Club Elo ratings: how football Elo works and how we measure team strength

Where did Elo come from? From chess to football

The Elo system was invented by Arpad Elo, a Hungarian-American physics professor at Marquette University in Milwaukee and a keen chess player. The United States Chess Federation (USCF) adopted his method in 1960, and FIDE followed in 1970. Since then "Elo rating" has become a byword for measuring a player's strength from results.

The idea was brilliant in its simplicity:

• every player has a number of points that describes their strength,
• the gap in points tells you what result to expect,
• after a game, points pass from the loser to the winner, and how many depends on how surprising the result was.

Beating a stronger opponent earns a lot of points, beating a weaker one only a few. Losing to a favourite costs little, slipping up against a minnow costs a lot. Nobody has to be judged "by eye": the ranking corrects itself, match by match.

Elo came to football in 1997, when Bob Runyan adapted the system to international matches (the World Football Elo Ratings). Later FIFA itself took it up: the women's ranking has been based on a modified Elo from the start, in 2003, and the men's ranking switched to an Elo-based formula in August 2018.

Why rank clubs rather than national teams?

Elo ratings for national teams are easy to find. Clubs are harder, yet it is clubs that play every week, their matches we watch at the weekend and they are what the arguments are about: how strong a club really is once you look beyond its own league, or how much better the Premier League really is than the rest of Europe.

A club Elo ranking also has practical advantages over the league table:

• it accounts for the opponents' strength: 10 points from matches against the top sides is not the same as 10 points from matches against the bottom of the table,
• it compares leagues: thanks to the cups, Arsenal, Flamengo and Kashima Antlers are on one scale,
• it does not reset in the summer: a club's strength carries over from season to season, and a promoted side does not start "from zero",
• it turns into chances: the Elo gap gives the probability of a win, a draw and a defeat in a specific match.

That last point mattered most to us. We wanted a ranking that does not just put clubs in order, but says something you can check about the next match.

How does Elo work in football?

The basis is the expected result. If the home side's rating is $d$ points higher than the away side's (with home advantage added), its expected result is:

A gap of 0 points gives E = 0.5, a match between equals. A gap of 200 points gives about 0.76, and 400 points about 0.91.

After the match the actual result $S$ is 1 for a win, 0.5 for a draw and 0 for a defeat. The change in rating is:

where $K$ sets how quickly the ranking reacts and $m$ is the goal difference multiplier. Whatever one club gains, the other loses, so points do not appear out of nowhere.

In chess that is enough. Football needs a few more things, because it has draws, home advantage, different leagues and clubs that go up and down.

How we calculate our club rankings

Our model is called club-elo-v1. Here are all its parts, with nothing hidden.

One pool for all leagues

Every club, from the Premier League, the Championship, LaLiga, the Bundesliga, the Eredivisie, the Brazilian and Japanese leagues and 30 other leagues, is rated on the same scale. We calculate the ranking chronologically from 1992, match by match, on results after 90 minutes (extra time and penalties do not affect the ranking, and we skip walkovers).

Besides the domestic leagues, the ranking takes in seven international competitions, always including their qualifying rounds:

• the Champions League, the Europa League and the Conference League,
• the Copa Libertadores and the Copa Sudamericana,
• the AFC Champions League (AFC Champions League Elite),
• the Club World Cup.

The K factor: 15 in the league, 30 in the cups

A league match moves the ranking with K = 15, and a match in any of these cups with K = 30. Cup matches are rare, and they say the most about the differences between leagues, so each of them has to weigh more.

Goal difference, in moderation

A 1-0 win and a draw have a multiplier of $m = 1$. With a goal difference $g$ of two or more, the multiplier grows as $g^{0.7}$: 2-0 gives 1.62, 3-0 gives 2.16 and 5-0 gives 3.09. A 3-0 means more than a 1-0, but not three times as much.

On top of that comes a brake for favourites: the bigger the winner's lead in the ranking before the match, the more we lower the multiplier. Without it, strong clubs that regularly thrash weaker ones would inflate their ratings beyond reason. A thrashing handed out by an outsider counts in full. The whole multiplier looks like this ($d_w$ is the winner's lead in the ranking before the match):

Home advantage for every league separately

We do not assume one value for the whole world. Each league has its own home advantage, which starts at 90 points and adjusts after every match: if home sides win more often than the model predicted, it goes up, and if less often, it goes down. For matches on neutral ground we add no home advantage.

The effect is interesting. At present (as of 24 September 2026):

In Brazil, with its distances and climate, the home ground is worth almost three times as much as in Japan. The Premier League's 53 points is less than LaLiga's 75 or the Bundesliga's 56. It can differ within one country too, although not much in England, where the Championship gives the home side 49 points. Belgium's top flight, the Pro League, gives 52, and its second tier, the Challenger Pro League, only 31.

