How Accurate Are Sports Predictions? · Part 2 of 7
Can a Sports Model Beat the Wisdom of the Crowd?
Where the record can tell them apart, the betting market’s win probabilities score better than BBMI’s. In college basketball and the NFL it cannot tell them apart.
BBMI Research · November 21, 2026
On Nov. 18, BBMI and the betting market disagreed about a college football game by 13 points. BBMI made Massachusetts a 16.8-point favorite at home against Akron. The market had Massachusetts by 3.5. Akron won, 34-33.
16.8BBMI: Massachusetts favored by 16.8 points over Akron
3.5Betting market: Massachusetts by 3.5
34–33Final score, Nov. 18. Akron won.
Part 1 of this series argued that one upset cannot tell us whether a forecast was any good. It also cannot tell us which forecaster is better. So the question here is this: across thousands of games where both make a forecast, which one provides more accurate probabilities?
Why the market is a hard benchmark
A betting line is not one person’s prediction. It reflects sportsbook models, professional bettors and a market that keeps reacting to new information. By game time, much of what is publicly known about a matchup is already in the price.
That makes the market a hard benchmark. BBMI is not being measured against a random guess. It is being measured against a forecast shaped by thousands of decisions.
The idea is older than sports betting. In 1907 the statistician Francis Galton studied a contest at a livestock fair in Plymouth, England, where 787 people guessed the dressed weight of an ox. The middle guess was 1,207 pounds. The ox weighed 1,198, a miss of less than 1%.
787 guesses at the weight of an ox, 1907
1,207 lbThe middle guess
1,198 lbWhat the ox weighed
Source: Francis Galton, “Vox Populi,” Nature, March 7, 1907.
What the record shows
BBMI publishes its win probabilities beside the market’s. The chart compares them with the Brier score from Part 1: the average squared gap between the forecast and the result, counting a win as 1 and a loss as 0. Lower is better, and saying 50% for every game scores 0.25. Each row scores both sides on the same games. In each sport the market number was captured at the same time as BBMI’s forecast or earlier, so the market did not have later information.
The market leads in three sports. Two are too close to call.
Interval excludes zeroInterval includes zero
BBMI scored betterMarket scored better
College football2026 · 631 games
+0.0111
College baseball2026 · 2,328 games
+0.0094
MLB2026 · 1,695 games
+0.0048
NFL2026 · 151 games
−0.0015
College basketball2025-26 · 2,151 games
−0.0032
−0.010+0.010+0.0200
Each dot is BBMI’s Brier score minus the market’s on the same games. The Brier score is the average squared gap between a forecast and the result, so lower is better, and saying 50% for every game scores 0.25. The line through each dot is the 95% interval, from resampling the games 10,000 times. A line that crosses zero means the record cannot say which forecast is better.
The market leads in college football, college baseball and MLB. In each, the interval on the gap excludes zero, so the lead is unlikely to be chance. The widest gap is in 2026 college football: 0.0111 over 631 games. The Akron game is one of them. In college baseball the gap is 0.0094 over 2,328 games, and in MLB it is 0.0048 over 1,695.
College basketball and the NFL are too close to call. Across 2,151 college basketball games from 2025-26, BBMI’s Brier score was 0.1958 and the market’s was 0.1990. BBMI’s is lower, but the interval on the difference runs from 0.0075 in BBMI’s favor to 0.0011 in the market’s.
The NFL result is similar on a smaller sample. Over 151 games in 2026, BBMI’s score was 0.2126 and the market’s was 0.2141, with an interval from 0.0108 in BBMI’s favor to 0.0079 in the market’s. In neither sport can this record say which forecast is better.
The numbers behind the chart
Sport
Games
BBMI
Market
Difference
95% interval
College football, 2026
631
0.1865
0.1754
+0.0111
+0.0033 to +0.0193
College baseball, 2026
2,328
0.2303
0.2209
+0.0094
+0.0044 to +0.0143
MLB, 2026
1,695
0.2467
0.2419
+0.0048
+0.0012 to +0.0085
NFL, 2026
151
0.2126
0.2141
−0.0015
−0.0108 to +0.0079
College basketball, 2025-26
2,151
0.1958
0.1990
−0.0032
−0.0075 to +0.0011
BBMI and Market are Brier scores on identical games. Difference is BBMI minus market. A negative number means BBMI scored better.
Beating the market on average and adding information beyond the market are different questions. A model can have a higher overall error and still identify particular games where the consensus is mistaken. Whether BBMI does that consistently requires a separate test.
Investors have run this test for decades
Investment managers face a similar test, although the measurement is different. An index fund aims to track a market benchmark. Active managers try to outperform it. S&P’s SPIVA research has kept score of that contest for more than 20 years.
In the SPIVA U.S. Mid-Year 2026 scorecard, 67% of actively managed large-cap U.S. funds underperformed the S&P 500 in the first half of 2026. Over the 20 years ending June 30, 2026, about 93% did. SPIVA measures fund returns after costs, not forecast error, so the two records are not the same kind of score.
Active large-cap U.S. funds that underperformed the S&P 500
67%In the first half of 2026
93%Over the 20 years ending June 30, 2026
Source: S&P Dow Jones Indices, SPIVA U.S. Mid-Year 2026 scorecard. Fund returns after costs.
The lesson is not that independent analysis is pointless. It is that a strong benchmark already incorporates a great deal of information, and claiming an advantage over it requires evidence.
The takeaway
The betting market is a formidable forecasting benchmark. In the sports where this record can separate the two, the market’s probabilities score better than BBMI’s. In college basketball and the NFL it cannot separate them.
That does not mean an independent model has no value. The important question is not whether a model makes convincing predictions. It is whether it can show, over a meaningful sample, that those predictions improve on what is already known. That holds in sports, in investing and in most fields that try to predict the future.
Next in the series: why predicting a player is harder than predicting a team.
Data: BBMI forecast and market records, scored on Nov. 21, 2026, using only games where both a BBMI probability and a market probability exist, and the 2026 season for football. For that reason these figures differ from the pooled figures on the Model Performance pages. In basketball and baseball the market probability comes mainly from the sportsbook’s moneyline price. Removing the sportsbook’s margin changes those scores by 0.0005 or less where it can be checked. In football the market posts a point spread, which is converted to a probability. In college football BBMI’s spread is converted the same way. In the NFL BBMI’s probability comes from its simulation. Each interval is for one sport taken alone and is not adjusted for making five comparisons.
Continue the series: How Accurate Are Sports Predictions?