Team and match goals — over/under, BTTS and the goals model
Match totals (over/under 2.5 goals), team goals and both teams to score are modelled from expected goals (xG), not raw scorelines. SharpXI fits attack and defence strengths across the league to price them.
How SharpXI models it
A Dixon-Coles bivariate-Poisson model, fit on team xG with recency weighting, estimates each team's attack and defence strength, home advantage and the low-score correlation. From that we derive over/under lines, team totals and BTTS. Goals is our most demanding market and we say so — it's the most efficiently priced market in football, and the track record shows the honest margin.
Go deeper: How we model team goals
Where the value tends to sit
Value in goals markets comes from teams whose xG diverges from their results — over- or under-performing finishing that the market anchors to.
Premier League team leaders this season
In the 26/27 Premier League season, Brighton led for goals a game at 3.20, ahead of Man City (2.60) and Brentford (2.00). That's against a Premier League team average of 1.41. Per-match rates from completed matches, not odds.
Per-match rates across the league (26/27 season, completed matches) — the teams whose underlying numbers drive this market.
| Team | Per match |
|---|---|
| Brighton | 3.20 |
| Man City | 2.60 |
| Brentford | 2.00 |
| Chelsea | 2.00 |
| Newcastle | 1.80 |
| Arsenal | 1.60 |
| Man Utd | 1.60 |
| Ipswich | 1.40 |
Frequently asked
Does SharpXI use xG for goals markets?
Yes. The goals model is fit on team expected goals rather than raw goals, which is a less noisy signal of underlying attacking and defensive quality.
Why is goals one of the hardest markets to beat?
Goals are the most efficiently priced numbers in football — the world's modelling effort concentrates here, so the margin over the base rate is always slim, and in some leagues it doesn't clear the baseline at all. The Premier League track record shows the honest number rather than hiding it.
SharpXI is a statistical tool, not a betting operator, and no model guarantees a result. The honest scoreboard is closing-line value over a large sample. 18+ — please gamble responsibly.