Elo ratings with closing odds: the Elo that learns from the market
Elo with odds is an experimental version of our football club ranking. After every match it moves points not only for the result but also for how the market priced the match just before kick-off, that is, for the closing odds. You will find both versions on the club rankings page, under the "Classic" and "With odds" switch.
Key takeaways
- On 35,455 matches from 2023–2026 it predicted results more accurately than classic Elo, in each of the three seasons.
- It closed 57% of the gap to the closing odds, but the market still predicts better, so it gives no edge over the bookmaker.
- The order of clubs barely changes (rank correlation 0.967), and Chelsea, Roma and Leverkusen are among the clubs that move up.
- It is an experimental version that will replace the classic ranking if its edge holds on current matches.
What results-only Elo cannot see
The classic club Elo ranking learns only from results. That is its strength, because every number can be checked, and its weakness, because the result of one match carries very little information. A 1-0 win from a stoppage-time penalty and a 1-0 win in a match where the opponent did not manage a shot on target look the same in the table. Elo also knows nothing about injuries, a change of manager or summer signings until it sees their effects in later results.
The market knows all this earlier. The closing odds include the line-ups announced an hour before the match, injuries, form and the money of thousands of bettors. In our article on when odds stop moving we showed that the moves in the odds on match day are information, not noise, and in the one on betting odds that closing odds judge the chances more accurately than odds from a few days earlier. If so, a ranking can learn from them.
There is a second, less obvious problem too. Results-only Elo exaggerates the differences between clubs. To make its 1X2 chances match the results, we have to flatten them: the coefficient in the mapping to chances is 0.00467, while the Elo formula would give 0.00576. The ranking with odds does not need this. Its coefficient for leagues with odds is exactly 0.00576, so a points gap means what the Elo formula says it means.
How the ranking with odds works
Everything we described for the classic ranking stays: one pool of 37 leagues and 7 cups, home advantage tracked separately for every league, cup results carried over to whole leagues, promotion and relegation. Only what happens to the points after a match changes.
In the classic ranking the home side's points change by , where is the result (1 for a win, 0.5 for a draw, 0 for a defeat), is the expected result from the Elo gap, is the goal difference multiplier and . The ranking with odds adds a second term:
is the expected result from the closing odds with the bookmaker margin removed: the chance of a home win plus half the chance of a draw. The result weighs half as much as in the classic ranking and the odds a lot, because closing odds are far less random than a single result. The away side loses as many points as the home side gains.
An example. The ranking gives the home side an expected result of 0.60. The closing odds without the margin give 52% for a home win and 26% for a draw, so . The match ends in a draw.
| For the result | For the closing odds | Total | |
|---|---|---|---|
| Classic ranking | 15 · (0.5 − 0.60) = −1.5 | – | −1.5 |
| Ranking with odds | 7.5 · (0.5 − 0.60) = −0.75 | 200 · (0.65 − 0.60) = +10 | +9.25 |
A draw is a weaker result than expected, so the classic ranking takes points off the home side. Before the match, though, the market rated it higher than the ranking did, so the ranking with odds moves it up. When the ranking and the market agree, the odds term is close to zero and the result decides the change.
Three rules keep it honest:
- A match's odds affect the ranking only after the match. We work out pre-match chances from a ranking that knows the odds only of matches already played.
- A match without closing odds counts as in the classic ranking. This mainly applies to older seasons of leagues for which we have no odds history.
- We do not show the odds. The odds are only material for the ranking to learn from; the site shows only Elo points and chances.
We take closing odds from two sources. Up to April 2025 they are Pinnacle's odds from 28 leagues, going back to the 2012/13 season. From April 2025 we use closing odds worked out from many bookmakers' prices, for all 37 leagues.
How we tested it
We recalculated the whole ranking from 1992, more than 323,000 matches, in 76 variants of the weights for the result and for the odds. We chose the weights on matches from 2016–2023 and report the results on matches from 2023–2026 that we did not use for the choice. The measure was log-loss, the same as for the classic ranking: a penalty for confidence in the wrong direction, where lower is better. The variant without the odds term reproduces the classic ranking to a fraction of a point, so we compare exactly the same matches and the same model, differing in one thing only.
The result does not hinge on one lucky setting. Odds weights from 150 to 300 and result weights from 5 to 12 give almost the same log-loss. At each of the six result weights we checked, adding the odds term improved the forecasts, whatever its weight.
