Skip to content

Latest commit

 

History

History
33 lines (23 loc) · 1.74 KB

potential.md

File metadata and controls

33 lines (23 loc) · 1.74 KB

Player potential backtest

Generated 2026-10-15T18:44:18+00:00. Target: log market value h years after a 1 June snapshot (transfermarkt-datasets, CC0). Time split: every training target is dated before the test snapshots.

2 years ahead

Train snapshots 2012-2021 (133373 player-years); test 2023-2024. 30% of test players had no valuation 2 years on and are not scored.

Method Players MAE (log) Median abs. % error Spearman 10-90% band coverage
Potential model 27699 0.597 42% 0.809 76%
Age curve (median change by age) 27699 0.653 45% 0.800 —
Persistence (value unchanged) 27699 0.780 53% 0.729 —
Potential model, age 21 or under 4789 0.894 61% 0.711 73%
Age curve, age 21 or under 4789 0.979 67% 0.690 —

Most used features: log_league, log_mv, age, minutes_share, age2, height

4 years ahead

Train snapshots 2010-2019 (99066 player-years); test 2021-2022. 48% of test players had no valuation 4 years on and are not scored.

Method Players MAE (log) Median abs. % error Spearman 10-90% band coverage
Potential model 23072 0.758 54% 0.719 76%
Age curve (median change by age) 23072 0.878 60% 0.707 —
Persistence (value unchanged) 23072 1.148 75% 0.533 —
Potential model, age 21 or under 4637 1.032 69% 0.581 76%
Age curve, age 21 or under 4637 1.151 73% 0.575 —

Most used features: log_league, log_mv, height, age, age2, mv_change_2y

Market value mixes ability with age, contract and hype. The dossier shows the projection as a level (compared with current Premier League players) next to its 10-90% band, never alone.