Value Betting and Expected Value, Without the Hype

Value betting means placing a bet only when the odds pay more than the true chance of the outcome deserves, so that the bet has positive expected value. In plain terms, the price is too generous. The idea is sound arithmetic. The trouble is that the true chance is never shown to you. Every value bet rests on your own estimate beating the bookmaker’s, and the bookmaker builds a margin into each price that you have to overcome first. This guide shows the expected-value sum with worked numbers, then sets out what the evidence says: how long it takes for luck to stop hiding skill, what happened to researchers who did find mispriced odds, and what Great Britain’s regulator found when it counted restricted accounts.

There are no tips here, and no method. The worked numbers are our own, and so is a 100,000-bettor simulation. The real-world figures come from a research paper posted on the arXiv preprint server, the Gambling Commission’s 2025 data on account restrictions, and Pinnacle’s published betting rules.

Expected Value in One Line

Expected value (EV) is the average result of a bet if you could repeat it many times. For a bet at decimal odds, where the price includes your stake, the sum is:

EV per unit staked = probability of winning × decimal odds − 1

The answer is the average profit or loss for each unit you stake. Here it is at a price of 2.50, with a few different guesses at the true chance:

True chance of winning	Sum	EV per unit staked
44%	0.44 × 2.50 − 1	+10%
42%	0.42 × 2.50 − 1	+5%
40%	0.40 × 2.50 − 1	0%
38%	0.38 × 2.50 − 1	−5%

The price is the same in every row. The only thing that changes is the probability, and a shift of two percentage points takes the bet from a 5% gain on average to break-even. That is the whole subject in one table. EV is easy to work out once you know the probability. The probability is the hard part.

The break-even point is 1 ÷ the decimal odds. At 2.50 it is 40%. A bet is “value” only if the true chance is above that line, and only by the amount it is above it.

Why Most Bets Start Below Zero

A bookmaker’s prices are not a neutral estimate. They carry a margin, and you can see it by turning each price into a probability and adding them up. Take a two-way market priced at 1.91 on each side:

• Each side: 1 ÷ 1.91 = 52.36%
• Both sides together: 104.71%

A fair book would add up to exactly 100%, because one side must win. The extra 4.71% is the overround. If the contest really is 50:50, the fair price would be 2.00, and the 1.91 you are offered has an EV of 0.50 × 1.91 − 1 = −4.5%. That is the starting point for an ordinary bet on an even contest: about 4.5p lost on every pound staked, on average, before any skill comes into it.

To get above zero at 1.91 you need the true chance to be above 52.36%. So you have to be right not only more often than a coin toss, but more than two percentage points more often than the market’s own fair estimate, on every bet you count as value. The implied probability calculator does this conversion for any market and strips out the margin so you can see the fair figure the bookmaker started from.

The same arithmetic explains why combining bets makes things worse rather than better. Each leg carries its own margin, and our guide to why accumulator payouts look better than they are shows how those margins multiply when legs are joined.

The Real Problem: Your Probability Is an Estimate

Value betting guides tend to start with “find a bet where your probability is higher than the bookmaker’s”. That hides the difficult step. You do not know the true probability. You have an estimate, and an estimate can be wrong in either direction.

Here is what that does to a single bet. Say you think a side wins 53% of the time and you find it at 2.00. On your numbers the EV is 0.53 × 2.00 − 1 = +6%. Now look at what the bet is really worth if your estimate is off:

If the true chance is	Your believed EV	Real EV
53% (you are exactly right)	+6%	+6%
51.5%	+6%	+3%
50%	+6%	0%
48%	+6%	−4%

A three-point error wipes out the whole edge. A five-point error turns it into a loss as big as an ordinary bet at a normal margin. And you cannot see which row you are in from inside one bet, or from a few dozen.

There is also a selection trap. If you compare your estimates with the prices and back only the cases where yours look better, you will tend to pick the ones where your estimate is too high. Those are the cases where you are most likely to be wrong in your own favour. The bets that look like value are partly the ones where your model has made its biggest mistakes.