European cups move the whole league

This is the most important part if we want to compare leagues. When a Premier League club meets a Dutch club in the Europa League, it is not only the ratings of those two teams that change. The result also carries over, to a lesser degree, to the other clubs of both leagues. If English clubs keep losing to Dutch ones, the whole Premier League slowly slides down against the Eredivisie, even the clubs that did not play in Europe that season. That is how the Premier League's clubs end up rated against LaLiga's or the Bundesliga's: through every European tie, not just the ones a given club plays.

Without this, clubs that do not play in Europe would never "find out" how strong their league is. The Europa League and the Conference League are especially valuable here, because every season they bring in clubs from mid-sized and smaller leagues, such as Norway's or Denmark's. The Copa Libertadores, the AFC Champions League and the Club World Cup, in turn, link leagues from other continents with Europe. That is how Flamengo or Kashima Antlers end up on the same scale as Arsenal and Bayern Munich.

Promotion, relegation and newcomers

A club that is promoted or relegated takes its rating with it. That matters, because a promoted side is not always the weakest team: of the 66 clubs promoted to the Premier League since 2004/05, 35 stayed up and 31, or 47%, went straight back down (more in the article on how many points it takes to stay up). Nor do we pull ratings back towards the average after a season: a club that was strong in May is still strong in July.

A club that was not in our data before (or has not played for more than 400 days) starts at the average rating of the teams that left the league, plus 45 points. The whole league is then shifted so that this bonus does not add points out of thin air. A good example is Wrexham. It dropped out of League Two in May 2008 with 1147 points, and when it came back in August 2023, after 15 years outside the Football League, it started again at 1111, as a newcomer. Two promotions in a row took it to 1447 by May 2025, and as of 24 September 2026 it has 1579 in the Championship: its highest rating so far, and 11th of the league's 24 clubs.

From Elo to 1X2 chances

The classic Elo formula gives one number, but in football we need three: the chance of a home win, a draw and an away win. We use an ordered logit model for this. First we work out the strength gap, where $\text{Elo}_1$ is the home side's rating, $\text{Elo}_2$ the away side's and $H$ the home advantage in that league (0 on neutral ground):

Then we turn it into three chances: $p_1$ for a home win, $p_X$ for a draw and $p_2$ for an away win (with $a = 0.004569$):

The parameter $c$ governs draws and is fitted separately for each league, because leagues really do differ here. The model expects the most draws in Argentina's top flight ($c = 0.644$) and the fewest in the Dutch Eerste Divisie (0.534); the Premier League has 0.584. Leagues with few matches are pulled towards the global value, so that a handful of chance draws cannot distort the parameter.

An example: before El Clásico on 25 October 2026, Barcelona had 2182 points and Real Madrid 2145. LaLiga's home advantage is 75 points, so $d = 2182 - 2145 + 75 = 112$. With $c = 0.605$ for LaLiga, that gives a 48% chance of a Barcelona win, 28% of a draw and 25% of a Real Madrid win.

The same formulas run the "Who wins?" calculator on the rankings page: pick any two clubs and you get the chances of 1, X and 2. The full version, with fair odds, a comparison with bookmaker odds you type in and the rating changes after each result, is the Elo calculator (how to use it). If the 1X2 market is new to you, start with our explanation of 1X2 betting.

The ranking also shows on our football tips list: the number next to a team's name is its Elo, and the chances from these formulas stand beside the 1X2 forecast. How to read a whole match row, we show step by step in how to read football tips.

How we calibrated the model

Each of the parameters above (K, home advantage, the goal difference exponent, the brake for favourites, the newcomer bonus, how strongly cup results carry over to the league) could be set differently. We did not pick them by feel.

We did it in three steps:

1. Data. 323,000 matches from 1992–2026, from 37 domestic leagues and 7 international cups.
2. Training. We tuned the league parameters so that the ranking predicts league results from 2012–2023 as well as possible. The cup parameters (K in the cups and how strongly cup results carry over to the league) we tuned on cup matches from 2008–2023. After adding the Europa League, the Conference League and the cups outside Europe we checked them again: K = 30 and the same carry-over strength still came out best. We also made sure that no league does systematically better or worse in the cups than the ranking predicts.
3. Test. We checked the model on matches from 2023–2026 that it had not seen while the parameters were being chosen.

The measure was log-loss, a penalty for confidence in the wrong direction. A model that gives 80% to an outcome that did not happen gets a big penalty. A model that says "I don't know" for every match gets an average one. The lower, the better.

