Premier League weekend, three widest gaps

Ten fixtures were assessed this weekend. Seven produced numbers close enough to the market that the difference sits inside our own error bars, which is the ordinary result and the one nobody writes about. Three did not.

#The away side at 3.40

Our composed number puts this at 33%, the price implies 29.4%, and the margin on the three-way market is 3.8%. Strip that out properly and the market is at about 28%, so the gap is five points.

Most of it comes from the research half rather than the model. Two independent sources have the home side missing both first-choice centre backs, and the fitted model, which reads results only, has no way of knowing that.

#The under, at 1.95

Both sides have played four matches in twelve days and the second of them travels on Friday. The scoring rates the model uses are drawn from a season in which neither side had a schedule like this, and fixture congestion depresses goals more reliably than it depresses results.

Our number is 56% under, against an implied 51%.

#The draw at 3.90

This one is the model's, not the research's. The two sides are separated by less than a goal of expected difference, and when two teams are that close the draw is structurally underpriced by markets that treat it as a residual rather than as an outcome with its own distribution.

#How to read the seven we said nothing about

The seven fixtures with no published gap are the important part of this page. A weekend where every match produces a disagreement is a weekend where the model is miscalibrated, not one where the market has collapsed. Agreement is the expected result and it is what most of these assessments produce.

#What this does not tell you

Three gaps in a weekend is three data points, and a weekend where all three lose is entirely ordinary. The reasoning above is what we think; the number is an estimate and not a promise, and the way to read a week of these is set out in what variance does to a good bet.