Betting formulas have the same weakness. Staking rules such as the Kelly criterion tell you how much to bet given an edge, but Wikipedia’s summary of the research notes that the criterion is only fully valid when the probabilities are fully known, and that overestimating your chance of winning pushes the stake past the best level and raises the risk of ruin. A formula cannot fix a bad input. It can make a bad input more expensive.

How Long Luck Can Hide Skill: A 100,000-Bettor Simulation

Suppose, for the sake of argument, that you really do have an edge. How soon would your results show it? We simulated 100,000 bettors of two kinds, each staking one unit per bet on even contests. The code is our own, and the random seed is fixed so the run can be repeated.

• Bettor A has a genuine, hypothetical 3% edge: a 51.5% chance at odds of 2.00.
• Bettor B has no edge and bets at 1.91 on true 50:50 contests, an EV of −4.5%.

Bets placed	A: share ahead	B: share ahead	A: median result	B: median result
100	57.9%	30.8%	+2.0 units	−4.5 units
500	73.5%	15.0%	+16.0 units	−22.5 units
1,000	82.1%	6.9%	+30.0 units	−45.0 units
2,500	93.1%	1.0%	+74.0 units	−112.5 units
5,000	98.2%	under 0.1%	+150.0 units	−225.0 units

Three things stand out.

A hundred bets prove almost nothing. After 100 bets, nearly a third of the bettors with no edge are ahead, and more than a third of the bettors with a real edge are behind. A season of weekend bets cannot tell the two apart. Anyone pointing to a profitable few months as proof of skill is reading noise.

Even a real edge spends a long time underwater. By bet 1,000, 16.2% of the edge bettors were still in loss. Their worst run from a peak to a low point during those 1,000 bets had a median of 26 units, and 15.2% of them saw a fall of 40 units or more. A bettor with an edge has to keep betting the same way through losing runs like that without knowing whether the edge was ever there.

The numbers favour the margin over time. At even-money prices, one bet’s result swings by about one unit either way, and luck grows with the square root of the number of bets. A 3% edge grows in a straight line. The two only become equal after about 1,100 bets. For Bettor B, the margin works the same way in the opposite direction, which is why only 1% of them are ahead after 2,500 bets.

This simulation is the best case for value betting. Bettor A’s edge is real, fixed and known to us. Nobody limits A’s account, every price is still there when A bets, and A never changes method after a bad month. Real betting offers none of that.

What Happened When Researchers Found Mispriced Odds

The best-known attempt to test value betting with real money is a 2017 paper by Lisandro Kaunitz, Shenjun Zhong and Javier Kreiner. They did not try to out-forecast bookmakers. They averaged the closing odds of up to 32 bookmakers across 479,440 football matches from 2005 to 2015, treated that average as an estimate of the true probability, and looked for individual prices that sat well above it.

Their results, in their own figures:

• Historical closing odds: a 3.5% return across 56,435 simulated bets.
• Three months of paper trading: a 5.5% return across 407 bets.
• Five months of real money, with a planned $50 stake per bet: a profit of $957.50 across 265 bets, which the paper reports as an 8.5% return.
• Paper and real combined: $2,086 across 672 bets, a 6.2% return.

That looks like proof that value betting works. Read on and the paper’s own analysis is more cautious. The authors compared their 672 live and paper bets with a random betting strategy over the same period and found a 1 in 11 chance that random bets would have done as well (a p-value of 0.089, short of the usual 0.05 bar for statistical significance). They also found that around 30% of the prices their system flagged had already changed by the time anyone could act on them, and that they missed many opportunities because no one was watching the screen around the clock.

Then the accounts were cut back. A few months into real-money betting, the authors wrote, bookmakers “started to severely limit our accounts”. Some bets were held for “manual inspection” before being accepted. Others were capped well below their $50 stake. The figures they published show limits as low as $1.25 and $10.45. They stopped betting because they could not continue the strategy. The authors called this discrimination against successful clients and gave the paper the word “rigged” in its title. That is their view, not a legal finding, but the limits themselves were shown in screenshots.

