English Premier League closing-line track record
How Pinnacle's closing prices fared against 90 finished English Premier League matches, 2026-08-21 to 2026-11-08. Updated 2026-11-15.
Pinnacle's closing favourite won 50.6% of the time, against the 51.3% its closing price implied on average. Matches were drawn 24.4% of the time against an implied 24.8%. The closing line's mean Brier score (0.597) and the opening line's (0.595) differ by 0.0014, within two standard errors (0.0074) of zero: on this sample the closing line is not measurably more accurate than the first price we recorded.
The closing price against the result
| Measure | What happened | What the closing price implied | 95% range of what happened |
|---|---|---|---|
| Closing favourite won | 50.6% (45 of 89) | 51.3% | 40.4%–60.7% |
| Match drawn | 24.4% (22 of 90) | 24.8% | 16.7%–34.2% |
| Price moved toward the result | 46.7% (42 of 90) | — | — |
Calibration by closing probability
| Closing implied probability | Outcomes | Mean implied | Happened | 95% range |
|---|---|---|---|---|
| 0–20% | 40 | 15.2% | 15.0% (6) | 7.1%–29.1% |
| 20–40% | 154 | 28.0% | 27.3% (42) | 20.9%–34.8% |
| 40–60% | 59 | 49.0% | 49.2% (29) | 36.8%–61.6% |
| 60–80% | 15 | 67.9% | 73.3% (11) | 48.0%–89.1% |
| 80–100% | 2 | 80.8% | 100.0% (2) | 34.2%–100.0% |
Did the closing line beat the opening line?
The closing line's mean Brier score (0.597) and the opening line's (0.595) differ by 0.0014, within two standard errors (0.0074) of zero: on this sample the closing line is not measurably more accurate than the first price we recorded.
| Per match | Matches | Share |
|---|---|---|
| Closing line scored better | 41 | 45.6% |
| Opening line scored better | 49 | 54.4% |
| Price never moved | 0 | 0.0% |
| Moved toward the result | 42 | 46.7% |
| Moved away from the result | 48 | 53.3% |
Of the 90 matches whose price moved, 46.7% moved toward what happened.
Biggest upsets
- ฮัลล์ซิตี 2-0 แมนเชสเตอร์ ยูไนเต็ด — ฮัลล์ซิตี closed at 9.00 (10.8% implied), 2026-08-22
- นอตทิงแฮมฟอเรสต์ 0-1 Coventry City — Coventry City closed at 6.05 (16.1% implied), 2026-09-19
- ไบรตันแอนด์โฮฟอัลเบียน 3-0 อาร์เซนอล — ไบรตันแอนด์โฮฟอัลเบียน closed at 5.35 (18.2% implied), 2026-09-19
- แอสตันวิลลา 2-1 แมนเชสเตอร์ ซิตี้ — แอสตันวิลลา closed at 5.10 (19.1% implied), 2026-10-24
- Coventry City 0-1 ฮัลล์ซิตี — ฮัลล์ซิตี closed at 5.10 (19.1% implied), 2026-08-29
- แมนเชสเตอร์ ซิตี้ 0-1 ไบรตันแอนด์โฮฟอัลเบียน — ไบรตันแอนด์โฮฟอัลเบียน closed at 4.30 (22.6% implied), 2026-10-31
- ทอตนัมฮอตสเปอร์ 1-3 แอสตันวิลลา — แอสตันวิลลา closed at 4.11 (23.7% implied), 2026-09-19
- เชลซี 0-2 ทอตนัมฮอตสเปอร์ — ทอตนัมฮอตสเปอร์ closed at 3.90 (24.9% implied), 2026-10-24
- นิวคาสเซิลยูไนเต็ด 2-1 ลิเวอร์พูล — นิวคาสเซิลยูไนเต็ด closed at 3.76 (25.9% implied), 2026-08-23
- Coventry City 4-1 นิวคาสเซิลยูไนเต็ด — Coventry City closed at 3.71 (26.2% implied), 2026-10-12
More English Premier League
Other competitions
How these numbers are computed
Data window. Every English Premier League final that kicked off in the 365 days before 2026-11-15; the matches that qualified run from 2026-08-21 to 2026-11-08. The study is recomputed once a day.
Which finals count. A football final is counted only when the last clock reading we recorded reached minute 89, the collector's evidence that full time was played; finals in other sports must pass the site's plausibility checks (no tied final in a sport that cannot end level). A final must also be the one the site publishes at its own result page, under the competition it belongs to. 91 finals qualified.
Where the prices come from. Our archive records the two bookmakers as separate events, and only one of them carries the score. Each final is paired with the Pinnacle event for the same match. First by name: the same two teams (names normalised the way the live odds board merges the two books) on the same UTC date, with kickoff times no more than 30 minutes apart. Where no name matches exactly, by the odds board's own test for one game listed twice, among Pinnacle events filed under the same competition: the same sport and competition class, both sides recognisably the same teams, kickoffs no more than 15 minutes apart — and when Pinnacle lists the fixture the other way round, its prices are put back on the right teams. A final with two equally close candidates, a candidate whose sides cannot be told apart, or a Pinnacle event claimed by two finals is left out rather than guessed. 90 of the 91 finals found their Pinnacle event, 0 of them by the second test.
Opening and closing prices. The price is Pinnacle's full-time moneyline (1X2 where the sport can end level, home/away where it cannot). The closing price is the last one we recorded at or before kickoff; the opening price is the first one we recorded — not necessarily the first price Pinnacle posted, because our collector records a market from the moment it first sees it. The median match was first priced 329.3 hours before kickoff. A match counts only when every outcome has both an opening and a closing price: 90 did.
Removing the margin.Each outcome's probability is its inverse price divided by the sum of the inverse prices of all outcomes (proportional normalisation), so the probabilities of one match add up to 100%.
The measures. The favourite is the outcome with the highest closing probability; a match where two outcomes tie for highest has no favourite. The draw row covers only sports that can end level. The calibration table places every outcome of every match in a band by its closing probability — lower bound included — and counts how often outcomes in that band happened. The Brier score of a match is the sum, over its outcomes, of the squared gap between the probability and what happened (1 for the outcome that occurred, 0 otherwise); 0 is a perfect forecast. The opening and closing scores are compared match by match, and a difference is called real only when it exceeds twice its standard error. "Moved toward the result" means the probability of what actually happened — the winner, or the draw — was higher at the close than at the open. The 10 biggest upsets are the winning teams with the lowest closing probability; each links its result page. The 95% ranges are Wilson score intervals.
Publication floor. A competition's numbers appear once 30 matches qualify; below that the page states what the archive holds and stays out of search results.
Window: 365 days. Pairing tolerance: 30 minutes by name, 15 by the second test. Pinnacle events examined: 73647. The companion study, bookmaker margins by league, measures what each book charges; this one measures how right the sharper book's last price was.