Can Open Data Spot the Next €30m Player?

Every summer, clubs spend billions chasing the same insight: which cheap young player is about to become expensive. The received wisdom is that you need scouts, contacts and proprietary tracking data. This project asked a smaller question — how far can you get with data anyone can download?

The answer: further than you'd think, with important caveats. A transparent screen built from open data ranked its top-20 players at a 55% "hit" rate against a 33% base — and a learned model across fourteen seasons pushed that to a 3× lift, while quietly confirming what football economists have argued for years about what actually drives a player's price.

The setup: open data, ten leagues, one question

The pool is the ten strongest leagues by Opta's Power Rankings — the traditional Big Five plus the Championship, Primeira Liga, Eredivisie, and the Brazilian and Argentine top flights. Performance data comes from FBref (StatsBomb-powered per-90 metrics for roughly 6,700 players); market values, ages, positions and transfer histories from a Transfermarkt-derived dataset.

The brief: find players who are young, cheap, playing meaningful-but-not-nailed-on minutes, and whose numbers suggest more than their price tag — including players who look mis-labelled by position. Every ingredient is grounded in the scouting and player-valuation literature: role-based profiling (Bergkamp et al., 2022), and the econometric determinants of transfer value (Poli, Besson & Ravenel, 2022; Franceschi et al., 2023).

The screen: four signals, one score

Each player gets four scores, always compared within their position so a full-back is never measured against a striker:

Performance — attacking output and creation per 90 (shots, key passes, shot- and goal-creating actions).
Intensity — pressing, tackling, carrying and recovery volume: an open-data proxy for "can handle a high-tempo game," since real GPS data is proprietary.
Misprofiling — a model that flags players whose statistical fingerprint matches a different role than the one they are listed in.
Underused — strong per-90 output on a low share of team minutes: the classic under-priced profile.

Does it actually work?

A screen is only worth anything if its picks come true. So I ran a strict point-in-time back-test: freeze everything we'd have known in June 2025, rank the players, then check what happened over the following year. A "hit" is a player who was either sold for a real fee (≥€10m) or whose market value rose by at least 50%.

Back-test lift chart — **Fig 1.** Among eligible young players, 33% "hit" over the next year. The screen's top-20 hit at 55% — a 1.66× lift — declining smoothly down the ranking. The score genuinely concentrates winners at the top.

Crucially, the names hold up. The screen's high-scoring calls that paid off include real breakouts and real sales:

Player	Club (24/25)	Pos, age	Fee	Value +
Abdoul Ouattara	Strasbourg	RB, 18	—	+1150%
Rodrigo Zalazar	SC Braga	MF, 24	€30m	+67%
Enzo Le Fée	Sunderland	MF, 24	€24m	+56%
Lazar Samardžić	Atalanta	MF, 23	€17m	—
Can Uzun	Frankfurt	AM, 19	—	+150%
Nico O'Reilly	Man City	FB, 19	—	+289%
Samuel Dahl	Benfica	LB, 21	€9m	+86%
Iván Fresneda	Sporting CP	RB, 19	—	+88%

A sample of high-ranked 2024/25 players and what happened next. Value change is Transfermarkt market value to mid-2026.

Going deeper: what drives a value rise?

The screen uses hand-picked weights. To ask the data directly, I built a second model — a gradient-boosting classifier trained on ~70,000 player-seasons from 2012–2025, predicting whether a player's value would jump ≥50% the following year, using only what was known at the time. Tested out-of-time on the most recent seasons, it scored a ROC-AUC of 0.81, with its top-decile picks rising at 44% against a 14% base.

0.81Out-of-time ROC-AUC (0.5 = coin flip)

3.1×Top-decile lift over base rate

70kPlayer-seasons, 2012–2025

Model lift by decile — **Fig 2.** Players sorted into ten buckets by model score. The rise-rate climbs steadily; the top bucket more than triples the base rate.

Feature importance chart — **Fig 3.** What the model leans on. **Age** dwarfs everything else, followed by current value level, recent momentum, then playing time and goals.

That first bar is the whole story of football's transfer market in one image. Age is destiny. Value level comes next — cheaper players have more room to grow in percentage terms — then momentum, minutes and goals. It's a striking, if slightly deflating, confirmation of the economics literature: youth is the single biggest lever on a player's price trajectory.

Age versus rise rate — **Fig 4.** The share of players whose value rises ≥50% falls sharply with age. An 18-year-old is several times more likely to "pop" than a 26-year-old with identical output.

The honest limitations

This is a real edge, not a crystal ball, and it's worth being precise about why:

Age does much of the work. Because "rise" is measured in percentages, young players have a mechanical advantage. The genuinely useful scouting question — which players rise beyond what their age predicts — is the harder one, and needs richer inputs.
The multi-season model uses basic stats only. Goals, assists and minutes are in the historical data; xG, progression and defensive volume are not. Whether advanced metrics predict value on top of age is the natural next study, and needs a multi-season FBref pull.
Market value is a crowd estimate, not ground truth, and it has momentum — so some of the model's skill is simply "the market keeps re-rating players it is already re-rating."
Coverage gaps. The Championship and Argentina are missing from the value dataset, and name-matching across sources resolves about 77% of players.

