Sunderland vs Arsenal: Our Biggest Disagreement With the Market So Far

Sunderland host Arsenal tonight, and our model has produced its largest gap against bookmaker pricing since we began recording forecasts.

Market probabilities have the margin removed. "Strong" is the highest rating our value scale issues, and this is the first time it has appeared in anything we've published.

Why the model likes the outsider

The single fact doing most of the work: Arsenal have zero wins in their last five matches.

They are still priced at 1.575, which implies a 60.2% chance of winning away at a promoted side. Our model reads the same form data and lands at 55.2%. Five points of disagreement on a short price is a lot, and it flows straight into the Sunderland number: 22.6% against 14.9% implied.

Arsenal's team news doesn't help the case for the price. Mosquera, William Saliba and Timber are all out. For a side averaging 61% possession, losing that much of the defensive spine matters more than the possession figure suggests — control of the ball is worth less when the structure behind it is patched.

Sunderland are not creating much, at 12 shots per game. But they don't need to create much at 6.35. That price only needs to be wrong by a little.

Now the part that should make you cautious

This morning we published our first forty graded forecasts. The headline finding was unflattering: 42.5% accuracy, and in forty attempts the model picked the draw exactly zero times while eleven matches finished level. We named the specific defect — the draw probability comes out too low — and said any pick where the draw sits under roughly 25% in a balanced fixture should be treated as suspect.

Look at the table again. Our draw probability here is 22.2%. Below the threshold we set ourselves a few hours ago.

So the honest reading is split. The direction of the disagreement is defensible: Arsenal's price looks short for a team that hasn't won in five, and our model is flagging that rather than following the market. The specific numbers carry a known bias, and that bias pushes weight away from the draw and onto the two win outcomes, inflating both the Sunderland figure and, in the other direction, the case against Arsenal.

The other thing worth noticing: our Premier League record so far is nought from five, and every one of those five picks was the home team. Tonight the model goes the other way and picks the away side. Whether that's the model correcting or simply being wrong in a new direction, we'll know by the final whistle.

What the model expects to see

A tight, low-scoring game. Over 2.5 goals comes out at 35.4%, both teams to score at 39.5%, and the most likely scoreline is 0-1. That is a model describing a nervous, low-tempo evening rather than an away side sweeping through a promoted team.

Confidence on the fixture is 0.64, rated High, which is above our recent average and mostly reflects that both squads have enough recent data to work from.

The record

This forecast was written down before kick-off, with the numbers above, and it will be graded automatically when the match finishes. If Arsenal win comfortably it stays on the track record as a miss, the same way the five Premier League misses from this week are still sitting there.

That's the deal we've made with the numbers: publish first, grade afterwards, delete nothing.

18+. A 7.6-point gap on a 6.35 shot is not a tip. It is a disagreement, from a model that has been wrong 23 times out of 40.