MEAN REVERSION COEFFICIENT
−0.42
Per unit of prior season points %
Key Findings
Two questions. Updated through 2026. Here is what the data shows.
1,648 UFA movement events across 284 NHL team-seasons, 2018–2026. Three research questions examined through regression analysis and descriptive comparison, with COVID-season sensitivity checks.
Does prior season performance predict the following season's change?
The research question comes from behavioral economics work on whether performance naturally corrects after extreme outcomes — independent of what a team does in the offseason. Applied to NHL free agency: is a team's offseason activity actually changing their trajectory, or are they riding a wave they cannot control?
Loading finding...
Mean reversion result
STATISTICAL SIGNIFICANCE
p < 0.001
Confirmed across all four model specifications
STANDARD ERROR
0.052
Tight confidence interval
MODELS CONFIRMED IN
4 of 4
Full sample and COVID-restricted
A coefficient of −0.42 means that for every 10 percentage points a team finished above average last season, the model predicts they will fall back approximately 4.2 percentage points this season — regardless of what they spent in free agency. This is mean reversion, not regression to mediocrity: exceptional teams get worse, struggling teams get better, and the offseason activity in between does not reliably change that trajectory.
Mean reversion across performance quartiles
Loading example...
| Prior season finish | Points % range | Avg change next season | Recent examples |
|---|---|---|---|
| Loading quartile data... | |||
Loading interpretation...
Does it matter if a team signs a player from a rival or from across the league?
A natural question for any front office: should you target players from other parts of the league rather than your own backyard? The idea is that a fresh environment - new opponents, new travel, new systems - might help a player perform better than re-treading the same matchups. We tested whether signings from different parts of the league produce different outcomes for the team that signed them.
Loading finding...
Performance change by player origin
Mean season-over-season points percentage change attributed to each signing, grouped by where the player came from.
| Player came from | Avg next-season change | Typical season-to-season variation | Signings analyzed | Best result in |
|---|---|---|---|---|
| Loading geography data... | ||||
The practical implication: when a team chases a player from a different conference, they should not expect a bigger improvement than if they had signed someone closer to home. The novelty of a new environment - different opponents, different cities, different systems - does not translate into a measurable team-level edge. Individual players may benefit from the fresh start, but at the team level the effect does not show up.
Each signing is matched to the team's overall performance change that season - the same method the NFL site uses. Variation shows how much individual team-seasons in each group differ from the group average. The fact that the variation in each group is much larger than the spread between groups is why geography is not a reliable predictor on its own. Signings where the previous team is unknown are excluded.
Does offseason UFA spending predict team improvement?
The question every front office faces every July: is spending big in free agency worth it? This analysis compares offseason UFA investment — measured two ways — against actual season-over-season performance change for all 32 NHL teams across the full 2018-2026 outcome window.
Loading finding...
Does spending on the right positions matter more than spending in total?
We tested two ways of measuring offseason spending. The first simply adds up every dollar a team committed to UFAs. The second weights those dollars by position - a dollar spent on a goalie counts more than a dollar spent on a fourth-line winger, reflecting how much each position is generally believed to affect team performance.
If position matters as much as front offices say it does, the position-weighted measure should predict team improvement better than raw dollars. We tested both, with and without the COVID-affected 2020 and 2021 seasons.
| Approach | Seasons included | What it measures | Prediction strength | Statistical confidence |
|---|---|---|---|---|
| Loading model comparison... | ||||
Prediction strength shows how much of the team-to-team variation in performance change the approach can explain. The remaining variation comes from factors not in the model.
Loading verdict...
Does any of this predict team improvement?
After the launch, three additional things were tested to see if they predict whether a team gets better next season:
- Net spending - what a team added in free agency minus what they lost. The Toronto Maple Leafs raised this question after they lost Mitch Marner's 102 points and never matched that loss in their replacements.
- Contract length - whether teams that lock in long-term deals fare differently than teams that prefer flexibility.
- Retention - whether teams that re-sign their own players fare differently than teams that let them walk.
Loading verdict...
| What was tested | Did it predict team improvement? |
|---|---|
| Loading results... | |
Loading interpretation...
Loading model robustness note...
What the null result does and does not mean
Offseason spending is not the primary driver of whether an NHL team improves next season. The strongest predictor is where they started. Teams and analysts who attribute a team's improvement primarily to their offseason moves should consider how much of that improvement would have happened regardless.
This analysis measures team-level aggregate spending against team-level performance change. It does not measure whether individual signings were good value, whether specific players improved after moving teams, or whether the composition of a roster matters independent of spending. Those are separate research questions.
This study covers UFA signings only — players with full market choice. RFAs, trades, draft picks, and development are all excluded from Version 1. A team that built through the draft rather than free agency may improve substantially despite a low MIS score. The model measures one input to roster construction, not the whole picture.
COVID sensitivity check
The 2020 season was suspended in March and completed in a bubble format. The 2021 season used temporary geographic divisions and a shortened schedule. Both seasons are flagged in the dataset and excluded from the restricted sample models.
Result: both findings — mean reversion dominance and spending null result — hold in the COVID-restricted sample. The results are not driven by the unusual competitive environment of those two seasons.
MEAN REVERSION IN RESTRICTED SAMPLE
Confirmed
SPENDING NULL IN RESTRICTED SAMPLE
Confirmed
Data and methodology
- Data sources: Spotrac (contract and transaction data), NHL API via nhlscraper (team performance data)
- Analysis window: 2018–2026 outcome seasons (2016 baseline out of scope)
- Dependent variable: season-over-season change in points percentage
- Primary independent variables: total AAV spent, weighted MIS
- Controls: prior season points percentage, season type, cap ceiling change
- Models: OLS regression, four specifications
- COVID seasons: flagged and excluded in sensitivity check
- RFA signings: excluded by scope — UFA only
- Minor league contracts: excluded from dataset