Checking model
FairValue2025–26 snapshot14 leagues5,820 players
FairValue estimates footballers' Transfermarkt market values, makes model predictions inspectable and investigates how prediction errors differ across groups.
Solid what the market says Outline what the model says
02The Market
The model learns from, and is audited on, a selected population: players with a Transfermarkt valuation in the last twelve months, valued while playing in one of fourteen covered leagues, with at least 450 club minutes in the two years before. It is not the global transfer market, and nothing here should be read as if it were.
Market Landscape →League Landscape →The fourteen-leagues story →
03The Discrepancy
We expected the market to value non-UEFA players below comparable UEFA players. The audit found the reverse. Compared with UEFA players in the same leagues, non-UEFA players' Transfermarkt values sit about +10% above what Model A predicts for them. How far off the model is, on average, is about the same for both groups. What differs is the direction of the miss.
+0.093
Headline: signed residual gap, non-UEFA minus UEFA, compared within leagues, Model A, log scale. 95% CI +0.063 to +0.124.
Compared within each league the gap is +0.093, the headline figure. For scale, a single player's typical miss is ±0.41 (×1.50): the group difference is small next to individual error, and consistent.
Same gap, two models
04The Players
One record per player, the same from every corner of FairValue: observed value, both model estimates, the inputs behind them, the valuation history and the audit context. Below, the most valuable player in each position.
Fourteen leagues, fourteen front-runners the most valuable player in each covered league
Observed Model A Ratio = observed ÷ Model A estimate (held-out). A ratio above ×1.00 means the market value is above the estimate. It is not a verdict that a player is over- or under-priced.
05The Research
Two gradient-boosted models, evaluated only on players they did not see in training (five-fold, out-of-fold). Every estimate can be taken apart input by input, and the errors are audited by group.
Models
Model Aperformance, age, position + league and club0.891
Model Bperformance, age and position only0.669
Held-out R² on the log scale. Context dominates: league and club explain a large share of the market's pricing. Typical miss for Model A is ×1.50.
Value a profile in the Valuation Studio →Explainability
SHAP contributions break each estimate into multipliers, one per input, that add up exactly to the served prediction. Global importance is shown separately and labelled as such.
See an explanation for Haaland →Valuation Journey
The journey walks from the model's average starting value through each input that raised or lowered it, to the final estimate.
Follow Inside a valuation: Khéphren Thuram →Fairness
Signed and absolute error gaps, tier error rates, league strata, a calibration check and club-clustered intervals, at two minute thresholds.
Open the Fairness Audit →Methods
Data, population, validation, explainability and what the study cannot say, on one page.
Read the Methods →Live model check
Checking modelChecking the registered models…
Centre-Forward · LaLiga · age 24
€00.0M
Centre-Forward · LaLiga · age 24
€17.1M
Elite
Premier League
Serie AItaly
Ligue 1France
BundesligaGermany
LaLigaSpain
Pro LeagueBelgium
Süper LigTürkiye
EredivisieNetherlands
Liga PortugalPortugal
Premier LigaRussia
SuperligaDenmark
Super LeagueGreece
PremiershipScotland
Premier LihaUkraine
