xgEdge

Model vs market

We price football matches. You can compare our price with the market.

This section sets the model's probability for each outcome beside a bookmaker's price for it. Fair odds are the price the model breaks even at, and the edge is how far the market's price sits above or below it.

No match kicks off in the next 72 hours.

Not betting advice. Estimates from a statistical model, published for reading. A positive edge is a statement about an average over many comparable prices, not about the match in front of you.

No match kicks off in the next 72 hours, so there is nothing to price yet.

How well calibrated is the model?

When the model gives an outcome 60%, does it happen about 60% of the time?

685 matches · 11645 forecasts

Brier score
0.210
Base rate 0.215
Log loss
0.608
Base rate 0.621

Lower is better for both. The base rate is a forecast that gave every selection how often it happened over this same sample, which it could only have known with hindsight.

View full model performance

Reliability: the model's probability against how often it happened Each point is a band of forecasts: across, the model's average probability in the band; up, how often those selections happened. A point on the diagonal is perfectly calibrated. A list of the plotted bands follows the chart. 0% 0% 25% 25% 50% 50% 75% 75% 100% 100% Too cautious Too confident

Model probability → · ↑ Happened

Points close to the diagonal mean outcomes happened about as often as the model predicted. The size of a point is its number of forecasts.
  • Within 2 pp
  • 2–4 pp off
  • More than 4 pp off
  • 0–10%: model average 8.2%, happened 9.5%, 84 forecasts.
  • 10–20%: model average 16.0%, happened 17.0%, 958 forecasts.
  • 20–30%: model average 25.5%, happened 25.4%, 1997 forecasts.
  • 30–40%: model average 34.8%, happened 37.4%, 1853 forecasts.
  • 40–50%: model average 44.9%, happened 45.9%, 1761 forecasts.
  • 50–60%: model average 54.9%, happened 54.9%, 1600 forecasts.
  • 60–70%: model average 64.7%, happened 61.3%, 1291 forecasts.
  • 70–80%: model average 75.0%, happened 73.8%, 1234 forecasts.
  • 80–90%: model average 84.1%, happened 82.2%, 793 forecasts.
  • 90–100%: model average 91.8%, happened 90.5%, 74 forecasts.

Qualified selection record

Settled qualified selections
876 across 412 matches
Won–lost–void
295 won, 572 lost, 9 void
Units
−94.36u
ROI
−10.8%

Equal stake of one unit per qualified selection.

Selections from the same match may be correlated.

Historical selections use the model probability and bookmaker price recorded at the last qualifying pre-match evaluation. Nothing is re-priced after the result.

How this record is calculated

The daily and sharp combinations and their graded results have their own pages: Slips · Results

View full model performance