All scores →
Loading…
Loading…
Loading…

Expected value: the one concept every bettor should understand

By 4Basketball Staff · · 11 min read

What expected value means, how to calculate it for any bet, why a winning bet can have negative value and a losing bet positive value, and how basketball's own shot-selection logic is the same idea.

This article is educational and describes how betting markets work. It is not advice, a prediction or an encouragement to bet. Sports betting is legal only in some places and only for adults, and over time the odds favor the house. If gambling is a problem for you or someone you know, call 1-800-GAMBLER in the United States, or look for the help line in your own country.

If you only take one idea away from the mathematics of betting, take this one. A bet is not good because it wins. A bet is not bad because it loses. A bet is good or bad according to the price you were paid for the risk you took, and the tool for measuring that is expected value, usually shortened to EV.

Expected value is the average result of a bet if you could place it an enormous number of times under identical conditions. It is not a prediction of what will happen tonight. It is a statement about what the bet is worth. Once you can calculate it, odds stop being a mystery of plus and minus signs and become simple comparisons between a price and a probability. This article explains the concept from scratch, works through examples, shows why it explains most of what is hard about betting, and ends with the connection to basketball itself, where coaches have been thinking about expected value, under other names, for years.

What expected value means

Take a simple game. I flip a fair coin. If it lands heads, I pay you 3. If it lands tails, you pay me 1. Should you play?

Half the time you win 3, and half the time you lose 1. On average, per flip, you gain (0.5 × 3) − (0.5 × 1) = +1. You do not get exactly +1 on any single flip; you get +3 or −1. But over a thousand flips, you would expect to finish roughly 1,000 ahead. The average, +1, is the expected value of one play.

The general formula for a bet is:

EV = (probability of winning × amount won) − (probability of losing × amount lost)

The "amount won" is the profit, not including the stake you get back. The "amount lost" is your stake. A bet with EV above zero is favorable to you in the long run. A bet with EV below zero is unfavorable. A bet with EV of zero is fair.

Two points need emphasis.

  1. EV is an average over many repetitions. It tells you nothing certain about one bet. A favorable bet loses often, and an unfavorable bet wins often.
  2. EV depends on the true probability, which is the unknown. You never know the true chance that a team will cover a spread. You only have estimates. The calculation is exact once the probability is given, but the probability itself is always a judgment.

Calculating EV for common bets

A standard spread bet

A spread bet at −110 risks 110 to win 100. Suppose the true chance of winning is 50 percent. Then:

EV = (0.50 × 100) − (0.50 × 110) = 50 − 55 = −5

You lose 5 on average per 110 risked, which is −4.5 percent of the stake. This is the house margin, explained in the vig and the hold.

Now suppose the true chance is 55 percent:

EV = (0.55 × 100) − (0.45 × 110) = 55 − 49.5 = +5.5

That is +5 percent on the stake. The same price becomes favorable if the true probability is high enough. The break-even point is the probability at which EV is zero. At −110 that is 110 ÷ 210 = 52.4 percent.

A moneyline underdog

An underdog at +150 pays 150 on a 100 stake. The implied probability is 100 ÷ 250 = 40 percent.

  • If the true chance is 40 percent: EV = (0.40 × 150) − (0.60 × 100) = 60 − 60 = 0. Fair.
  • If the true chance is 45 percent: EV = (0.45 × 150) − (0.55 × 100) = 67.5 − 55 = +12.5, or 12.5 percent of the stake.
  • If the true chance is 35 percent: EV = (0.35 × 150) − (0.65 × 100) = 52.5 − 65 = −12.5.

Notice how sensitive the result is. A five-point change in the true probability moves the EV by 12.5 percentage points of the stake. This is why small errors in estimating probability matter so much, and why the estimate is the hard part.

A shortcut with decimal odds

With decimal odds D and a probability p, EV per unit staked is simply p × D − 1. At 2.50 (+150) and p = 0.45, that is 0.45 × 2.5 − 1 = +0.125. Same answer, less arithmetic. For converting between formats, see our guide to reading odds.

Why a winning bet can be a bad bet

Here is the idea that separates people who understand betting from those who do not. A bet can win and still have been a bad bet. A bet can lose and still have been a good one.

