NFL

# NFL Betting: How Predictive Are Preseason Win Totals and Super Bowl Odds?

Logic indicates that preseason sportsbook odds are pretty efficient, but how do they look when put to the test?

Look, I'm going to level with you on this one.

I want to make better NFL season win simulations, and to do that, I figure I need to test a few things. I've looked at how much early-round NFL Draft picks actually impact team offenses year-over-year, and now, I want to dig into the accuracy of preseason win total and Super Bowl odds.

Are teams with win totals around 7.5 generally ending up there, or are bookmakers way off? Intuitively, we know that the answer is that they aren't that far off -- else books would get peppered on these every year.

But what does a deep dive uncover?

### The Process

I pulled historical pre-season Super Bowl odds and win total over/unders from 2012 through 2020 via Pro-Football-Reference and compared those results to a few things.

That would include actual win totals, expected point differentials, and even playoff makes and misses.

Rather than be looking to point out the biggest hits or misses, I'm primarily aiming to see what -- if anything -- jumps out with how accurate these things are.

### Win/Loss Totals

I figure this is a fine place to start from a high-level view.

If you take each team's win total in this 9-year sample (which is 288 team seasons) and compare it to their preseason win total, the average differential is -0.20. That's pretty damn close.

But allow me to clarify: the -0.20 number is derived from a straightforward average.

That means that if Team A finishes two wins above their projected total and Team B finishes two below, the average is zero games off the pace. So...let's look at it another way.

Let's use some absolute values. Using absolute values, each of those teams from the example above would still be two games off their preseason total, but the average differential would be 2.0 games and not 0.0.

Using those absolute values of the win differentials across the full sample, we see a discrepancy of 2.19 games between preseason win totals and actual win totals.

That means that, on average, teams are finishing more than 2.0 games off of their projected win total pace. So for a projected 8-win team, they could get to 6 or 10 wins, for example.

(The standard deviation of the absolute differentials is 1.54 games.)

While this seems well within the expected range of outcomes, that's a pretty wide range on an eight-win team from the preseason. Especially if we're using this to project things like schedule strength and team efficiency. A 10-win team is not a 6-win team in this regard.

### Expected Win/Loss Totals

Now, you can say whatever you want about stats such as expected wins based on point differential.

You don't get your bets paid out based on expected wins, and I understand that, but if the goal is to see how predictive preseason numbers are, then I want to look at expected wins.

Using each team's point differential, I calculated their Pythagorean wins.

In doing this, the average differential of preseason win totals and Pythagorean wins remained close to what we saw with raw wins: -0.21.

But if we look at the absolute value differential between preseason win totals and Pythagorean wins, it reduces to 1.69. That's a change of a full half of a game when compared to a team's outright win total.

Put another way: bookmakers are better at predicting expected wins than they are at predicting actual wins.

What really stands out is the sheer number of teams that wind up a lot closer to their prediction if we use expected wins than actual wins.

Preseason Win Totals
vs. Performance
Avg.
Win
Differential
Avg.
Absolute
Win
Differential
Teams Off
by 1.5+
Games
Teams Off
by 2.5+
Games
Teams Off
by 3.5+
Games
vs. Actual Wins -0.19 2.19 66.7% 44.1% 26.0%
vs. Pythagorean Wins -0.21 1.69 49.7% 24.3% 10.1%

Almost across the board (excepting the probably useless average win differential), we see the effectiveness of preseason win totals fare better when compared to expected performance than actual performance.

That is basically another way of stating that the NFL is pretty random and that better teams don't always win their games.

For my purposes of simulating seasons and anticipating team efficiency, it's nice to know that -- while predicting wins is hard -- predicting team efficiency and expected wins is pretty significantly easier (though still a little elusive, of course).

### Over/Under Buckets

I don't want to go too far down the rabbit hole on which levels of over/unders are most predictive, but I'll at least touch on it briefly. Removing the 14 teams that pushed on their win totals, we still have 274 team seasons to look at.

