Why a betting model alone is not enough to beat the bookmaker
A good football model is not enough to beat the bookmaker. We tested this on our own team strength ratings: they predict results clearly better than league averages and just as well as a well-tuned club Elo rating. But when we compared them with the prices of Pinnacle, the bookmaker with the lowest margin and the sharpest lines, none of 89 simple betting rules passed the test, and bets placed where the model disagreed most with the market lost money. This guide explains where the gap comes from and what it means in practice.
Key takeaways
- Bets where the model gave one side at least 10 percentage points more than the market lost between 8% and 15% of the amount staked.
- Disagreeing with the market did not predict the price move, so it usually means the market knows something the model does not.
- Even an Elo rating that learns from closing odds closes 57% of the gap to the market, but it does not overtake it.
- You can trust a betting system's results only if its rules were fixed before the period shown and the results were compared with the closing line.
The model predicts well. The market predicts better
We check the attack and defence ratings and the club Elo ratings on matches the model has not seen before. On 35,455 league matches from the 2023/24–2025/26 seasons, our goals model predicts results clearly better than a simple league average and as well as the Elo ranking. That is a solid result: the model understands who is strong, where they play and how many goals they usually score.
The trouble is that the bookmaker knows this too. Market odds take in many models, news on line-ups, injuries and motivation and, above all, the money of bettors who pay for what they know. A good statistical model knows only part of this. To make money from it, being good is not enough. You have to be better than the market, and by enough to cover the margin.
How we tested it
The easiest way to fool yourself is to test an idea on the same data you came up with it on. So the test had two parts:
- Choosing the rules: seasons 2019–2023. On this data we looked for simple rules, such as "back the team the model rates X percentage points higher than the market" or "bet over when the model expects more goals than the line". That gave 89 candidates.
- The test: seasons 2023–2026. The rules that looked promising were checked on later matches that we had not seen while choosing them. In every match the model knew only what had happened before it.
The benchmark was Pinnacle's prices, because that is the hardest market to beat. If a model cannot beat it, an edge you see at another bookmaker usually comes from that bookmaker's higher margin or weaker line, not from the model knowing something the market does not.
Results: zero out of 89
- None of the 89 rules passed the test. Some looked good in 2019–2023, but the edge disappeared in the following seasons.
- The biggest disagreement with the market meant the biggest losses. Bets where our model rated one side's chances at least 10 percentage points higher than the market lost between 8% and 15% of the amount staked.
- Disagreement did not predict the price move. If the model knew something before the market, the price should have moved its way before kick-off. It did not.
The second point matters most. A big gap between the model and the market looks like an opportunity. In practice it usually means the market knows something the model does not: that the reserve goalkeeper is playing, that the team is saving its strength for a cup tie, or that the club has changed its manager.
Why the market is so hard to beat
The margin
Bookmaker odds are not a fair price. The probabilities implied by 1X2 odds always add up to more than 100%, and the excess is the margin. Even if your read of the chances is exactly as good as the market's, the margin means you lose in the long run. How to work out a price without the margin is explained in our guide on how to calculate fair odds.
The closing line
The price just before kick-off, the closing line, is the best available estimate of the chances. By then the market has taken in the line-ups, injuries, the weather and the money of the best-informed bettors. Professionals measure themselves by whether they regularly get a better price than the close. If a model often "sees value" and the price then moves the other way, that is a sign there was no value. How far the price still moves on match day, we show in when do odds stop moving.
The market is models too
Big bookmakers and professional bettors have models at least as good as ours, and usually better, because they have more data. Elo ratings, xG and attack and defence strength are publicly known ideas. Everything you can read from them is already in the price. Even our Elo rating with closing odds, which learns from closing prices, closes the gap to the market by 57% but does not overtake it. For the same reason AI betting tips do not beat the odds either: in tests from 2026, the leading language models predicted the World Cup slightly worse than the market.
Where the promises of "value" from simple models come from
Multiple testing
If you check 89 rules, a few of them will almost certainly show a profit in the past by pure chance. With many attempts, the best result always looks better than the rule's true value. That is why a test on later data, not used in the choice, matters so much. Anyone who shows only the best result out of many attempts is showing luck, not an edge. Football streaks fool people the same way: the 37 leagues we track have about 141 active streaks of 6 or more matches a day, so some streak always looks impressive.
A small sample
Betting results swing a lot. A few dozen or even a few hundred tips are too few to tell an edge from a lucky run. A system that "made 20% last month" may be entirely down to chance. Match stats have the same problem: a team conceding 1.3 goals a game lets in about 8 goals per 15-minute spell over 38 matches, so 11 late goals instead of 8 is ordinary chance.
Comparing with the wrong price
A model that "beats" odds with a high margin often does not beat the margin-free price or the closing line. Your price being better than a worse price does not mean it is better than the market.
What this means in practice
- Use statistics to understand matches, not to hunt for bankers. Team strength ratings, xG and Elo describe well who is strong and why. That knowledge helps you watch and analyse football; it does not guarantee a win.
- Do not trust systems that promise a profit. If someone sells a betting system with past results, ask whether the rules were fixed before the period they show and whether the results were compared with the closing line. Usually the answer is no.
- Do not chase a system. If a method stops working, adding more and more conditions until it "shows a profit" again on past data is exactly the multiple testing mistake.
- Set your rules before your first bet. A fixed, low stake, loss limits and a budget you can accept losing. We cover this in 10 rules to protect your bankroll.
- Treat betting as entertainment. Bet only with licensed bookmakers and only amounts you can afford to lose.
FAQ
Can you beat the bookmaker in the long run?
A few people manage it, usually thanks to better information, faster reactions or working many markets at once, not thanks to a simple model. Most bettors lose in the long run, mainly because of the margin.
If your model does not make a profit, why do you publish it?
Because it describes team strength well and helps you understand matches. We are also honest about what it cannot do.
What is the closing line?
It is the price just before kick-off. It is seen as the best available estimate of the chances, because it includes all the information the market has gathered up to the last moment.
Do value bets actually exist?
Yes, but they are rare, small and hard to confirm. A gap between a simple model and the market is, in most cases, the model's mistake, not an opportunity.