So the one well-documented case reads like this: a team of data scientists, using a strategy that needed no better forecasts than the bookmakers’ own, made a small return whose live results were not statistically clear, and were limited out of the market within months.

Bookmakers Can Limit Winners, and Often Do

That experience is not unusual, and in Great Britain the regulator has now put numbers on it. In early 2025 the Gambling Commission asked some of the largest online betting operators for data on commercial restrictions placed on active accounts during 2024. Its chief executive, Andrew Rhodes, published the headline findings in July 2025:

Finding (2024, sample of large GB operators)	Figure
Active customer accounts in the sample	14,923,840
Accounts restricted in some form	643,779 (4.31%)
Restricted accounts given a stake limit (“stake factoring”)	62.17%
Stake-limited accounts cut to 1%–9% of the normal maximum	36.22%
Stake-limited accounts cut to 1% or less	22.41%
All active customers in lifetime profit	25.42%
Restricted customers in lifetime profit	46.78%

The Commission notes that one person may hold accounts with several operators, so these are accounts, not people. Even so, two points are plain. Restricted customers were far more likely to be ahead than customers in general. And about three in four active customers overall, 72.54%, were in lifetime loss.

The regulator also set out its position. Operators “are entitled to act in their commercial interests and manage liabilities”, it is not the Commission’s role to tell them how, and “being a successful bettor is not a protected characteristic in discrimination law”. Bookmakers’ own terms say much the same. Pinnacle’s general rules state that all bets “will be accepted or rejected purely at Pinnacle’s discretion”, and that if there is an obvious error in the odds, bets on that market may be voided. A price that looks generous because it was a mistake may not stand.

Why Few Bettors Keep an Edge

Put the pieces together and the reasons are not mysterious:

• The margin comes first. On a typical two-way price you start about 4.5% behind, and you have to clear that before any skill shows.
• Your probability is a guess. Small errors in your estimate are enough to remove an edge entirely, and picking bets where you disagree with the market favours your mistakes.
• Luck hides skill for a long time. Even a real 3% edge only matches ordinary luck after about 1,100 even-money bets.
• Good prices move. In the Kaunitz study, about 30% of flagged prices had gone by the time they were checked.
• Winning accounts get limited. The GB data and the researchers’ own experience both show stakes being cut once a customer looks profitable.
• People change course. A losing run is exactly when most bettors start chasing, raising stakes or switching method, which throws away whatever edge there might have been.

None of this means every price is fair. Bookmakers do make mistakes, and the study above found some. It means a reader should treat claims of steady value-betting profit with deep suspicion, especially from anyone selling picks, software or courses.

Claims to Treat With Suspicion

• “Value betting is a long-term winning system.” Only if the edge is real, measured over thousands of bets, and the accounts stay open. The evidence above says all three are rare together.
• “Our tips are value bets.” A tipster’s record over a few hundred bets cannot tell skill from luck, as the simulation shows. A seller earns from subscriptions whether or not the tips win.
• “Long odds are where the value is.” A long price is value only if the outcome is more likely than the price says. Long shots also swing harder, so short records mean even less.
• “You can’t lose if you only bet value.” You can lose for months with a genuine edge. Without one, you lose at the rate of the margin.

If you are new to how odds and margins work in general, the plain-English guide to sports betting for beginners covers prices, bet types and budgets before any of this matters.

What Expected Value Is Actually Good For

EV is still worth understanding, just not as a promise of profit. It is the clearest way to see what a bet costs you. A price with a 2.6% margin costs less over time than one with 5%, so comparing margins tells you which way of betting loses money more slowly. It shows why an accumulator costs more than the same picks bet as singles. And it is a useful check on your own thinking: if you cannot say how you got your probability, you are not value betting, you are guessing with extra steps.

Set a budget before you bet, treat the money as spent, and never raise stakes to win back a loss. If betting has stopped being something you can take or leave, the free, confidential support on our responsible gambling page is a good first step. You must be 18 or over, and you should check the law where you live before opening any betting account.