The Portable Coach: Which NFL Coaching Effects Transfer Across Rosters, and How Confidently

NFL front offices hire and fire coaches as though coaching effects are large, transferable, and
known. We show they are none of those things uniformly: coaching authority is role-specific,
resolution-specific, and only partly portable across rosters - and exactly one locus is forecastable
precisely enough to price. We apply a single partition (the head-coach / offensive-coordinator /
defensive-coordinator staff triple at each team-season) at four resolutions under one signal-to-
noise discipline (B = 1000 permutation null, cross-season holdout, propensity-matched event de-
sign) to nflverse play-by-play 2002–2024. Offense is two-locus: head coaches author macro philos-
ophy (mid-season-firing pass-rate ratio 1.36×, p = 0.002; off-season pace ratio 1.17×, p = 0.021,
while offensive coordinators author schematic fine-grain off-season red-zone shotgun-rate ratio
1.34×, p = 0.0002). The decision-relevant quantity is the predictive-validation slope: how much
of a coordinator’s prior-tenure schematic profile reappears at a new team. Under a hierar-
chical random-coefficient model the offensive-coordinator slope is 0.469 (95% CI [0.32, 0.62]),
robust to team-quality adjustment, and broad - six distinct concepts port at slopes 0.41–0.50.
The head-coach slope is 0.279 (95% CI [−0.01, 0.56]), bracketing zero: head-coach portabil-
ity is real but lives in which coach, not a population constant. Under proper role isolation,
defensive-coordinator changes move macro tendencies acutely in the modern era (opponent plays
6.37×, p = 0.007). The same framework applied to the NBA surfaces a different surviving locus
(head-coach perimeter defense), demonstrating the method transfers while the sport-specific
finding does not. We give the front-office reading throughout and report, rather than hide, the
production deployment’s honest detection limits.