Value Bets: Where Our Prediction Model Odds Beat the Bookmakers’

Every set of odds is a percentage in disguise. We work out our own percentage for the same bet, and this page lists every one where our number is higher than theirs.

Odds of 10/11 mean the bookmaker thinks it lands 52% of the time. Evens is 50%. 1/2 is 67%. That is all a price is.

We work out the same figure from the records instead — how often a player has had two shots in his last 6 matches, how often two sides have both scored in their last 20 — and put the two numbers side by side. Where ours is higher, the odds are longer than the record says they should be. That gap is the last column in the table.

The biggest gaps come first, whatever division they are in. A League One corners line sitting above most of the Premier League is worth seeing, and grouping by league would hide it.

What Value Betting Actually Means

A value bet is not a bet you expect to win. It is a bet priced longer than its real chance of winning, and most of them lose. A 40% shot at 3/1 is a good bet, and it loses six times in ten.

That distinction is the whole thing, and it is why this page shows percentages rather than confidence. If we say a selection is a 67% chance at 5/6, we are saying it should win about 2 times in 3, and the price pays as though it wins closer to 11 times in 20. Back enough bets in that shape and the arithmetic works. Back one and anything can happen.

The bookmaker is not trying to predict the result. They are setting a price that balances the money coming in with a margin built on top. Most of the time that price is a better estimate than ours — they have more information, more money riding on being right, and they adjust as bets arrive. Finding value means finding the occasions where it is not, and being honest that those occasions are rarer than any tipster page suggests.

---

How We Try to Beat the Bookies

We do not have inside information, and we do not claim to. What we have is a model that reads what has actually happened, applied consistently to markets the bookmaker prices with less attention than the match result.

Nobody prices a first goalscorer market carelessly. The match odds on Arsenal against Chelsea have been argued over by every trader in the industry. But the line on how often a specific full-back gets fouled, or how many corners two sides produce between them, gets less scrutiny — and those are markets where a count of what has happened in the last six matches is a genuinely useful input.

That is where our disagreements cluster. Look down the table, and you will see fouls drawn, shots, corners and cards far more often than match results. That is not a stylistic preference. It is where a stats model has something to offer and where it does not.

We publish our figure and the bookmaker’s side by side so you can judge the disagreement rather than take our word for it. Most tipster pages give you a selection and a reason. This gives you a number you can check against the price yourself.

---

Model Predictions Against the Odds

Our model reads counts, not form or reputation. For a player line, it takes his last 6 matches, asks how often he cleared the line, and checks the rate per 90 minutes behind it so one big return does not flatter an average. For a team market, it uses both sides’ last 20 matches.

Where the model is good. Corners, cards, shots, fouls drawn and goals at both ends. All of these are counting exercises — how often has this happened — and a 20-match sample answers that reasonably well.

Where it is not. Match results. A result depends on squad quality, and counting past matches cannot see it. Our model cannot tell that Manchester City’s bench is worth more than Coventry’s first XI, so on lopsided fixtures it sits 20 points or more from the market, and the market is right. We publish the price on results rather than argue with it, and we say so rather than quietly filling the gap.

We have measured this rather than assumed it. Tested across a European round, our match-result figure sat an average of 11 points from the market even after adjusting for league strength. That is not a prediction; it is a guess with decimals on it.

---

What the Numbers Mean

A market percentage includes the bookmaker’s cut, which on these markets runs around 5%. Add up the implied probabilities on either side of any line, and they come to more than 100 — that surplus is the margin, and it is why an edge of two or three points is not really a disagreement at all.

The table starts at 8 points for that reason. Below it, our number and the price are broadly saying the same thing.

It stops at 20 points as well, and that bound matters more than the first. When our figure sits far above the bookmaker’s, the likelier explanation is a flaw in ours rather than a mistake in theirs — and our own settled results say so.

Our own settled record, split by the size of the gap, is published at the foot of this page. The pattern that produced the ceiling is that the widest disagreements have performed worst — by a distance, and consistently enough that we stopped publishing them.

A 34-point disagreement is not an edge. It is a warning that something in our calculation is wrong — a player who has moved club, a promoted side whose figures describe a lower division, a rate built on 300 minutes.

Selections outside both bounds are still published, on the individual competition pages, with the reasoning that produced them. They are absent here because this table is about where the model and the market differ by an amount we can defend.

---

What the Model Reads

Player lines come from the player’s own last 6 matches, checked against his rate per 90 minutes. A player under 450 minutes does not appear at all, because 4 goals in 266 minutes is a hot streak rather than a rate.

We also check that he is playing. Starts this season are counted separately from the 20-match sample, which matters in September when most of that sample is last season. A forward with 4.31 shots per 90 who has started once in five is not a selection, however good the rate looks.

Team markets use both sides’ last 20 matches. Where a side has changed division, those figures get weighted by level, because a promoted team’s record against weaker opposition flatters them — Coventry’s Championship corner count, and Lincoln’s League One matches are both discounted in the rows above.

Card markets need the referee. Where no official has been appointed, we use a league average of 4.20 cards a game, the mean across the 25 referees we have measured, and any selection built that way is checked across the plausible range rather than at the average alone.

---

The Bets We Get Wrong

Our full record is at the foot of this page, including every losing bet and the running return. We do not summarise it here, because a figure typed into a paragraph goes stale the moment the next round settles and the table below never does.

Two patterns have already cost us enough to change the rules.

Attacking props on heavy underdogs. A player’s shot count or fouls-drawn record looks the same whoever he is playing for, but a side with 20% of the ball gives him far fewer chances. Every one we published on a side priced below 25% lost in August, and we now exclude them.

Figures from the wrong division. A promoted side brings its lower-league record with it, and a card or corner count built that way is inflated. We found this by pricing a League One fixture at 72% where the market said 33%, then watching all three of those selections lose. The market was closer, and even the market was high.

Both rules came from losing money rather than from theory. That is the only way we know of finding out whether a rule is worth having.

---

No value-bet results settled yet.

---

By Competition

• Premier League best bets for every round
• Championship best bets fixture by fixture
• La Liga predictions
• Bundesliga predictions
• Champions League predictions

Every one of those pages carries the same list, filtered to its own competition. The numbers cannot disagree with each other because there is only one set.