Home › Method and accuracy

Method and accuracy

How the forecasts are made and how they have done.

The forecast takes each club's results so far and its history, then plays the rest of the season 10,000 times. Every figure on the site is a summary of those simulated seasons: a chance is how often something happened in them, expected points the average final total, the range where 80% of them ended.

Before a season a club's expectation comes from its record the season before, pulled toward the middle because tables always even out, or, for a club new to the division, from what similar clubs did in the past. Match results then move the estimate of each club's strength round by round; home advantage and the draw rate are taken from the division's own history.

The model for each season is chosen in advance. Candidate models are scored on the seasons before it and the best is frozen for the season; nothing is picked with hindsight. Past seasons shown on the full forecast pages are reconstructed under that rule and say so.

What it does not know. Transfers, injuries, managers, cup runs and money. Before a ball is kicked the forecast has no skill for the title in the top tier or for promotion in the third; the summer decides too much. It earns its keep from the first rounds on.

The four divisions

DivisionSeasonPSimsUpdated
Pro League2026-2731/18210,0002026-09-10
First Division2026-2721/21010,0002026-09-06
Second Division2026-2713/21010,0002026-09-06
Third Division2026-275/51210,0002026-09-09

Every forecast on the site is rebuilt the morning after each matchday.

How the forecasts have done

Backtest: each past season forecast with only the seasons before it, the results revealed round by round, checked after rounds 2, 5, 10, 15 and 20.

Show the backtest numbers

Pro League

Seasons tested
11
Club-seasons
152
80% band held the real total
86%
Mean error before the season
8.4 points
Mean error after round 20
2.3 points

Brier score (lower is better), against equal chances

Title0.0410.067
AFC place0.0820.206
Goes down0.0510.124

Stated chance against what happened, AFC place

Pooled over every in-season checkpoint. The backtest scores the top four, which is this season's AFC allocation of 4. In seasons when fewer places were awarded the live forecast counts fewer, so treat this as the top-four record.

Read the 5-15% band with care: the model has said 9% and it has happened 1% of the time over 182 forecasts. In that range it has been too confident.

Forecast bandForecastsSaidHappened
0–5%6481%0%
5–15%1829%1%
15–30%14022%30%
30–50%13240%47%
50–70%14060%59%
70–100%27890%90%

First Division

Seasons tested
11
Club-seasons
146
80% band held the real total
79%
Mean error before the season
9.5 points
Mean error after round 20
3.0 points

Brier score (lower is better), against equal chances

Title0.0490.070
Goes up0.0790.127
Goes down0.0640.127

Stated chance against what happened, Goes up

Pooled over every in-season checkpoint. The backtest scores the top two, the same zone the page forecasts.

Read the 70-100% band with care: the model has said 89% and it has happened 73% of the time over 74 forecasts. In that range it has been too confident.

Forecast bandForecastsSaidHappened
0–5%8821%1%
5–15%1649%5%
15–30%14023%37%
30–50%8839%39%
50–70%11259%57%
70–100%7489%73%

Second Division

Seasons tested
3
Club-seasons
41
80% band held the real total
74%
Mean error before the season
11.6 points
Mean error after round 20
3.7 points

Brier score (lower is better), against equal chances

Title0.0490.068
Top two0.0630.125
Goes down0.0710.125

Stated chance against what happened, Top two

Pooled over every in-season checkpoint. The backtest scores the top two, the same zone the page forecasts.

Only 3 seasons and 41 club-seasons could be tested here, too few to say much. Treat this division's figures as a rough guide.

Forecast bandForecastsSaidHappened
0–5%2321%0%
5–15%509%8%
15–30%5222%27%
30–50%4237%33%
50–70%856%50%
70–100%2684%92%

Third Division

Seasons tested
2
Club-seasons
33
80% band held the real total
73%
Mean error before the season
17.9 points
Mean error after round 20
4.6 points

Brier score (lower is better), against equal chances

Title0.0380.057
Top two0.0500.107
Goes down0.0620.107

Stated chance against what happened, Top two

Pooled over every in-season checkpoint. The backtest scores the top two, the same zone the page forecasts.

Only 2 seasons and 33 club-seasons could be tested here, too few to say much. Treat this division's figures as a rough guide.

Read the 5-15% band with care: the model has said 8% and it has happened 0% of the time over 60 forecasts. In that range it has been too confident.

Forecast bandForecastsSaidHappened
0–5%1881%1%
5–15%608%0%
15–30%3823%16%
30–50%2436%58%
50–70%861%75%
70–100%1293%100%
Follow

Independent forecasts with no connection to the UAE Football Association or the UAE Pro League. Club crests are the clubs' own, shown for identification. Built 2026-09-11.

Results and fixtures come from the UAE Football Association feed and the Pro League. The archive is from the UAE FA, RSSSF and Wikipedia; where sources disagree the correction is recorded beside what it changed.