Suppose a team is priced at +150, which implies 40 percent, and its true chance is only 30 percent. You bet and the team wins. You collected 150. But the EV was (0.30 × 150) − (0.70 × 100) = 45 − 70 = −25. If you made that bet a hundred times, you would expect to lose about 2,500 on 10,000 risked. The one win was luck.

The reverse is equally true. If you bet an underdog at +150 with a true chance of 45 percent and it loses, you made a good bet that did not work out. If you would make the same bet again, the decision was right even though the outcome was wrong.

This is hard to accept because the human mind judges decisions by results. It is the reason that so many bettors credit themselves for wins and blame bad luck for losses, when the evidence of a single outcome is close to meaningless. Judging a bet by its EV, not its result, is the discipline at the center of the subject. It also protects against a darker pattern: if a bad bet wins, you may be rewarded for the mistake and repeat it.

The real difficulty: estimating the probability

The formula is easy. The input is not. Where does the true probability come from?

Nowhere, strictly speaking. It is not observable. Every estimate is a model, and models can be wrong. For an individual game, your estimate competes with the market's, and the market has a lot of information and a lot of smart participants. Here are the ways people try to form an estimate and the trouble with each.

  • Gut feeling. Overconfident, and unreliable at the precision required. The difference between 52 and 55 percent is tiny to intuition and enormous to EV.
  • Team records and statistics. Useful, but already known to the market and therefore already in the price.
  • Models. Better, if built carefully and tested out of sample. They require data and honesty about their errors.
  • The market itself. Many sophisticated bettors treat the no-vig closing price as the best available estimate of the true probability, and ask whether they can get a better price than that. This is the idea behind closing line value, a topic we cover in detail.

The uncomfortable point is that the market is hard to beat. If the market's no-vig probability is 58 percent for a team priced at −150, and you take that price, your EV is (0.58 × 100) − (0.42 × 150) = 58 − 63 = −5, about −3.3 percent of the stake. To have positive EV, you need to believe the true chance is above the break-even rate of 60 percent, and be right. Our article on why beating the closing line is hard explains the evidence that most people cannot.

EV and the margin

The house margin is, in effect, a negative EV applied to every bet. On a standard −110 market with the two sides equally likely, each bettor's EV is −4.5 percent before any skill is applied. Skill has to overcome that before it produces a profit.

For combined bets the problem is worse. In a parlay, EV is computed on the product of the probabilities, and each leg's margin compounds. A three-leg parlay of coin-flip legs at −110 has an EV of about −13 percent. We examine the numbers in parlays and teasers. The general lesson is that the more legs, the lower the EV per dollar, unless each leg has an edge.

The law of large numbers, and its limits

EV describes the long run. The law of large numbers says that as the number of repetitions grows, the average result approaches the expected value. That is true, and it is why casinos and sportsbooks are reliable businesses. But the law works slowly for individual bettors. Variance, the size of the swings around the average, is large relative to the edge in most betting situations.

As an illustration, a bettor with a genuine 54 percent win rate on −110 bets, which is an extraordinary result, has an EV of about +3 percent per bet. After 100 bets the typical swing is several times larger than the expected profit, and there is a substantial chance, around a third, of being behind. We explain this in more detail in our guide to bankroll management. So EV is the right guide to what to do, but it is a poor predictor of what will happen in any short run.

This also means you can rarely tell from a short record whether a bettor has positive EV. A hundred bets, a thousand bets, even several thousand may not separate skill from luck. That is one reason that process measures, such as how your prices compare with closing prices, are used instead.

Expected value in basketball itself

You do not need a sportsbook to meet EV. Basketball has been transformed by it.

Consider the choice between a long two-pointer and a three-pointer. If a player makes 42 percent of long twos, each attempt is worth 0.42 × 2 = 0.84 points. If the same player makes 36 percent of threes, each attempt is worth 0.36 × 3 = 1.08 points. The three is a worse bet in terms of how often it goes in, but a better bet in terms of what it is worth. Nobody remembers the percentages; the logic is the expected points per shot, and it is the reason that teams have moved their shot selection toward threes and shots at the rim, as covered in why teams stopped taking midrange shots and the three-point revolution in the numbers.

The same thinking lies behind points per possession, which values each trip down the floor by its expected points rather than by whether it ended in a make. A possession that produces a good shot and misses is still a good possession. A possession that ends in a lucky heave that goes in was still a poor one. Coaches who judge decisions rather than outcomes are doing the same thing the sound bettor does.

