Player-level Elo ratings for professional football

Every player has their own Elo rating, and so does every coach. When two teams meet, we look at the actual lineup — who starts, who's on the bench, who's in the dugout — and weight each one by how much their position influences the result. Attackers carry the most weight. The coach, midfielders, defenders, goalkeepers and bench players each get smaller shares. That gives a team strength for that specific match. Two team strengths plus a per-league home advantage produce a probability for home win, draw, and away win.

After the final whistle every player's Elo moves up or down based on the gap between prediction and result — scaled by minutes played, margin of victory, and how big an upset it was. Big upsets shift ratings more than expected wins. Players age out gradually too, with attackers peaking around 24 and goalkeepers around 28, so the model never gets stuck rating someone for who they were five years ago.

Somewhere along the way I stumbled into a useful by-product: scoring every match a player plays against the average performance for their position, and adding those scores up. Elo Above Replacement (EAR). Only later did I realise this is essentially football's version of baseball's Wins Above Replacement (WAR) — same idea, different sport.

EAR doesn't just sharpen the predictions. Week after week it shows you which player is genuinely lifting his team above the level of his teammates — the kind of signal you can't see in goals or assists alone. The highest career EARs in football belong to exactly the names you'd hope they would — Messi, Ronaldo, Kimmich. The strongest sanity check this model has ever passed.

football-data.co.uk is the best free archive of bookmaker closing odds anywhere on the internet. I started checking my predictions against Pinnacle Closing — the sharpest line in football — to see how close a model built only from player ratings could get.

The answer: its match predictions approach that line. Approaching it is not beating it, so the betting tip service stopped on 4 October 2026; the bets and their record stay on the betting page. What it does show is that the player ratings underneath can be relied on — and that is what the rest of this site is built on: league simulations.

Everything you see here is free and stays free — no paywalls, no email gates, no “upgrade to see more”. If you'd like to keep the lights on (about €30 per month of API + server bills), donations live on the support page.

And if you're a developer or data scientist who'd rather pull the numbers straight into your own code — ratings, full history, EAR, market-value estimates, predictions, transfer-fit and more, all as JSON — there's a REST API. It has a free tier, so you can start without paying a cent; the paid plans exist only to cover heavier usage. The website itself stays free either way.