Every rating system makes the same promise: give it enough results and it will tell you who should win. MMA keeps testing that promise. Favourites lose often enough to keep the sport unpredictable, but not so often that the models are worthless. Understanding where the uncertainty comes from is more useful than pretending it isn’t there.
How Ranking Systems Estimate Win Probability
Most modern rating systems descend from Elo. Each fighter carries a number, and the gap between two numbers converts into an expected win probability. In the standard Elo form, with a rating difference d, the favorite’s expected score is:
E = 1 / (1 + 10−d/400)
A 100-point edge works out to roughly 64%. A 200-point edge is about 76%. After each fight, ratings move in proportion to how surprising the result was, scaled by a K-factor.
Glicko adds a ratings deviation, a measure of how confident the system is in each number, which matters in a sport where fighters compete two or three times a year. Whole-History Rating goes further and re-estimates a fighter’s strength across their entire career rather than updating one bout at a time.
What all of them leave out is the information that never reaches the results column: injuries, weight cuts, camp changes, stylistic matchups, a move up a division, the specific danger of one opponent’s best weapon. Betting markets price some of that in, which is one reason closing odds remain a tough benchmark for any pure rating model.
Anatomy of an Upset
Upsets tend to cluster around a handful of recurring causes.
- Styles. Ratings compress a fighter into a single number, but matchups are not transitive. Holly Holm’s win over Ronda Rousey at UFC 193 in 2015 was a disciplined striker keeping a dominant grappler at range, the kind of stylistic problem a rating cannot see.
- Short notice. Late replacements carry hidden variables in both directions. Michael Bisping took his UFC 199 rematch with Luke Rockhold in 2016 on roughly two weeks’ notice and knocked him out in the first round to win the middleweight title.
- Layoffs and activity. Long gaps widen the real uncertainty around a fighter’s current level, yet many systems treat a fighter returning from two years out much as they did before the break. Glicko-style deviation helps here; plain Elo does not.
- One clean shot. MMA has a high-variance finishing mechanic most sports lack. Leon Edwards was behind on the cards against Kamaru Usman at UFC 278 in 2022 when a head kick in the final minute of the fifth round took the welterweight title.
The last category matters most. In a sport with four-ounce gloves, a fighter can be outclassed for 24 minutes and still win. Even a well-calibrated 80% favourite should expect to lose one fight in five.
How Often Favourites Actually Lose
The honest test of any model is calibration: when it says 70%, does the favourite win roughly seven times in ten?
Large samples of UFC results suggest betting favourites win somewhere around two-thirds of the time, with heavier favourites winning more often, as they should. The market is, broadly, well calibrated. Rating systems built from results alone tend to sit a little behind closing odds on raw accuracy, but they are close enough to be useful, and they are transparent in a way the market is not. You can see exactly why a system rates a fighter where it does.
The misses are worth studying too. Many of the biggest come in fights involving short-notice replacements, fighters changing weight class, long layoffs and opponents with thin records, which is exactly where the inputs are weakest. That tells you something important: much of MMA’s unpredictability is not pure randomness but uncertainty about the inputs. Better data narrows it. It never removes it.
True Randomness, by Contrast
Compare that with a game where there is nothing to model at all.
In a certified number draw, the result comes from a random number generator or draw machine that has been independently tested to show every combination is equally likely and each draw is independent of the last. An online lottery such as Emirates Draw, which streams its draws live and runs on independently certified draw systems, is a clean example. There are no styles, no layoffs, no short-notice replacements and no hidden form. Last week’s winning numbers carry exactly zero information about this week’s.
That is the key difference. In MMA, a better model, more data or a sharper read of a matchup can genuinely improve a forecast. In a certified random draw, nothing can. Tables of “hot” and “cold” numbers, pattern-spotting and prediction systems all fail for the same reason: the process has no memory. The only figure that matters is the fixed probability of your numbers matching, and nothing you do moves it.
Takeaway: Know Which Kind of Uncertainty You’re Facing
A title fight and a weekly draw both involve uncertainty, but not the same kind.
MMA uncertainty is mostly epistemic. It comes from what we don’t know about the fighters, and it shrinks with better information. That is why ratings improve with data, why sharp bettors look for edges around short-notice bouts and weight changes, and why calibration is worth checking.
Lottery uncertainty is aleatory. It is built into the process by design, and no amount of information reduces it. There is no edge to find.
Before you put money on either, be clear which one you’re dealing with. On fight night, research can help, though the variance stays large. In a certified draw, treat a ticket as entertainment with a known price and long odds, set a budget, and leave the prediction systems alone.