Value bet: what the edge actually measures

What is a value bet?

A value bet is one where your estimated probability is higher than the one the odds need to break even. You're comparing two numbers: a price, and an estimate that doesn't come from that price.

People often get the term wrong. "Value bet" doesn't mean "bet to place" or "bet that will win". It's a gap worked out from assumptions, and it's only as good as those assumptions.

The starting point: two sources

A value calculation needs the odds and a probability estimated independently of them. Without that independence, the calculation measures nothing.

The classic mistake is to use the bookmaker's implied probability as your estimate. The result is negative by construction, by the size of the margin. You're comparing a price with itself, so all you find is the overround. Even margin-free odds aren't an independent estimate. They're extracted from market prices and keep all their errors.

A usable probability has to come from somewhere else: a statistical model, a consensus of several sources, solid outside data, or a personal analysis that's labelled as such.

The formulas

Two formulas are enough. They say the same thing from two angles.

theoretical edge = estimated probability × decimal odds − 1

expected value = theoretical edge × stake

With odds of 2.20 and an estimated probability of 50%: 0.50 × 2.20 − 1 = +0.10, so +10%. On a £10 stake, the theoretical expected value is +£1.

Here's another way to read the same sum. The fair price for 50% is 2.00 (see fair odds). The 2.20 on offer is higher, hence the positive edge. Both approaches are equivalent. The second often reads better, because it compares two prices rather than a price and a percentage.

In this calculator, two of the three fields are easy: the odds and the stake are right in front of you. The third, your estimated probability, I can't fill in for you, and neither can the platform. It drives the whole result.

Where does the estimated probability come from?

Four sources come up most often, each with its own blind spots.

A statistical model is the most rigorous on paper, as long as it's calibrated and tested out of sample. A model that's never been tried on data it hasn't seen proves nothing.

A personal estimate is legitimate but rarely calibrated. Naming the favourite is easy, since the market does it too. Putting a number on the gap to within a few points is another matter, and intuition slips quickly there.

A consensus of several sources holds up well against one-off errors. It stays correlated with the market, though. If every source copies the same prices, it's no longer independent of the price it's meant to judge.

That leaves outside data: line-ups, form, playing conditions, injuries. It helps if it's newer than what the market has already absorbed, and that's increasingly rare. How fast information gets absorbed is exactly what odds movements show.

Why can an estimated probability be wrong?

Five mechanisms, and none of them shows in the final figure.

• Overfitting catches models calibrated on too few matches. They reproduce the noise in their sample and look brilliant on their own history.
• A missing input, such as a late withdrawal or something riding on the standings, shifts the estimate without making it look absurd.
• A source that says 60% where the outcome lands 45% of the time produces positive edges in a row. All of them wrong.
• Keeping only the matches where the model "got it right" creates a track record that doesn't exist. That's selection bias.
• The market absorbs information continuously. Yesterday's estimate can lag behind a price that has moved since.

A +10% edge from a solid estimate and a +10% edge from an overfitted model are written down exactly the same way.

Are margin-free odds a value bet?

No. The question comes up a lot.

Margin-free odds describe what the market estimates once the overround has been spread back out. They're almost always higher than the displayed odds, because the margin has been taken out. That difference only measures the margin. It isn't a value gap.

For theoretical value, you need an estimate that disagrees with the market. A simple restatement of the market won't do. It's the same distinction as between margin-free odds and a model's fair price, covered on fair odds.

Value bet or winning bet?

The two ideas are independent, and mixing them up is expensive.

A value bet can lose. With a +10% edge on an outcome estimated at 50%, it loses one time in two. The calculation expects that.

A winning bet can have been taken at a bad price. An outcome at 2.00 taken at 1.70 is still a bad price, even if it comes in. The result says nothing about the quality of the price. To get around that problem, some people look at closing line value, which compares two prices without ever looking at the match result.

Neither idea makes a bet risk-free. They're theoretical expected values over a large number of repeats, while every single bet stays uncertain and losses are possible. Responsible gambling matters here for a simple reason: a positive expectation on paper protects you from no real loss.

OddScore's position

OddScore helps you read the market price.

The platform converts odds into probabilities, strips out each operator's margin, estimates a reference price and follows how it moves. It doesn't produce a sporting probability independent of the market. So it has nothing to point at a value bet with, and no page on the site will. OddScore gives no staking advice: the AI reading sums up what the market says, it isn't a bet to follow.

Put it into practice

Apply this guide to real odds.

Dig into the market

Odds movements are only part of the story. Here are the next topics to read.

Frequently asked questions

Sources & methodology