The top of the board

Here are the ten highest-scoring players the model can fully profile for 2025/26 — the sharp end of the top score decile, the group that (per the back-test) hits at roughly three times the base rate. The final column is why each one surfaced: the signals doing the lifting.

Player	Club	Value	Score	Why the model likes him
Héctor FortRight-Back · 18	ElcheESP · La Liga	€12m	0.67	elite attacking output vs positional peers; strong per-90 output on limited minutes
Sidny Lopes CabralLeft-Back · 22	EstrelaPOR · Primeira Liga	€4m	0.58	elite attacking output vs positional peers
Kian Fitz-JimCentral Midfield · 22	AjaxNED · Eredivisie	€3m	0.57	plays like a more advanced role than his label; high pressing, carrying & recovery volume
Zachary AthekameRight-Back · 20	MilanITA · Serie A	€15m	0.57	elite attacking output vs positional peers; plays like a more advanced role than his label
AdrielGoalkeeper · 24	AVS FutebolPOR · Primeira Liga	€1m	0.56	elite attacking output vs positional peers; plays like a more advanced role than his label
Rayane BounidaAttacking Midfield · 19	AjaxNED · Eredivisie	€5m	0.55	elite attacking output vs positional peers; plays like a more advanced role than his label
Riyad IdrissiLeft-Back · 20	CagliariITA · Serie A	€4m	0.54	elite attacking output vs positional peers; high pressing, carrying & recovery volume
Sem SteijnAttacking Midfield · 23	FeyenoordNED · Eredivisie	€11m	0.54	elite attacking output vs positional peers; plays like a more advanced role than his label
MaikRight-Back · 20	São PauloBRA · Serie A	€2m	0.54	elite attacking output vs positional peers; strong per-90 output on limited minutes
Simon JanssenLeft-Back · 24	ExcelsiorNED · Eredivisie	€0m	0.54	elite attacking output vs positional peers; strong per-90 output on limited minutes

Score is the composite screen (performance, intensity, misprofiling, underused), normalised within position. Table limited to players with full Transfermarkt profiles; several top-decile Championship names are omitted as that league is absent from the value dataset.

Ones to watch for 2026/27

Picked in September 2026, using the latest full season of data and the market values at the time.

So who does it like now? Running the screen over the current season and filtering for the young-and-still-cheap profile, three names stand out — chosen for a spread of position, league and risk. These are the model’s calls, not certainties: think of them as where to point a scout’s eye first.

Right-back · 18 · Elche (La Liga)€12m

Héctor Fort

The confident pick

The screen’s number one, and the profile is textbook. A Barcelona-schooled attacking full-back posting an elite output score while playing only a fraction of his side’s minutes — the classic “producing above his role” signature. Full-backs who create at 18 re-rate faster than almost any profile in football, and €12m leaves plenty of headroom.

Attacking mid · 19 · Ajax (Eredivisie)€5m

Rayane Bounida

The value play

Priced like a prospect, performing like a producer. One of the most-talked-about talents in the Ajax academy, with a profile that spans creator and second-striker roles — the positional flexibility that tends to lift value. Ajax’s well-worn pathway to big sales makes the “pop” mechanism concrete. The cheapest of the three and the highest ceiling.

Winger · 18 · Cruzeiro (Brasileirão)€10m

Keny Arroyo

The high-upside flyer

The market-inefficiency bet. An 18-year-old winger thriving in Brazil is textbook arbitrage — South American wide players routinely see their value multiply on the move to Europe. Higher variance than the other two, but he sits on the cheap side of a transfer market that systematically under-prices talent before it crosses the Atlantic.

Ranked by the open-data screen, which the back-test showed carries a real edge (1.66× at the top) — not a guarantee. Profiles are built from output plus the screen’s signal scores; richer radars follow once advanced 2025/26 data is added.

The takeaway

Free data plus transparent methods gets you a screen that beats the base rate meaningfully and a determinants model that is honest about its own limits. It won't replace a scouting department — but as a first filter over 6,700 players, it points the eye at the right names, and it can tell you why.

Methods & references

Performance data: FBref advanced per-90 stats, 2024/25, nine of the ten target leagues, collected via a browser session. Value, position and age: a Transfermarkt-derived Kaggle dataset (davidcariboo/player-scores). Players are resolved across sources by name plus season-accurate club (~77% matched). The screen standardises every metric within positional peer groups. The determinants model is a HistGradientBoostingClassifier; identity and outcome windows are strictly separated in time to prevent leakage. All figures are generated directly from the underlying data.

Bergkamp et al. (2022). How soccer scouts identify talented players. European Journal of Sport Science. doi:10.1080/17461391.2021.1916081
Poli, Besson & Ravenel (2022). Econometric approach to assessing the transfer fees and values of professional football players. Economies. doi:10.3390/economies10010004
Franceschi, Brocard, Follert & Gouguet (2023). Determinants of football players' valuation: a systematic review. Journal of Economic Surveys. doi:10.1111/joes.12552