For readability, I'll bucket squads into percentiles. Just know that the sample average does total 8.17 projected wins, more than the 8.00 that we'd see over an NFL season (with no ties). Naturally, the under is more likely to hit for that reason alone. That's not an error. Books can make win totals higher or lower than anticipated and adjust juice accordingly.

Win Total Over Rate
By Team Win Projection
Teams Overs Over %
9.0 or Greater 93 42 45.2%
Between 7.5 and 8.5 105 53 50.5%
7.0 or Fewer 76 34 44.7%
All 274 129 47.1%

The high-end and low-end win projections seem to be the least likely to hit (on the over, that is), and the middle-tier teams are probably your best place to be looking for overs.

I think this makes sense, given the assumed inflation on last year's good teams (who statistically will regress to some degree and also face tougher schedules by virtue of their success the prior season).

The opposite would apply for good teams that struggled in the prior year to more middling records and are treated as 7.5-to 8.5-win teams entering the season.

I don't want to get ahead of myself, but this breathes some caution against an over-reliance on last year's results to predict the upcoming season.

### Super Bowl Odds

Super Bowl odds will be tougher to compare because, well, only one team wins the Super Bowl.

But a few things do jump out: the past nine Super Bowl winners all had a win/loss over/under set at 8.5 or higher, the average being 10.11 overall. That could at least be a starting point despite being a bit obvious.

The lone long shot to win that actually won was the Philadelphia Eagles in 2017. They were +4000 to win the Super Bowl, making them the only team longer than +1600 to win in this sample. That season, they ranked 13th in preseason Super Bowl win odds. The eight other winners were all sixth or better.

From this perspective, the odds are very accurate at identifying the eventual Super Bowl champions -- or at least the teams that have a real shot at it.

Chalk is chalk for a reason.

### Predicting Playoff Performance

I don't have historical playoff odds, but we can work backwards by looking at teams that did make the playoffs and see their preseason win projections and Super Bowl odds.

It's possible that bookmakers can narrow down the championship-caliber teams but are less successful in predicting which teams make the playoffs.

Averages By Playoff/
Non-Playoff Teams
Average
Preseason
Win Projection
Average
Super Bowl
Win Odds
Missed Playoffs 7.66 +7,924

A few other quick notes on this:
- All nine Super Bowl favorites made the playoffs.
- But only 56.0% of teams ranked second through sixth in Super Bowl odds made the playoffs, kind of a low number for the chalkiest teams.
- 10.8% of teams with bottom-six preseason Super Bowl odds made the playoffs, so we generally know who is bad.

This also is pretty clear if we use preseason win projections:

Playoff Make Percentage
by Preseason
Win Projection
Playoffs
Playoffs
11 or Greater 14 14 100%
10.5 22 14 64%
10 17 6 35%
9.5 20 10 50%
9 25 9 36%
8.5 49 28 57%
8 27 8 30%
7.5 32 8 25%
7 23 5 22%
6.5 18 3 17%
6 17 3 18%
5.5 13 1 8%
5 or Under 11 1 9%

What I think is most important is that these are logical but not linear, which is to be expected.

As there's a lot of information here already I don't want to get too far into which types of teams typically hit the over or under (which will be a separate piece), so I think we can leave it at that.

### Conclusions

Overall, preseason win totals and Super Bowl odds seem to do a pretty solid job of predicting how teams will fare, which any logical person could've seen coming.

What's a little more impactful is that while, yes, teams will play a few wins off their preseason projection on average, it's less drastic when we look at expected wins.

That should remind us of the volatility inherent in football and actually closing out some games over the course of the season.

My hypothesis is that teams with turnover or scoring luck will be most likely to hit their unders the next season, but I'll be digging into that more later this offseason.

I think the grandest, most sweepingest takeaway of all is that while these numbers do a good job of getting us started anticipating the season, we can still find ways to improve upon them and bet them. And we should certainly not copy-and-paste last year's overachieving teams into the top of the league for a second straight year.