How accurate are UFCal's MMA fight predictions?
How the model's pre-fight probabilities actually fared, graded against every decided fight in the database. Nothing is cherry-picked: these are the same numbers shown on bout pages before each fight, recomputed live.
UFCal's Power Index has correctly picked the winner in 62.5% of the 19417 MMA fights graded so far, with a Brier score of 0.239. It is well calibrated: when the model gives a fighter a 50%-55% chance, that favorite wins 56.2% of the time.
- Favorite accuracy
- 62.5%
- Fights graded
- 19417
- Brier score
- 0.239
When the model says 70%, does the favorite actually win about 70% of the time? Predicted vs. observed, by confidence bucket.
| Model says | Avg predicted | Favorite won | Fights |
|---|---|---|---|
| 50%–55% | 52.2% | 56.2% | 7168 |
| 55%–60% | 57.4% | 59% | 5379 |
| 60%–65% | 62.3% | 67.7% | 3734 |
| 65%–70% | 67.2% | 73.9% | 2121 |
| 70%–75% | 72.1% | 77.6% | 1172 |
| 75%–80% | 77% | 79.2% | 518 |
| 80%–100% | 82.7% | 82.3% | 192 |
| Year | Fights | Accuracy |
|---|---|---|
| 2026 | 946 | 62.3% |
| 2025 | 1802 | 61.3% |
| 2024 | 1781 | 61.6% |
| 2023 | 1793 | 64.4% |
| 2022 | 1497 | 63.5% |
| 2021 | 1443 | 62.1% |
| 2020 | 1054 | 63.7% |
| 2019 | 1469 | 61.3% |
| 2018 | 1253 | 60.2% |
| 2017 | 1049 | 62.7% |
| 2016 | 956 | 62.6% |
| 2015 | 781 | 60.8% |
| 2014 | 755 | 64.9% |
| 2013 | 505 | 63.8% |
| 2012 | 407 | 61.2% |
| 2011 | 335 | 63% |
| 2010 | 275 | 61.5% |
| 2009 | 230 | 57.8% |
| 2008 | 241 | 62.2% |
| 2007 | 196 | 66.3% |
| 2006 | 171 | 67.3% |
| 2005 | 92 | 70.7% |
| 2004 | 46 | 50% |
| 2003 | 39 | 61.5% |
| 2002 | 52 | 67.3% |
| 2001 | 38 | 71.1% |
| 2000 | 42 | 66.7% |
| 1999 | 39 | 59% |
| 1998 | 23 | 65.2% |
| 1997 | 29 | 65.5% |
| 1996 | 30 | 73.3% |
| 1995 | 27 | 70.4% |
| 1994 | 18 | 77.8% |
| 1993 | 3 | 66.7% |
Every number on this page is recomputed live from the stored pre-fight probabilities, the same ones printed on bout pages before each fight happens. Draws are excluded (no verdict to grade) and dead-even 50/50 prices grade no pick. Ratings and probabilities span every organization we track, back to 1993.