. That is the gap between one-shot and repeated decisions: over many small independent bets the average asserts itself; on a single all-in bet it is beside the point.

A raffle ticket costs $2. There is a 1-in-1,000 chance of a $1,000 prize and nothing otherwise. The expected value of a ticket is (1/1000 × $1,000) − $2 = $1 − $2 = −$1. Every ticket loses a dollar on average, so as a money-making scheme it fails. But notice what makes it harmless: the stake is trivial. Losing $2 changes nothing, so a slightly negative expected value bought for a bit of fun is a perfectly reasonable choice. Size, not just sign, decides.

Now a bet with positive expected value that you should still refuse. A stranger offers even odds on a coin, but you must wager everything you own and may play only once. On a fair coin the expected value is zero, and even a slight edge makes it positive — yet a single unlucky flip leaves you destitute with no second try. Contrast an investor placing 1% of a portfolio on each of hundreds of favourable, independent bets: the same positive edge, but now the average is allowed to do its work and ruin is off the table. Identical expected value per bet, opposite wisdom.

A classic puzzle, posed by Nicolas Bernoulli and analysed by his cousin Daniel Bernoulli around 1738, shows expected value alone failing to match sensible behaviour. A coin is flipped until the first heads; if heads first lands on flip n, you win 2 to the power n. The probabilities shrink exactly as fast as the payouts grow, so every term contributes the same amount and the expected value is infinite — the sum never stops adding. By pure expected value you should pay any finite price to play. Yet almost nobody would pay even $20, because the giant payouts are astronomically unlikely and one play cannot deliver a long-run average. Daniel Bernoulli's proposed resolution — that people value extra money less as they have more, and weigh outcomes by usefulness rather than raw dollars — is often cited as an early root of risk-adjusted decision-making. The historical framing is debated, but the lesson is durable: expected value does not, by itself, capture how a person should value risk.

• "Positive expected value means take the bet." Not if a loss is ruinous or you play only once.
• "Insurance is a bad deal because its expected value is negative." It rationally trades a small certain cost for protection against catastrophe.
• "Expected value predicts what will happen." It predicts the long-run average, not any single result.
• "Bigger expected value always wins." Two bets with the same expected value can differ entirely in their risk of ruin.

Face a series of bets with clearly stated odds and stakes, and decide which to accept — watching for the positive-expected-value bets that would still wipe you out.

Think Like a Maester: Before betting on the average, ask whether you can afford to be around for it. Expected value rewards those who get to keep playing; ruin quietly revokes that privilege.

1. How do you calculate the expected value of an uncertain choice?
2. Why does someone rationally buy insurance with negative expected value?
3. Explain why a positive-expected-value bet can still be a bad idea.
4. What is the difference between a one-shot and a repeated decision here?
5. What does the St. Petersburg paradox illustrate about expected value?

Expected value weights each outcome by its probability, giving a single number to compare uncertain choices and a firm guide across many repeated, survivable bets. But it is a statement about the long run, and its authority collapses when a loss is ruinous or a decision is made only once — the point dramatised by the St. Petersburg paradox, where an infinite expected value still tempts no sensible person. Use expected value, but first make sure you can afford to be around for the average it promises.

• Probability Distributions
• Risk vs Uncertainty