In the leagues the gain over textbook Elo is small, because Elo on its own is already very good. In the cups the difference is clear, because there what matters is comparing leagues, and the classic approach cannot do that.

Do the Elo ratings hold up?

On the rankings page we publish a test result that updates itself. We take all matches from the last 12 months and for each one work out the pre-match chances, with parameters fitted only on the five years before. So the model does not know the results it is judged on.

As of 24 September 2026 (13,361 league and cup matches since 24 September 2025):

• The ranking's favourite won 48.8% of matches. It sounds modest, but in football one match in four ends in a draw, and favourites often have only a 40–50% chance. A model that picked 70% of favourites correctly would have to know the future.
• Log-loss 1.024 against 1.073 for outcome frequencies alone.
• Calibration stays on the diagonal. This is the most important test: when the model says "30%", the outcome should happen roughly 3 times in 10.

Most of the differences are fractions of a percentage point. To be fair, in the 90–100% band the model predicted 91.5% on average and 72.7% happened. That is only 11 cases, too few to draw conclusions from, but we will keep an eye on it.

Club Elo in practice: real matches

Formulas are one thing, but Elo is easiest to see in real matches. All the numbers come from our ranking.

Augsburg win in Munich: one of the biggest upsets of the season

On 24 January 2026 Augsburg (1723) won 2-1 away at Bayern Munich (2208). The model gave it a 4.5% chance, the lowest of any winner among the recent matches of ranked clubs.

Yet Augsburg gained "only" 14 points. It was a one-goal win, so the multiplier was 1, and K in the league is 15. The most a match like this can earn is 15 points. Elo deliberately does not turn the ranking upside down after one upset: if Augsburg really is stronger, it will show in the next matches.

Barcelona and Liverpool, May 2019

On 1 May 2019 Barcelona beat Liverpool 3-0 in the Champions League semi-final and gained 20 points, while Liverpool lost 25. Three days later Barcelona lost to Celta Vigo (−25), and three days after that it lost 4-0 at Anfield and dropped 34 points. In the second leg Liverpool won back everything it had lost in the first, and a little more.

Manchester United in Paris, 2019: a 3-1 worth 36 points

In February 2019 Manchester United met PSG in the Champions League last 16. In the first leg, at Old Trafford, it lost 2-0 and dropped 27 points. Before the second leg, in Paris, United had 1999 points and PSG 2100. With the hosts' home advantage, the model gave United only a 20% chance of winning.

United won 3-1. Let's do the sums:

• the gap from United's side (87 was the home advantage in the Champions League at the time): $d = 1999 - 2100 - 87 = -188$,
• United's expected result: $E = 1 / (1 + 10^{188/400}) \approx 0.25$,
• the multiplier for a two-goal margin: $m = 2^{0.7} \approx 1.62$ (no brake, because United was the outsider),
• the change in a cup match, so with $K = 30$:

United jumped to 2035. On goals the tie finished level, 3-3, and United went through on away goals, but in the ranking the two legs together left it 9 points up: the win in Paris brought more than the home defeat had cost, because the model expected much less of it.

Liverpool v Tottenham 5-1: why did a big win earn only 4.5 points?

On 27 April 2025 Liverpool beat Tottenham 5-1 at Anfield and sealed the Premier League title. Even so, it gained only 4.5 points. Why?

• Liverpool was a clear favourite: 2229 against 1964, plus the Premier League's home advantage. The model gave it a 71% chance of winning, and its expected result was 0.863.
• It was a league match, so K = 15, not 30.
• The brake for favourites kicked in: the multiplier for a four-goal margin (2.64) dropped to about 2.20.

The result:

Elo rewards what it did not expect, and Liverpool thrashing Tottenham at Anfield was largely expected.

Bayern Munich v Arsenal 0-2 in 2013

In February 2013 Arsenal met Bayern Munich in the Champions League last 16. In the first leg it lost 3-1 at home and dropped 15 points, and in the second leg, in Munich, it won 2-0 and got 42 back: the model had given it only a 12% chance of winning. That was not enough to go through, because Bayern advanced on away goals, but Elo does not look at the aggregate score. A win over an opponent like that is always worth something.

Bodø/Glimt and the Champions League, 2025/26

The highest rating of Bodø/Glimt in our data is 1988 points, from 11 March 2026, reached during its Champions League run: a 2-2 in Dortmund, then wins over Manchester City (3-1, +40) and at Atlético Madrid, two over Inter and a 3-0 over Sporting CP. Between the end of November and mid-March it rose by almost 150 points without playing a single league match. This example shows why the ranking includes the European cups: for clubs from mid-sized leagues they are often the only chance to measure themselves against the rest of Europe. Without those matches, the ranking would know much less about them.

