How accurate are our football predictions?
Across 169 finished matches from 28 June 2026 to 22 November 2026 (Serie A, World Cup, Premier League, Bundesliga, Ligue 1, La Liga), our model's most likely result was right 51% of the time, against 48% for simply backing the home side. Its both-teams-to-score calls were right 49% of the time, its over/under 2.5 goals calls 54%, and its single most likely scoreline 4%.
Recalculated automatically from the match results in our database. Updated 22 November 2026.
Hit rate by market
| Market | Our model | Naive baseline | Matches |
|---|---|---|---|
| Match result (1X2) | 51% | 48% (always home win) | 169 |
| Both teams to score | 49% | 54% (always the more common answer) | 169 |
| Over/under 2.5 goals | 54% | 53% (always the more common answer) | 169 |
| Exact score | 4% | — | 169 |
Brier score for match-result probabilities: 0.627 (lower is better; 0.667 is what you get by saying one-third each).
Are the percentages honest? Calibration
A good model's 60% calls should come true about 60% of the time. Here is how often the model's result pick actually happened, grouped by how confident it was.
| Model confidence | Average stated | Actually happened | Matches |
|---|---|---|---|
| Under 45% | 40% | 47% | 74 |
| 45–55% | 50% | 47% | 43 |
| 55–65% | 59% | 61% | 28 |
| 65% and over | 72% | 58% | 24 |
By competition
| Competition | Matches | Result | BTTS | Over/under 2.5 |
|---|---|---|---|---|
| Serie A | 34 | 50% | 62% | 59% |
| World Cup | 32 | 66% | 63% | 63% |
| Premier League | 28 | 50% | 39% | 54% |
| Bundesliga | 26 | 38% | 50% | 50% |
| Ligue 1 | 25 | 36% | 40% | 40% |
| La Liga | 24 | 63% | 29% | 58% |
Small samples swing a lot: treat any competition with fewer than 50 matches as indicative only.
Latest 15 calls
| Match | Model said | Final score | Result call |
|---|---|---|---|
| Augsburg v MainzBundesliga · 22 November 2026 | Augsburg win (49%) · 1-1 | 3-1 | Right |
| Paris FC v Angers SCOLigue 1 · 22 November 2026 | Paris FC win (44%) · 1-1 | 4-0 | Right |
| Hull City v Brighton HovePremier League · 22 November 2026 | Hull City win (51%) · 1-1 | 2-0 | Right |
| Real Madrid v CeltaLa Liga · 21 November 2026 | Real Madrid win (70%) · 2-0 | 2-1 | Right |
| Napoli v TorinoSerie A · 21 November 2026 | Napoli win (65%) · 2-1 | 1-0 | Right |
| Strasbourg v BrestLigue 1 · 21 November 2026 | Strasbourg win (54%) · 1-1 | 0-1 | Wrong |
| Lille v Olympique LyonLigue 1 · 21 November 2026 | Lille win (44%) · 0-0 | 0-1 | Wrong |
| RC Lens v ToulouseLigue 1 · 21 November 2026 | RC Lens win (58%) · 2-1 | 1-1 | Wrong |
| Monaco v AuxerreLigue 1 · 21 November 2026 | Monaco win (75%) · 3-1 | 1-1 | Wrong |
| Schalke v StuttgartBundesliga · 21 November 2026 | Stuttgart win (40%) · 1-1 | 0-0 | Wrong |
| Newcastle v ArsenalPremier League · 21 November 2026 | Arsenal win (39%) · 0-0 | 2-2 | Wrong |
| Sevilla FC v Real BetisLa Liga · 21 November 2026 | Draw (41%) · 0-0 | 1-1 | Right |
| Parma v RomaSerie A · 21 November 2026 | Parma win (40%) · 1-1 | 1-5 | Wrong |
| Nice v PSGLigue 1 · 21 November 2026 | Nice win (52%) · 2-1 | 2-3 | Wrong |
| Rayo Vallecano v ValenciaLa Liga · 21 November 2026 | Valencia win (61%) · 1-2 | 0-1 | Right |
How we measure it
The model is a Dixon–Coles-adjusted Poisson model: it turns each team's recent scoring and conceding (time-weighted, split home and away, adjusted for opponent strength) into expected goals, then into probabilities for every scoreline. To score it fairly, we re-run it for each finished match using only the games that had finished before that kick-off, then compare its calls with the real result. That is the same model behind our predictions for today, BTTS, over/under and correct score pages.
We publish the misses alongside the hits. A model that is right half the time on match results is normal for football, a low-scoring sport where draws and upsets are common.
FAQ
How is the accuracy measured?
For every finished match we re-run the same Poisson model that powers our live prediction pages, using only the games that had finished before that kick-off. No result is ever seen in advance. Each call is then marked right or wrong against the real final score. Matches where either team had played fewer than three games are left out, so early-season guesses do not distort the numbers.
Can a prediction model guarantee winning bets?
No. These are probabilities, not certainties, and a well-calibrated 60% call still fails four times in ten. Bookmaker odds include a margin, so even an accurate model does not mean profit. If you bet, do it for fun, only with money you can afford to lose, and only if you are 18 or over.
Not betting advice. Past accuracy does not guarantee future results. 18+ only if you bet; free, confidential help at BeGambleAware.org.