Bandwagon Axle 2.0
Axle 2.0's win chances, graded against the final score for every game it gave one before kickoff, since Oct 29. Recomputed nightly, last on Nov 4.
All versions since Oct 28: 77% straight up on 479 games, Brier 0.17. This counts the lines of Axle 1.1, Axle 2.0; the numbers above are Axle 2.0's own.
By league
| League | Games | Favorite won | Brier | Closer than the market |
|---|---|---|---|---|
| NCAA Division I men's basketball | 137 | 83% 114 of 137 | 0.14 | – |
| NCAA Division I women's basketball | 116 | 87% 101 of 116 | 0.12 | – |
| NCAA Division I FBS football | 54 | 81% 44 of 54 | 0.15 | 31 of 53 |
| National Basketball Association | 39 | 62% 24 of 39 | 0.23 | 12 of 39 |
| National Hockey League | 33 | 61% 20 of 33 | 0.23 | 15 of 33 |
| Major League Soccer three-way | 15 | 60% 9 of 15 | 0.57 | – |
| National Football League | 14 | 64% 9 of 14 | 0.19 | 7 of 14 |
| Premier League three-way | 10 | 60% 6 of 10 | 0.64 | 5 of 10 |
| La Liga three-way | 10 | 80% 8 of 10 | 0.47 | – |
| Serie A three-way | 10 | 70% 7 of 10 | 0.51 | – |
4 leagues have fewer than 10 graded games and are not shown yet.
Soccer is graded three-way: the favorite is the largest of home, draw and away, and its Brier score counts all three. A three-way Brier score runs from 0 to 2, so it is not comparable with the others.
Calibration
465 gamesWhen Axle gives the home side 60%, does the home side win about 60% of the time? Up is how often it won, across is what Axle said.
10 or more games 5 to 9 fewer than 5
The numbers behind the chart
| Axle said | Games | Said (average) | Home side won |
|---|---|---|---|
| 10–20% | 10 | 14% | 10% |
| 20–30% | 20 | 26% | 25% |
| 30–40% | 31 | 36% | 48% |
| 40–50% | 59 | 46% | 49% |
| 50–60% | 57 | 55% | 53% |
| 60–70% | 79 | 65% | 75% |
| 70–80% | 92 | 75% | 86% |
| 80–90% | 92 | 84% | 96% |
| 90–100% | 25 | 94% | 92% |
Lately
straight up on 465 games · Brier 0.17 · closer than the market in 70 of 149
straight up on 465 games · Brier 0.17 · closer than the market in 70 of 149
How to read this
Straight up is how often the side Axle gave the better chance went on to win. It is the easiest number to understand and the easiest to misread: a favorite at 90% and one at 55% count the same, and a coin-flip game is nearly all luck.
Calibration matters more than the win rate. A model that says 70% should be right about 7 times in 10 over many games. The chart and the Brier score check exactly that: they reward honest chances and punish confident misses. A model can win a high share of its picks and still be badly calibrated.
Draws. Soccer has three results. Axle gives a chance to home, draw and away, the favorite is the largest of the three, and a draw counts as a result like any other. Where a draw is impossible, a tied score is left out.
Small samples. A rate on a few games says little, so a league is shown from 10 graded games, the overall figure from 50, and the comparison with the markets from 30 games that have a market price. Every percentage has its games beside it.
The markets are a comparison, never an input. "Closer than the market" counts the games where Axle's chance for what actually happened was higher than the prediction markets' devigged price before kickoff. Nothing here is a bet or a pick; Bandwagon's model, not a sportsbook line. We do not take wagers.
The longer, out-of-sample backtest for every league, the constants and the version history are on the methodology page. Today's lines.
Bandwagon Axle 2.0 was right 78% of the time straight up on 465 games, graded against the final score from the line stored before kickoff. Brier score, calibration and a league-by-league table.