The link goes both ways. Basketball analysts who build win-probability models, as described in simulating a season, use the same arithmetic of probabilities and payoffs. If you can see why a coach might prefer a lower-percentage shot with a higher value, you can see why a bettor might ignore a likely winner with a poor price.

Pushes, boosts and more than two outcomes

The simple formula handles two outcomes, win and lose. Real bets sometimes have more, and the formula extends naturally: multiply each outcome's probability by its result and add them up.

Pushes. A spread or total with a whole number can push, returning the stake. A push contributes zero to the EV, so the formula becomes (probability of winning × profit) − (probability of losing × stake), with the push probability simply left out. If a bet at −110 has a 47 percent chance to win, a 6 percent chance to push and a 47 percent chance to lose, the EV is 0.47 × 100 − 0.47 × 110 = 47 − 51.7 = −4.7. The push does not rescue the bet, but it lowers the risk. It also explains why half-point differences matter in markets where whole numbers are common targets.

Boosted prices. Sportsbooks sometimes offer a "boost" that raises the payout on a selection. A boost improves the price, which raises EV, but the question to ask is the same: how does the boosted break-even probability compare with the true chance? A boost on a long shot may still leave the bet at a negative EV, and a boost that is limited to small stakes caps the benefit. Our article on limits, boosts and promotions goes into the details.

Many outcomes. A futures market lists many teams. The EV of backing one of them is p × payout − (1 − p) × stake, with p being the true chance that the team wins. Because the board's implied probabilities add to well over 100 percent, the average EV across the whole board is negative, and by a larger amount than on a single game. See futures bets.

Comparing two bets

EV also helps you choose between options, not only decide whether to bet. Suppose a bettor thinks a team has a 62 percent chance to win and sees two prices on it: −150 at one place and −135 at another.

  • At −150, the stake needed to win 100 is 150: EV = (0.62 × 100) − (0.38 × 150) = 62 − 57 = +5, which is +3.3 percent of the stake.
  • At −135, the stake is 135: EV = (0.62 × 100) − (0.38 × 135) = 62 − 51.3 = +10.7, which is +7.9 percent of the stake.

The same belief is worth more than twice as much at the better price. Nothing about the team changed, and nothing about the estimate changed. Only the price did. This is why even bettors with no edge in forecasting benefit from comparing prices, and it is the whole logic of shopping for lines. It also shows the limit of the exercise: the estimate of 62 percent was assumed, and if it was really 58 percent, both bets are losers.

What EV does not tell you

EV is a powerful tool, but it has blind spots.

  • It ignores risk. Two bets can have the same EV and very different swings. A long shot and a favorite might each be worth +5 percent, but the long shot loses most of the time. Your tolerance for variance and the size of your bankroll matter.
  • It ignores utility. A person may reasonably pay a premium for entertainment, as one does at a concert or a restaurant. Negative EV is not a moral failing, it is a price. The mistake is to pretend that the price is not there.
  • It cannot rescue bad data. A precise formula fed an inaccurate probability gives a precise wrong answer.
  • It is a long-run statement. If you will only make a handful of bets, EV and results may diverge for a long time.

A short checklist

  1. State the price. Convert it into a break-even probability.
  2. State your estimate. Be honest about how it was formed and how uncertain it is.
  3. Compare. If the estimate is above the break-even rate by a margin larger than your error, the EV may be positive. If not, it is not.
  4. Remember the margin. Every standard bet starts about 4.5 percent in the hole.
  5. Judge the decision, not the result. A single outcome says almost nothing.
  6. Respect variance. Even positive EV bets lose frequently, so size stakes accordingly.

What to take away

Expected value is a way of asking what a bet is worth rather than whether it will win. It combines the probability and the payoff into one average, and it tells you that a bet can be wrong even when it wins and right even when it loses. The price sets a break-even probability, the house margin makes that bar higher than 50 percent, and the hardest part is estimating the true chance better than the market does. For most people, understanding EV is not a route to profit. It is a way of seeing how the product works, and of knowing what you are paying for when you pay for a bet. You can practice reading prices this way on the odds page, with no money involved.

If betting stops being fun or starts to cost more than you can afford, stop and talk to someone. In the United States the National Problem Gambling Helpline is 1-800-GAMBLER, and most other countries have a similar service.

Search more on this

Keep reading