The NFT Rating

Match forecasts

How likely is a win, a draw or a defeat? The rating gives the probabilities, and every rated match since 1872 shows how reliable they are.

Skill: 14%; Better than guessing from base rates (Brier score)
Brier score: 0.536; Brier score of the forecasts (base rates only: 0.625)
Draw peak: 29%; Chance of a draw between equally rated teams
Backtest: 36,318; Matches backtested

Upcoming matches

Planned matches between two rated teams, with each team's current rating. The team playing in its own country gets 100 points extra.

No planned matches entered yet

As soon as upcoming internationals are in the database, their forecasts show up here. Meanwhile, see below what the model said about the latest matches.

The latest matches, forecast in hindsight

The 20 most recent rated matches with the forecast the ratings before kick-off would have given. ✓ marks what actually happened.

Date	HomeHome / Away	Away	Home win	Draw	Away win	Likely score	Score	Chance of the result
5 Oct 2026Neutral ground	5 Oct 2026Grenada 1,282Bonaire 1,136	Bonaire 1,136	57%	26%	17% ✓Actual result	1–0 13%	1–4	17%
4 Oct 2026Neutral ground	4 Oct 2026Aruba 1,253Anguilla 906	Anguilla 906	80% ✓Actual result	17%	4%	2–0 14%	3–2	80%
4 Oct 2026	4 Oct 2026Turks & Caicos Islands 994British Virgin Islands 958	British Virgin Islands 958	55% ✓Actual result	27%	18%	1–0 13%	3–0	55%
4 Oct 2026	4 Oct 2026US Virgin Islands 1,019Bahamas 971	Bahamas 971	57% ✓Actual result	26%	17%	1–0 13%	3–2	57%
4 Oct 2026	4 Oct 2026Greece 1,838Germany 1,951	Germany 1,951	34%	29% ✓Actual result	37%	1–1 13%	0–0	29%
4 Oct 2026	4 Oct 2026Netherlands 2,018Serbia 1,725	Serbia 1,725	84% ✓Actual result	14%	2%	2–0 14%	2–1	84%
4 Oct 2026	4 Oct 2026Portugal 2,091Norway 1,955	Norway 1,955	68% ✓Actual result	23%	9%	1–0 13%	2–1	68%
4 Oct 2026	4 Oct 2026Wales 1,701Denmark 1,874	Denmark 1,874	26%	28%	46% ✓Actual result	1–1 13%	0–1	46%
4 Oct 2026Neutral ground	4 Oct 2026Ireland 1,750Israel 1,638	Israel 1,638	52%	27% ✓Actual result	21%	1–0 12%	1–1	27%
4 Oct 2026	4 Oct 2026Kosovo 1,726Austria 1,887	Austria 1,887	27%	28% ✓Actual result	44%	1–1 13%	1–1	28%
4 Oct 2026	4 Oct 2026Azerbaijan 1,353Lithuania 1,316	Lithuania 1,316	56% ✓Actual result	27%	18%	1–0 13%	1–0	56%
4 Oct 2026	4 Oct 2026Malta 1,283Andorra 1,094	Andorra 1,094	74% ✓Actual result	20%	6%	2–0 13%	1–0	74%
4 Oct 2026	4 Oct 2026Egypt 1,835South Africa 1,701	South Africa 1,701	68%	23%	9% ✓Actual result	1–0 13%	0–1	9%
4 Oct 2026	4 Oct 2026Morocco 1,992Mali 1,721	Mali 1,721	82%	16% ✓Actual result	3%	2–0 14%	1–1	16%
4 Oct 2026	4 Oct 2026Kyrgyzstan 1,445Lebanon 1,434	Lebanon 1,434	52% ✓Actual result	27%	21%	1–0 12%	3–1	52%
4 Oct 2026Neutral ground	4 Oct 2026Senegal 1,888Comoros 1,478	Comoros 1,478	85% ✓Actual result	13%	2%	2–0 14%	1–0	85%
3 Oct 2026	3 Oct 2026Belize 1,293French Guiana 1,420	French Guiana 1,420	32%	29%	39% ✓Actual result	1–1 13%	0–2	39%
3 Oct 2026Neutral ground	3 Oct 2026Sint Maarten 1,291Saint Vincent & Grenadines 1,414	Saint Vincent & Grenadines 1,414	19%	27%	54% ✓Actual result	0–1 12%	2–4	54%
3 Oct 2026	3 Oct 2026Croatia 1,935England 2,166	England 2,166	19%	27%	55% ✓Actual result	0–1 12%	0–7	55%
3 Oct 2026	3 Oct 2026Spain 2,327Czechia 1,706	Czechia 1,706	97% ✓Actual result	2%	0%	4–0 13%	3–1	97%

