> libero

A One-Year Value Target Read Three Quarters of the Decline After 28, Every Season

Defined as the last valuation 180 to 365 days after a player's last match, a one-year value target had a median horizon of 225 days. It read average growth as −2.9% where a year gives −7.5%, and in each of four seasons it read 66% to 80% of the decline of every band over 28. The rows it dropped, the obvious suspect, account for 0.3 points.

Oct 20264 min read

Libero's trajectory model predicts the change in a player's market value over the year after a season. Under that window, 10.4% of player-seasons had no valuation and were dropped, which made survivorship the first worry: players with no update inside a year might be the ones falling out of view. The valuations themselves say otherwise.

Valuations arrive in two waves, and the window ends between them

Days from a season's last match to the player's first valuation after day 180.

Figure 1: 74% of player-seasons get their next valuation 180 to 240 days out, the winter update. The summer update lands at 360 to 420 days, two thirds of it after day 390. The old window takes the winter one for most players and loses the rest; a window of 365 ± 120 days catches the summer one.

Horizon, Not Survivorship

Under the old rule the median horizon was 225 days, and 62% of targets were read before day 270. Taking the valuation nearest to day 365 within ±120 days moves the median to 382 and drops 2.0% of rows. Of the rows the old rule dropped, 87% did have a later valuation; their update simply came after day 365.

On the rows both targets keep, average growth goes from −2.7% to −7.2%. Adding back the dropped rows takes it to −7.5%. The dropped rows do skew old (21% of over-34 seasons against 8% of under-21s), which is why they average −10.5%, but they are a tenth of the data, so nearly all of the bias comes from the horizon.

A seven-month horizon understates the decline after 28 by a fifth to a third

Value change by age band on the same player-seasons, under each target.

Last valuation 180 to 365 days outNearest to 365 days, ±120

Figure 2: Same rows, two targets. Every band but 25 to 27 moves away from zero: 31 to 33 year olds lose 42% over a year where the short target read 32%, and under-21s gain 83% rather than 74%.

The old target measured a different quantity from a one-year change: a mix of seven-month and eleven-month changes weighted by when each player happened to be revalued. The mix is the same every season, because the revaluation calendar is: the old target reads 66% to 80% of the one-year decline for every band over 28, in each of 21-22 to 24-25. A test season is compressed exactly like the seasons before it, so any out-of-time check scored against the same target has nothing to flag. A model trained on it learns a flattened age curve, and learns it confidently.

What is still open

The timing of a revaluation is not random: a player in the news is revalued sooner. Whether the 365 ± 120 target leaves any of that in, through which players get an off-cycle update inside the window, is not something these counts can show.