Tool · Study 05
Will his game travel?
Estimate how much of a player's attacking output may survive a move between European leagues.
What this predictsThe model predicts post-transfer attacking output using player history, league context and transfer characteristics. It does not predict overall player quality or career success.
Pick a player, a league and a club.
The model estimates his first-season xG + xA per 90 after the move — and how unsure it is.
See the model tested against Manchester United's 2024 recruitment
What usually happens when players move?
Output drops after any strong season, move or no move — . So Study 5 compares movers with : same club, role, output and age. Against them, the average move cost about 3%: movers kept 97% (95% 93–101%).
One sets the direction: hardest to enter is the Premier League, easiest the Bundesliga and Ligue 1.
| League | Moving into | Moving out of | Moves in |
|---|---|---|---|
| Premier League | 82%76–87% | 117%107–127% | 137 |
| La Liga | 100%92–108% | 100%91–109% | 86 |
| Serie A | 96%89–103% | 99%91–108% | 97 |
| Bundesliga | 119%107–130% | 90%82–98% | 47 |
| Ligue 1 | 118%108–129% | 89%82–95% | 61 |
Study 05 · 428 summer moves 2016/17–2025/26 · xG + xA per 90 vs comparable stayers · 95% intervals
How much does context improve the prediction?
| Model | ||
|---|---|---|
| Naivepost = pre | 0.129 | 0.58 |
| Role averagepull toward the role average | 0.117 | 0.65 |
| Contextage, minutes, both clubs' strength, leagues | 0.102 | 0.74 |
| Final modelcontext + three-season history | 0.096 | 0.76 |
Each step adds information, while the final model is evaluated only once on unseen transfers. Lower MAE and higher R² are better.
Model validation
142
unseen transfers
- Test period
- 2023/24 — 2025/26
- Locked
- Model fixed before the test seasons were opened
- Beat the naive guess by
- 0.033 xG + xA per 90 (CI 0.018–0.049)
- ≥75% probability
- 0.80 · 0.179 vs 0.224
- 80% ranges
- covered 87% of test moves
This prediction model was fixed before the test seasons and evaluated once on 142 unseen moves. The calculator uses the same specification refitted on all 428 moves.
Known weakness
Where the model can fail
The model is built primarily around attacking output. It cannot see defensive actions, completed passes, carries or possession. And its league effect is a fixed amount, not a percentage, so low-output central midfielders moving into the Premier League are under-predicted by about 0.06 xG + xA per 90.
Manuel Ugarte
Manchester United, summer 2024 — standing at 1 June 2024 with only what was known then.
xG + xA per 90 · predicted range 0.01–0.05 · Study 05 case study
Ugarte is a useful example of why model output should not be interpreted as a complete player evaluation: he was bought for defensive work this data cannot see.
What this doesn't tell you
The model does not predict
- Injuries
- Adaptation outside attacking output
- Defensive contribution
- Possession or carrying contribution
- Transfer fee
- Contract
- Team tactics
- Fixture congestion
- Cup competition
- Overall player value
ScopeForwards and midfielders with ≥ 900 minutes in 2025/26 · defenders and goalkeepers excluded
DataUnderstat · football-data.co.uk odds (club strength) · Transfermarkt-derived dates of birth
How the prediction is made
Technical detailStudy 5's final model
Target. First-season xG + xA per 90 after a summer league move (≥ 900 minutes before and after). = after ÷ before.
Inputs. Last season's output and the three-season level (both by role group), seasons of history, age (with a curve), minutes, origin and destination league, and both clubs' pre-match market strength, plus a promoted-club term. Nothing else: role, consistency, profile distinctiveness, penalty duties and loans were tested and added nothing.
Uncertainty. From the model's own out-of-sample errors in rolling folds (2019/20–2025/26), per role group: that gives the 80% prediction range and the chance of keeping ≥ 75%, which is only reported when last season's output is at least 0.10.
Precomputed. Every eligible player × destination club (834 × 101) was predicted with the research's production model; this page only looks results up. Inputs as of 2025/26, for a move in 2026/27.
No “transfer score”. The research has no validated 0–100 rating, so this tool shows the model's actual outputs and their uncertainty instead.