Results: how much more accurate
Log-loss of the 1X2 chances on league matches from 2023–2026:
| Matches | Number | Classic | With odds | Difference |
|---|---|---|---|---|
| All | 35,455 | 1.0218 | 1.0116 | −0.0101 |
| With closing odds | 30,292 | 1.0195 | 1.0075 | −0.0120 |
| Without closing odds | 5,163 | 1.0348 | 1.0356 | +0.0008 |
| 2023/24 season | 11,667 | 1.0166 | 1.0069 | −0.0097 |
| 2024/25 season | 11,923 | 1.0194 | 1.0110 | −0.0084 |
| 2025/26 season | 11,865 | 1.0292 | 1.0169 | −0.0123 |
For comparison: outcome frequencies alone, with no ranking at all, have a log-loss of 1.0743, and all the tuning of the classic ranking over textbook Elo improved it by 0.0019. The ranking with odds improves the result by 0.0101, with a 95% confidence interval from −0.0115 to −0.0088. It was better in each of the three seasons. On matches without closing odds the difference is within the bounds of chance: there both rankings compute almost the same thing.
The most important comparison, though, is the one with the market. On matches with closing odds, the odds themselves have a log-loss of 0.9986. Classic Elo trails them by 0.0209, the ranking with odds by 0.0089. The gap shrank by 57%, but it did not disappear: the market still predicts better.
What changes in the ranking
The order of clubs is very similar in both versions: the rank correlation is 0.967. Half of the 734 clubs move by no more than 31 places. The ranking with odds is not more jumpy either: the average change after a league match is 8.1 points, against 7.0 in the classic one. The points themselves cannot be compared between the versions, though, because the ranking with odds has a smaller spread.
The top of the ranking with odds (as of 24 September 2026):
| Rank | Club | Elo with odds | Rank in classic |
|---|---|---|---|
| 1 | Bayern Munich | 2159 | 2 |
| 2 | Barcelona | 2147 | 5 |
| 3 | Manchester City | 2128 | 3 |
| 4 | Arsenal | 2104 | 1 |
| 5 | Real Madrid | 2090 | 8 |
| 6 | PSG | 2048 | 4 |
| 7 | Liverpool | 2029 | 6 |
| 8 | Manchester United | 2023 | 9 |
| 9 | Inter | 2006 | 11 |
| 10 | Chelsea | 1999 | 18 |
The biggest differences concern clubs whose results drifted away from how the market priced them. Chelsea, Roma and Leverkusen are 8–11 places higher in the ranking with odds than in the classic one, because the market rated them higher than their recent results alone would suggest.
What the ranking with odds does not do
It gives no edge over the bookmaker. It learns from the market, so by definition it cannot beat it, and the market still predicts more accurately. We described the same lesson in the guide on why a betting model is not enough: everything you can read from public data is already in the price. Score predictions show it too: in 2,660 Premier League matches, a Skrill Final Score pick worked out from the odds landed 12.9% of the time, and one from Elo 12.1%. A chatbot does no better: in tests from 2026, AI betting tips from the leading language models predicted matches worse than the odds.
The ranking with odds is a better map of team strength. It helps where there are no odds yet, or never will be: for matches scheduled weeks ahead, for comparing clubs from different leagues in the "Who wins?" calculator, and for the question of who is really strong when the table says otherwise.
Why it is an experimental version
The test on 2023–2026 is still a test on history. That is why on the ranking with odds page we publish a live test: the pre-match chances from the last 12 months and the classic ranking's log-loss on exactly the same matches. If the edge holds on new matches, the ranking with odds will replace the classic one. Until then the classic ranking stays the main one, and it is the source of the chances shown with matches, the biggest upsets in the leagues and the Elo calculator.
You reach the ranking with odds through the "With odds" switch on the club rankings page. It works on every club's page too: there you will see the club's Elo with odds history and its recent matches with this version's chances.
FAQ
Is the ranking with odds just bookmaker odds turned into points?
No. The ranking knows the closing odds only of matches already played and uses them to improve its estimate of team strength. About an upcoming match it knows only what follows from the teams' earlier matches. That is why it can give chances for any pair of clubs, including a pairing nobody prices, and why its forecasts are still weaker than the closing odds.
Why is Arsenal only fourth in the ranking with odds?
In the classic ranking Arsenal has 2285 points, 36 more than Bayern, because it has recently won more often than the ranking predicted. Before its matches the market rated it highly, but not higher than Bayern, Barcelona and Manchester City. The ranking with odds combines both pieces of information, so it puts Arsenal just behind them.
Can I bet based on it?
The ranking with odds describes team strength better, but it does not point to bets with an edge: the market it learns from still predicts more accurately. Treat it as a starting point for your own view of a match. Bet only with licensed bookmakers and only amounts you can afford to lose.
Where do you get the closing odds from?
The history up to April 2025 is Pinnacle's closing odds from 28 leagues since the 2012/13 season. From April 2025 it is closing odds worked out from many bookmakers' prices for all 37 leagues. We always remove the margin from the odds, so the ranking learns from clean chances.
What happens to a match that had no odds?
We count it exactly as in the classic ranking: for the result, with and the goal difference multiplier. On such matches both rankings predict practically the same.
When will the ranking with odds replace the classic one?
When the test on current matches confirms the edge seen in the history. We publish it on the ranking with odds page and update it every day, so anyone can check how it is going.