TL;DR

• An NFL team's "net rest" relative to their opponent in any given week has long been touted as a meaningful and self-evident factor influencing

• "Short week" advantage/disadvantage on a standalone basis (e.g., opponent played on Monday and their next game is Sunday) has limited signal and could be a fadable narrative for betting purposes

• Most importantly, our research suggests a more nuanced reality, where the timing, context, and clustering of net rest days matter more than their sheer quantity in any given isolated week

• The most signal for predicting point differentials comes from an NFL team’s clustered net rest relative to their opponent's over a rolling 3-week period

• Key insight: excess net rest advantages ("net-net rest") have a more consistent and deeper impact on point differentials than the more universally appreciated advantage of a bye-week

Each spring when the NFL schedule is released, one of the more popular clickbait observations by fantasy and betting analysts is the implied rest discrepancies between teams.

The concept of net rest advantages and disadvantages has long been touted as a meaningful and self-evident factor influencing relative team performance.

And this makes intuitive sense: having more time to rest, recover, practice, and gameplan than your opponent in between games and in aggregate over the course of a season should accrue some sort of material advantage (or disadvantage)…amirite?

Stress-testing net rest assumptions

Instead of relying on squishy heuristics, our study seeks to empirically test this commonly held belief and quantify the value (if any and to what extent) net rest advantages and disadvantages impact the relative future performance of NFL teams.

The results of our analysis challenge the industry’s overemphasis on simple season-long aggregate and weekly net rest and suggests a more nuanced reality, where the timing, context, and clustering of net rest days matter more than their sheer quantity in any given isolated week or in totality over any single season.

According to our research, the most signal from rest-related features for predicting point differentials comes from evaluating an NFL team’s clustered net rest relative to their opponent's over rolling 2-week, 3-week, and 4-week periods, with the most consistent signal in rolling 3-week periods.

Indeed, perhaps the most noteworthy takeaway: relative excess net rest advantages over the course of a rolling 3-week window period appears to have a more consistent and deeper impact on point differentials than the more universally appreciated advantage of a bye-week.

Let’s get into the details.

Data preparation

Our starting dataset spans NFL regular season games from 2000 to 2023, using the NFLfastR package. This includes public game- and team-level information such as schedules, scores, spreads, and week-over-week rest and net rest metrics (manually recalculated due to some errors in the raw NFLfastR dataset).

We then engineered and tested various custom rest and net rest-related features over multiple rolling window periods (being sure to avoid overlaps between seasons).

It should be noted the usual caveats associated with any sort of single-variable analysis hold. Our research provides insight into one aspect of performance, but it does not account for the multitude of other variables and complex interactions that influence NFL outcomes that could ostensibly obfuscate the actual impact of rest as a single factor, particularly non-linear relationships.

To help disentangle the effects of rest from other variables, we leveraged more sophisticated multivariate techniques using Random Forest and XGBoost machine learning models. For this analysis, we trained on data from 2000 to 2018 and tested on data from 2019 to 2023. Some examples of meaningful contextual variables we included were, among others, time zone differences, elevation differences, seasonally-adjusted local temperature differences, miles traveled, consecutive road games, and a custom modeled situational difficulty percentile feature.

Goals

We then sought to evaluate the predictive value of:

1. aggregate net rest on an overall season basis,

2. net rest on a game-by-game basis,

3. aggregate net rest over rolling n week periods, and most importantly,

4. aggregate net rest relative to their opponents’ aggregate net rest over rolling n week periods (“excess net rest” or “net-net rest”).

Impact of net rest on season performance

Our analysis suggests there is limited signal in a team's aggregate net rest predicting season-long performance in the NFL.

Season wins

Contrary to popular belief, total net rest over a season showed minimal correlation with total season wins. Counterintuitively, there’s even a slight negative correlation of total net rest to total season wins.

Season point differentials

Likewise, total net rest over a season showed minimal (again, even slightly negative) correlation with total season point differential.

Season win total futures

How about correlation to betting market's expected team performance using win total futures as a proxy? Probably not. Similar to our other observations, MAYBE there is some minimal signal at the very tail end of the net rest disadvantage spectrum, but the sample size is very small. Otherwise, zero correlation.

Season ATS

No discernible insight that aggregate season net rest advantages/disadvantages have any sort of outsized impact on against-the-spread results either. Similar to previous results, the correlation is actually negative (but really, mostly just noise).

Of course, the observations from these relatively simple analyses are not necessarily dispositive that a team's aggregate net rest has absolutely zero impact on total season wins or performance.

But what is clear is net rest’s impact (if any) is not as self-evident as many believe and discerning its impact requires a more complex approach.

Are net rest discrepancies getting worse?

Separately, there’s been accusations that the magnitude of net rest discrepancies are getting worse season-over-season due to poor scheduling by the NFL league office (ostensibly due to incompetence and/or to accommodate TV networks, etc.).

However, while the Ravens’ 16 days of net rest this season is indeed a standout, the dispersion of net rest among teams is broadly consistent with past historical precedent.

Impact of net rest on weekly performance

Similarly, our analysis suggests there is limited signal in a team's weekly net rest advantages/disadvantages predicting any single game performance in the NFL.

Distribution of outcomes

The impact of weekly net rest on team performance and single-game point differentials is dubious at best.

Particularly in the 1 to 5 days range, where rest advantages actually skew more toward negative outcomes.

In the 6+ days range (a proxy for bye weeks and bye week-like equivalents), rest advantages become more noticeably pronounced.

