He found that after controlling for team strength, there is no effect at all of momentum carrying over. Of course, people still believe in momentum, with fans of the Steelers (4-0 in their last four games) and Vikings (win over Bears in Week 17) having high hopes for the upcoming season.
I decided to do a follow up on Barnwell’s study, but focusing on how a team’s record against the spread (ATS) at the end of one season translated into ATS performance at the beginning of the next.
To do this, I looked at each team’s last four games of the regular season, going back to the 1985 season (the earliest season for which we have data). I then calculated the percentage of times that the team covered the spread in those last four games where the game wasn’t a push. For example, the New England Patriots went 2-2 ATS in their last four games, or 50 percent, and the Indianapolis Colts went 1-2 ATS with one game that was a push, or 33 percent.
I then looked at how these teams did in their first game of the new season. Below is a table showing the relationship between a team’s ATS percentage for the final four games with those teams’ corresponding percent chance of covering the spread in Week 1, along with the number of teams in that category.
As you can see, teams who did worse against the pointspread at the end of one season did better against the spread during the first week of the next, while it was the opposite for teams who covered the spread often in the final few weeks. If we plot these points, we get a chart that looks like this:
Seeing this, the negative relationship is fairly apparent. Despite the nice line however, the relationship isn’t quite significant: running a logistic regression on whether a team will cover the spread in Week 1 (looking only at games that didn’t push) the variable containing the team’s percentage ATS during the final four weeks of the previous season has a p-value of only 0.13.
Still, this is pretty good considering there are many other differences between teams that we aren’t attempting to control for: whether a team has changed coaches, whether they had a new quarterback, whether they had a lot of roster turnover. These things would certainly affect the spread, and might alter how betters perceive a team’s momentum.
The fact that teams that had a lot of success at the end of one season do poorly at the beginning of the next is a due to regression to the mean combined with the public not understanding this. As Barnwell proved, a team that finishes strong does not perform significantly better at the start of next season – there is no evidence of momentum. But, for some teams at least, the betting public believes that the team’s form will continue into the next season, leading to the relationship that we see.
It should be emphasized again that this relationship is very slight, as expressed by the not-quite-significant p-value. To put it into terms of money, however, if you had had bet against the 54 teams that had not lost a game ATS in their final four games of the previous season you would have won 30 of those bets. If you bet $1 on each game, with 5 percent vig, you would have made a total of $4.57, for a return of a little over 8 percent. Betting on teams who had underperformed at the end of the previous season would net you only a little more, with a return of just over 12 percent.
So you shouldn’t blindly base all your bets on how a team did in the past few games of the previous season, especially when the team itself changed a lot. Then the public may no longer think they have momentum, but it has been profitable and useful to keep this information in mind.
If you were wondering which teams you want to be wary/enthusiastic about for Week 1 of this NFL season, there are three teams that were undefeated against the spread in the last four weeks of the 2014 season, and another three that didn’t cover the spread once. The teams without a win are Miami, San Francisco and Tennessee, while Pittsburgh, Seattle and Dallas were perfect ATS down the stretch. Keep an eye on them.
One last note is that one very obvious expansion of this post would be to look to see if there is this effect during the season as well. For example, if a team goes on a four game run relative to expectations (i.e. doesn’t lose against the spread once) will that team have too high of expectations in its next game? An idea for a future post, perhaps?
Harrison Chase is the Co-President of the Harvard Sports Analysis Collective, a student-run organization at Harvard College dedicated to the quantitative analysis of sports strategy and management.