Do the forecasts hold up?

Every rated match grouped by the predicted chance of a home win, compared with how often the home team actually won. On the dashed line, forecasts and reality agree.

Each point is one group of matches. Above the dashed line, teams won more often than forecast.

Predicted home win	Matches	Predicted win	Actual wins	Predicted draws	Actual draws
0–10 %	3,010	4.7%	10.1%	16.3%	17.6%
10–20 %	3,232	15.2%	23.1%	25.5%	27.0%
20–30 %	3,718	25.1%	31.0%	28.1%	28.3%
30–40 %	4,306	35.0%	38.6%	28.9%	28.7%
40–50 %	4,693	45.1%	48.2%	28.3%	28.5%
50–60 %	4,688	54.9%	55.3%	26.6%	26.4%
60–70 %	4,434	64.9%	63.6%	23.7%	22.5%
70–80 %	3,928	74.9%	73.3%	19.5%	18.3%
80–90 %	2,631	84.7%	82.1%	13.1%	13.4%
90–100 %	1,678	94.2%	91.0%	5.0%	7.3%

Model against simpler guesses

Average error over 36,318 matches (24,500 since 2000). Lower is better.

Method	Brier score	Brier since 2000	Log-loss	Log-loss since 2000	Better than base rates
NFT Rating with draw model	0.5363	0.5307	0.9114	0.9025	14.2%
NFT Rating with a constant draw share	0.5382	0.5328	0.9161	0.9076	13.9%
Base rates only	0.6251	0.6320	1.0395	1.0489	0.0%

Brier score: the squared error of the three probabilities added up (0 = perfect). Log-loss: how surprised the forecast was by the actual result. Base rates only: the same chances for every match (how often home teams win, draw and lose).

Draws by rating gap

The bigger the gap between the teams (home advantage included), the rarer the draw: observed against the fitted curve.

Rating gap	Matches	Actual draws	Predicted draws
0–50	6,147	28.3%	28.8%
50–100	5,903	28.6%	28.2%
100–150	5,264	27.5%	27.0%
150–200	4,491	24.6%	25.2%
200–250	3,708	23.4%	23.1%
250–300	2,949	20.4%	20.6%
300–350	2,287	17.8%	18.1%
350–400	1,631	15.3%	15.3%
400–450	1,161	12.7%	12.1%
450–500	849	9.9%	9.2%
500–550	602	10.8%	7.1%
550–600	421	5.5%	5.3%
600+	905	4.0%	2.7%

How it works

The rating says how many points a team can expect, a draw counting half. To split that into win, draw and defeat, the share of draws is fitted to every rated match as a bell curve of the rating gap: 29% between equally rated teams, falling off with a width of 473 points.

The draw chance is then taken equally from both sides: win = expected points − half the draw chance, defeat = the rest. So a win plus half a draw is exactly what the rating expects.

For the likely score, each team's goals follow a Poisson distribution whose average grows with the rating gap: 1.23 goals each for equally rated teams.

The backtest uses the ratings each team had before kick-off, so they only include what was known at the time. Only the two parameters of the draw curve are fitted on the whole history. A shoot-out counts as a draw.