Average and median outcomes

Consistent with the observations above, average and median point differential for a team with a rest advantage of 1-5 days (a sample of approx. 1400 games) are decisively negative. In at least 50% of the sample, the team with positive net rest lose the game by more than 2 points with an average deficit over the entire sample of more than 1/2 a point).

This seems to imply that the "short week" advantage/disadvantage on a standalone isolated basis (e.g., opponent played on Monday and their next game is Sunday) seems to be moot and a fadable narrative for betting purposes.

The distinction, again, is with respect to a rest advantage of 6+ days (a sample of approx. 570 games).

The most frequent rest advantage is 7 days (approx. 470 games), while 6 days (approx. 50 games) and 8 days (approx. 45 games) are much smaller and therefore produce more visibly noisy results.

While obvious, it should be noted this advantage is not a repeatable phenomenon during the regular season, given a team only has one bye week. However, the corresponding disadvantage is indeed a repeatable phenomenon during the season, given a team may face several opponents coming off a bye week.

Alas, bye week advantages/disadvantages are also well known, generally appreciated, and therefore likely properly accounted for and priced into the market (perhaps even overly so).

Early season vs late season

Another common heuristic in fantasy and betting communities is rest advantages are more meaningful later in the season vs earlier in the season.

The evidence suggests that may not be the case (or at least isn't as self-evident as many assume without torturing the data).

While maybe somewhat surprising, perhaps this isn't as outlandish as it may seem on its face.

A plausible rationale is added time and rest advantages matter more early on as teams adjust to the new season and optimize gameplans, strategy, and chemistry among new players, coaches, and schemes.

The net impact of this early and mid-season fine-tuning may outweigh the diminishing returns of conducting the same exercise later in the season and/or any marginal health-related benefits of simple extended recovery time.

In fact, if we were to disaggregate how teams allocate their time during periods of rest advantages, we could theorize that earlier in the season is when teams are healthier and can more opportunistically use this time to enhance and optimize strategy and performance.

Conversely, later in the season is when teams are more injured and fatigued and are instead forced to allocate all or a majority of this time to player-level rehabilitations and less so to aggregate team-level improvements.

Rolling average net rest

The most substantial signal was found in evaluating a team’s aggregate net rest over rolling window periods, particularly 3-weeks. While any amount of net rest advantage/disadvantage within these rolling periods all yielded corresponding expected positive/negative point differentials, we wanted to sensitivity-test certain thresholds of excess net rest that amplified the signal.

To do so, we tested various excess net rest thresholds (for example, a 3-week rolling average of +/- 4 net rest days relative to their opponent). We used various machine learning techniques to attempt to disaggregate the impact of our rolling rest metrics among other variables and isolate the relative importance of each feature in a multivariate context. We also also wanted to ensure the thresholds were sensible and not overly precise or otherwise overfit to the data.

The 2-week, 3-week, and 4-week rolling periods with thresholds of approximately 3 days of rolling average excess net rest showed the most stable correlations with game outcomes and the most noticeable impact on machine learning model predictions.

Performance improved when the net rest thresholds were applied in conjunction with an opponent's rolling average net rest deficit, as our research suggested the shadow of fatigue incongruently lengthens over time. In other words, the detrimental effects of clustered rest/time disadvantages became more pronounced as the deficit compounds vs the additive effects of increasing rest/time advantages.

In practice, this means we need a defined surplus of relative net rest days to trigger a reliable advantage, but perhaps most importantly, the opponent must have at least some deficit of net rest days to trigger a reliable disadvantage. For example, a rolling 3-week net-net rest advantage of 3 days in the form of a rolling 3-week net rest of +4 days and an opponent's rolling 3-week net rest of +1 day would not yield as considerable an advantage as a rolling 3-week net rest of 0 days and an opponent's rolling 3-week net rest of -3 days.

This observation is consistent with even weekly results where we found the most consistent "meaningful" rest advantages: when a team has at least 10 days rest and the opponent has 7 days or less rest. The average point differential was +0.84 and the median was +2. Key Takeaway: the sequencing and composition of net rest advantages likely matter more than just their absolute values.

According to our research, proxies for team strength also played a meaningful role in the predictive efficacy of our custom rest metrics. There was limited value of rest advantages or disadvantages when involving the best teams or the worst teams and/or in matchups when the closing spread implied a large strength discrepancy (in other words, games with a TD spread or higher). In these matchups, the other variables in play overwhelmed and diluted any discernible impacts of rest.

Rolling 3-week excess net rest

• Teams with a rolling 3-week average of more than 3 excess net rest days (and the opponent had a rolling average of less than -3 days net rest) had an average point differential of +1.82 and a median of +3 over a sample of 186 games.

• This situation happens 15x in the 2024 season.

  

  
  

• The average spread spread differential was +1.14.

These rolling window periods and defined thresholds appear to more accurately measure the comprehensive impact over time of a team having a surplus or deficit of rest relative to opponents.

Takeaways

By avoiding a hyperfocus on any single isolated week, the rolling window period approach appears to more adeptly capture the compounding nature of wear and tear and cycle of fatigue and recovery, along with the time sensitivity of quality game-planning, adjustment, and overall strategic preparation.

All of these elements have historically been associated with the net rest concept and related metrics, but their actual impacts on performance appear to be far more relevant when net rest advantages/disadvantages are evaluated on a clustered and relative basis vs a simple standalone weekly basis.

We believe incorporating rolling net-net rest metrics into handicapping models can potentially provide a competitive edge not fully appreciated or priced-in by markets.

Rolling 2-week, 3-week, and 4-week average net rest data and charting capabilities will be available this preseason on the Sportfolio Terminal.

Thanks for reading.