World Cup 2026 Semifinal: xG Model vs Bookmaker Odds

Expected Goals (xG) has been the standard metric for evaluating football chances since roughly 2015. Unlike the scoreline, xG measures the quality of created chances — regardless of who converts them. For a tournament with high stakes such as the World Cup semifinals, it's the best publicly available forecasting foundation.

**What xG actually measures.** Every shot gets a probability between 0 and 1 of becoming a goal — based on distance, angle, body part, action type (open play, set piece, counter) and keeper position. A penalty is xG ≈ 0.78, a long-range shot from 25m more like 0.03. Across a season, xG is a very good predictor of future goal output — significantly better than actual goals, which are heavily variance-driven.

**xG in single matches — with caution.** Over a single semifinal, xG has limited predictive power. A team can win xG 0.8 vs 2.4 (big-chance conversion, keeper performance, red cards). For predictions you need xG values across the last 8–12 games — that averages out the variance.

**How bookmakers use xG.** Big bookmakers run their own xG models but additionally weight market flow (where the money sits), historical tournament performance and team news. That creates systematic deviations: brand teams (Brazil, Germany, France) often get slightly better odds than their xG justifies — because heavy public money sits there and the bookmaker shortens the odds to balance the book.

**Where value can appear.** (1) Underdogs with a strong xG trend that fly under the media radar. (2) Over/Under 2.5 goals when combined xG > 3.0 but the Over price is 2.10. (3) Handicap −1 on a favourite whose xG differential shows +1.5 but odds are only mildly favourite (see [World Cup 2026 Handicap Betting](/en/blog/world-cup-2026-handicap-betting-examples)).

**xG data sources.** Publicly available via Understat, FBref, StatsBomb Open Data or Opta-based portals. FIFA and several stats providers publish live xG during matches. Always use at least two sources — models differ by up to 20%.

**Working with xG: a worked example.** Semifinal: Team A xG average 1.9, Team B xG average 1.3 (offensive). Team A xG-against 0.9, Team B xG-against 1.4 (defensive). Estimated goals: A ≈ 1.9 × (1.4 / tournament avg 1.15) ≈ 2.3; B ≈ 1.3 × (0.9 / 1.15) ≈ 1.0. Forecast ~2.3:1.0. If bookmaker odds on A −1 handicap sit at 2.20 but Poisson probability of a ≥ 2-goal difference is 46%, fair odds are 2.17. Marginal, but slightly positive. More on the maths in [Value Betting Explained](/en/blog/value-betting-explained).

**Model limits.** xG ignores weather, late injuries, tactical shifts. In knockout ties with extra time, xG per 90 minutes is misleading. And: bookmakers often have a few minutes' team-news head start — you can't outbid that.

**Responsibility.** Even the best model doesn't guarantee wins. Turning a value analysis into an emotional bet loses long-term. Stake never more than 1–2% of bankroll per bet (Kelly fraction, see [Kelly Calculator](/en/kelly-criterion-calculator)) and keep the BZgA hotline (0800 1 372 700) in mind when betting matters more than the match.

**Bottom line.** xG is the best publicly available foundation for World Cup predictions — but no oracle. Anyone averaging xG across multiple sources, doing Poisson properly and only betting when there's a clear deviation from bookmaker odds finds slight edges across a tournament. The rest is discipline. Reminder: operators listed on Casinokeller only accept players residing in Germany.