FairValue

Checking model

Esc

  • EnvironmentOverviewWhat FairValue measures, what it found, where to go next.
  • EnvironmentTerminalExplore, compare and track players; market, league and model views.
  • EnvironmentStoriesValuations, head-to-heads and research findings, told as stories.
  • TerminalPlayer ExplorerSearch, filter and sort the player export.
  • TerminalComparison StudioTwo to four players on one scale.
  • TerminalWatchlistsNamed collections, kept in this browser.
  • TerminalMarket LandscapeHow observed value is distributed, and where the model drifts.

↑ ↓ to move · Enter to open · Esc to closeLoading player index…

Valuation story

Inside a valuation: Khéphren Thuram

From the model's €2.2M starting point to €41.4M, one input at a time, and where that lands against an observed €38M.

Built from Published SHAP output for the served Model A (reports/explainability); the player was in its training data.10 chapters

Observed €38MModel A €41.4M (served, in-sample)Serie A · Central Midfield

Portrait of Khéphren Thuram

Khéphren

Thuram

CM

№ 463618World Class €34M +0.11

Close-up

Image: FootyLogos · image credits

Chapter 01: The valuation

Khéphren Thuram: 25, central midfield in Serie A. Transfermarkt's valuation is €38M, in the World Class tier. This story follows how Model A arrives at its own figure, one input at a time.

Chapter 02: Where every estimate starts

Before it looks at this player, Model A sits at €2.2M: its average output across the players it was trained on. Every player's journey starts here. Each input then multiplies the estimate up or down.

Chapter 03: Appearances

Appearances (96) raises the estimate ×1.95: from €2.2M to €4.2M.

Across all modelled players this input moves Model A's estimate by ×1.28 on average, in one direction or the other. Here it moves it more than usual.

Chapter 04: League

League (Serie A, Italy) raises the estimate ×1.77: from €4.2M to €7.5M.

Across all modelled players this input moves Model A's estimate by ×1.96 on average, in one direction or the other. Here it moves it less than usual.

Chapter 05: Club points per game

Club points per game (1.82) raises the estimate ×1.57: from €7.5M to €11.8M.

Across all modelled players this input moves Model A's estimate by ×1.38 on average, in one direction or the other. Here it moves it more than usual.

Chapter 06: Age

Age (25.2 yrs) raises the estimate ×1.42: from €11.8M to €16.8M.

Across all modelled players this input moves Model A's estimate by ×1.60 on average, in one direction or the other. Here it moves it less than usual.

Chapter 07: Minutes (last 365 days)

Minutes (last 365 days) (3,394) raises the estimate ×1.35: from €16.8M to €22.7M.

Across all modelled players this input moves Model A's estimate by ×1.19 on average, in one direction or the other. Here it moves it more than usual.

Chapter 08: Everything else

The remaining 15 inputs are individually small. Together they raise the estimate ×1.82, to €41.4M.

Chapter 09: The estimate

After all 20 inputs, Model A says €41.4M. Transfermarkt says €38M. The residual, log(observed) − log(estimate), is −0.085: the observed value is below the estimate, so the model over-predicts by ×1.09.

The served model was trained on this player, so this is an in-sample figure; on unseen players a typical miss is ×1.50.

Chapter 10: What this is not

SHAP describes the model's arithmetic for these inputs. It is not a price, a scouting judgement or a statement about what causes value, and the order of the steps is by size, not sequence.

The same record, with Model B and the player's traits, is in the dossier.

Starting point€2.2Mobserved €38M

Every Model A valuation starts at €2.2M, the model's average output over the players it was trained on (log 14.592).

raises the estimate lowers it Model A estimate observed value (dashed line)Log scale: equal lengths are equal multiplications. Inputs are ordered by size of effect, not in any causal sequence. Published SHAP run for the served Model A; this player was in its training data.