Market · Arbitrage
What the market pays, against what it buys in wins
For each public stat a GM sees in the platform season: the premium the market pays per standard deviation, in cap share, beside the wins that stat is worth in the first season of the deal. Where the two disagree the market is mispricing it, if the test below holds.
- Out-of-sample test
- passed
- Partial rank correlation
- 0.207
- Proof deals
- 228
Per standard deviation of each stat
Market premium against wins
Market fit explains 72% of cap share; the win fit 31% of first-season WAR. Ranges are 95% (two standard errors).
| Stat | One SD | Market pays | Range | Wins (WAR) | Range |
|---|---|---|---|---|---|
| points per game | 0.299 | +18.7 to +56.5% | −0.00 to +0.95 | ||
| goals per game | 0.164 | −13.5 to +14.0% | −0.50 to +0.46 | ||
| ice time per game (min) | 3.1 | +5.0 to +31.6% | −0.59 to +0.20 | ||
| plus-minus per game | 0.216 | −10.6 to +0.2% | −0.15 to +0.24 | ||
| penalty minutes per game | 0.252 | −4.8 to +5.6% | −0.23 to +0.13 | ||
| shots per game | 0.789 | −14.2 to +5.2% | −0.23 to +0.47 | ||
| games played, share of schedule | 0.110 | −5.3 to +5.3% | −0.21 to +0.16 | ||
| age | 2.9 | −15.3 to +0.4% | −0.22 to +0.37 | ||
| age squared (centred) | 13.2 | −8.7 to +1.4% | −0.19 to +0.17 | ||
| defenceman | 0.454 | −9.7 to +16.7% | −0.26 to +0.63 | ||
| UFA at signing | 0.499 | −4.8 to +12.4% | −0.40 to +0.18 |
Each row holds the other stats fixed. Correlated stats (points, shots, ice time) share credit, so a single row is read with its range.
By signing year
What the market paid for, summer by summer
The market fit refitted on each first season with 40 or more deals.
| First season | Deals | points per game | ice time per game (min) | shots per game | age | defenceman | UFA at signing |
|---|---|---|---|---|---|---|---|
| 2023-24 | 50 | +64.8% ±23 | +2.6% ±18 | −5.0% ±14 | −6.9% ±11 | +17.4% ±20 | +3.5% ±12 |
| 2024-25 | 69 | +18.0% ±20 | +44.5% ±14 | +10.3% ±15 | −14.9% ±14 | −13.0% ±15 | +16.1% ±15 |
| 2025-26 | 159 | +8.9% ±17 | +56.8% ±9 | +22.0% ±12 | +0.9% ±11 | −19.6% ±11 | +1.0% ±11 |
The test
The table carries information the price does not
Weights frozen on 2022-23 and 2023-24 deals, tested on 228 deals with first seasons 2024-25 and 2025-26: stat-predicted WAR against realised WAR with the cap share held fixed. Partial rank correlation 0.207, one-sided permutation p 0.0010 (register p5-10-arbitrage-v2, Benjamini-Hochberg across the register).