Skip to main content
BestSlip

Build smarter slips. Compare better odds. Track your performance.

Predictions

Combined Analysis

Where your model and the market disagree, and whether those disagreements have been right.

This one needs your predictions

The comparison is between your model and the market, so it needs predictions of yours to compare. Those are private to your account — nobody else can read them and this page cannot be shown without one.

The method below is readable without an account, and this page is not one of the paid ones — it needs a prediction of yours, which is a different thing from needing a subscription.

Log inCreate an accountBuild a model first

How the comparison works

The two numbers are made comparable before they are subtracted. A posted price implies more than 100% across a market, because the difference is the book's cut. Reading a probability straight off it would make every market look likelier than it is and would hand your model a free edge on every outcome. So the margin comes out first — twice, by two different methods, because removing it requires assuming how the book spread it and the common assumption inflates long prices.

The market side is a consensus, not one book. Every book that priced the whole three-way is de-vigged on its own and the results are averaged. De-vigging each book separately matters: averaging raw implied probabilities would weight the widest book heaviest, which is the one whose numbers are furthest from a probability to begin with.

A gap is not an edge. The figure shown is the size of a disagreement between your model and the market. It becomes value only if your model is the one that is right, and the market's closing prices are the hardest forecast there is to beat. That is what the second panel is for: it grades both sides against what actually happened rather than taking the disagreement at face value.

Both sides are scored by the same function. Your model's Brier score and the market's come from one implementation, not two. The entire output of this page is a difference between two scores, and a difference between two copies of a formula would measure the copies as much as the forecasts — a disagreement in a rounding or a floor would read as your model beating the market.

The comparison is also paired: the two scores are differenced game by game rather than averaged separately. Most of the variation in a Brier score is which games were on the slate, and both forecasters share it, so cancelling it is what makes a verdict possible after a month of predictions instead of a season.

The market half of all this is recorded by Bookmaker Intelligence, which is also where the closing prices, the sampling resolution and the rest of the caveats live. The model half is Custom Model.

Bet Slip

Loading slip…