Tottenham: a slide you cannot see in one match

Elo also shows trends well. After a 2-0 win in Frankfurt in the Champions League on 28 January 2026 and a 2-2 with Manchester City four days later, Tottenham had 2046 points. Then came league defeats: away at Manchester United (−9), then against Newcastle, Arsenal, Fulham, Crystal Palace (1-3, −16), Nottingham Forest (0-3, −20) and Sunderland, and in the Champions League a 5-2 defeat at Atlético Madrid (−20). Single wins, such as 3-2 over Atlético in the second leg (+17), only slowed the fall. None of the defeats was a disaster, but by 18 April Tottenham had lost more than 100 points and dropped to 1944. It is a good example of Elo reacting to form gradually rather than in jumps.

A draw can cost points too

Sunderland drew 2-2 at home with Arsenal in November 2025 and gained 7 points, because it was a huge outsider (a 4.6% chance of winning). Liverpool, on the other hand, drew 1-1 at home with Sunderland a month later as clear favourites (84%) and lost 7 points. Arsenal likewise: a 0-0 at home with Sporting CP in the Champions League at a 72% chance of winning cost it 11 points. The same score on the board can mean a good day or a bad one, depending on what was expected.

Who else publishes club Elo ratings?

We are not the first. Here are the main club strength rankings and how ours differs from them.

Of the others, only FiveThirtyEight's SPI published both match chances and a public test of their accuracy, and it closed in June 2023. ClubElo gives chances, but only for upcoming matches and without a published test.

How our ranking differs from the others

ClubElo is a great project, and we modelled a lot on it, because it is the best public implementation of Elo for clubs. Still, we went for a few things you will not find together anywhere else:

1. We show that it works. The rankings page has a calibration test from the last 12 months, calculated on matches the model did not know. Most rankings ask you to take them on trust.
2. Draws have their own model for each league. A draw in Argentina's top flight and one in the Dutch second tier are two different stories, and we account for that.
3. Every match is transparent. Each club's page shows its recent matches with the pre-match chances and the exact change in rating. You can check every number.
4. Strength, not opinion. We mix no market values, expert ratings or bookmaker odds into the classic ranking. It is a pure record of what happened on the pitch. Closing odds are used by a separate, experimental version, which we describe below.
5. Second tiers on the same scale. The pool includes second divisions such as the Championship and Germany's 2. Bundesliga, rated on the same scale as the Premier League. You can check whether a second-tier leader would be a mid-table side one level up or a candidate for relegation.
6. A calculator for any pair. Arsenal against Flamengo, or Bayern against Kashima Antlers? Pick two clubs and you get the chances, with the right home advantage: the league's in a match within one league, the cup one in a match between clubs from different leagues.

Experimental version: the ranking with odds

Alongside the classic ranking we calculate an experimental version of it. After every match it moves points not only for the result but also for the closing odds, that is, for how the market priced the match just before kick-off. On 35,455 league matches from 2023–2026 it predicted results more accurately than the classic version: a log-loss of 1.0116 against 1.0218. A match's odds affect the ranking only after the match, and we do not show the odds themselves on the site. How it works and what it changes in the order of clubs, we describe in the article on Elo ratings with closing odds. You can switch between the two versions on the club rankings page.

Limitations worth remembering

Elo knows only results. It knows nothing about injuries, a change of manager, summer signings or a manager fielding a reserve side in the league match before a European second leg. That is why:

• at the start of a season the ranking can lag behind clubs that strengthened or weakened a lot over the summer,
• one upset changes little, and the ranking reacts to a real change in a club's strength only after a few matches,
• the American leagues (MLS and the USL Championship) do not play clubs from other leagues in our data, so they compare clubs only within their own league. Their Elo is not on a common scale with the rest of the world, which is why the rankings page shows them separately,
• leagues outside Europe are linked to the rest of the world by few matches, mainly in the Copa Libertadores, the AFC Champions League and the Club World Cup. Comparing them with European leagues is therefore less certain than comparing two European leagues,
• chances from the ranking describe team strength, not a ready-made tip. Elo is a good starting point for your own view of a match, but it will not replace thinking or sensible bankroll management.

FAQ

Summary

Elo is a way of measuring strength brought over from chess, and it works surprisingly well in football, as long as it is adapted to the game. Our club rankings add to classic Elo a moderated goal difference, home advantage and draws for each league separately, the carry-over of results from seven cups, from the Conference League to the Copa Libertadores, to whole leagues, and fair treatment of promoted clubs. We tuned the parameters on more than 320,000 matches and publish the evidence that the forecasts hold up. Check where your club stands and compare it with any rival in the calculator.