In celebration of the Chiefs covering last week, in a game I never would have bet without my "mid-season coach firing" angle, I have decided to start an exciting series where I explore some of the common betting angles that I see around Twitter and tout sites these days. We won't be looking at any angles based on advanced stats or sophisticated models here - those are for nerds and losers. Just simple, straightforward angles that anyone can use, without handicapping at all.
It goes without saying that I think most angles/trends are pretty much total garbage. In most cases, the angle is so far-fetched that it is not worth investigating. And when the angle does have value, usually it is a symptom of something that fundamental handicapping can better measure, not a cause. Still, there are many angles that are at least plausible, and a few that definitely should affect the betting lines. We will stick to these high-potential angles in this series, and attempt to answer the following questions:
- Does this angle impact team performance, above and beyond a very basic NFL stat model?
- Was this angle reflected in past betting lines?
- Would betting based on this angle have resulted in a profit in the past?
We start with the idea of "revenge". We define revenge as an in-season rematch, where one team lost their previous game to their current opponent. Note that these are always divisional games, as we are excluding playoff games here. The thinking behind the angle is that teams are somehow more "fired-up" given that they lost the last game, making them more likely to cover in the next meeting. We look at this revenge angle in three ways: whether or not the team beat the other the game before, whether they covered the spread in the game before, and their cover differential in the previous game.
We start with the easiest test of all: how much (and in what direction) should we weigh the result of the first game? Looking at the second matchup from the perspective of the home team, we find that since 2007, home teams that lost the first game between the two teams won by an average of 0.84 points per game, while home teams that won the first game won by an average of 4.46 points per game. In other words, rather than a "revenge" effect, the team that won the first game between the two teams is, as we would expect, a better team on average.
Of course, the "revenge" idea is that teams will somehow out-perform their typical talent level. Perhaps if there had been no revenge, the teams that won the first leg would have won by more, as the road team would not have had the "power of revenge" that allowed them to keep it close. To measure this, we use a simple model to measure the talent level of the teams. We include two variables in this simple model. The first is the point differential per game of each team up to the point that the game took place, including last season's point differential weighted at 10% and four games of the mean value of 0. And the second variable is yards per pass attempt to-date of the quarterback who started the game for each team, including last season's yards-per-attempt weighted at 50%, and 80 pass attempts of 5 yards per attempt (this is an intentionally low level designed to penalize backups).
While a simple model, if there are any large revenge effects, this model should show them, especially since revenge games occur late in the season when average past scoring margin isn't horrible at measuring team strength:
We see that after adjusting for the strength of the teams and starting QB, the team that won the last game can be expected to perform a little worse, which does support the angle. However, the impact is only -1.28 points per game and is not statistically significant (42% of the time this could have occurred due to chance).
The same model can also be run based on the home team's win margin in the last game between the two teams, as well as the home team's cover differential. Neither of these showed any influence at all on game outcomes.
We now turn to whether bettors weigh revenge in their estimation of the betting line. We limit our sample to games since 2012 where we have good line data, and run the same model as above, only this time, instead of modeling the home team's win margin, we model the number of points by which the Pinnacle Sports closing line favored the home team:
The point spread places absolutely no weight at all on the "revenge factor". I was a little surprised by this as I would have expected some inefficiency here as perhaps bettors would lean towards teams that won the last encounter. Once again, there was no improvement in testing for home win margin, or home cover differential in the previous game.
We can now test whether betting the revenge angle would have won in past games. Since 2012, we find that a wagering strategy of "taking the revenge", where one bets the home team if the home team lost the last game between the two teams, and the away team if the home team won the last game, would have shown a small profit of 6.35 units over a 297-game sample against Pinnacle closing lines, while "fading the revenge" would have lost 19.96 units.
Looking deeper into the past numbers, home teams coming off a loss to the other team were favored by only 0.61 points per game and covered by an average of 1.26 points, while teams coming off a win were favored by 4.96 points per game and covered by an average of 0.16 points. It may well be that instead of measuring revenge, we simply have uncovered a strategy that bets more home underdogs. While these aren't as good as they once were as the market is more efficient these days, home underdogs are probably still a slightly better bet than road favorites. More likely though, the "success" of the system was just due to variance.
Overall there just aren't enough revenge games to be statistically sure if this angle is valid, but it at least would have won in the past. I certainly would never place any wagers based on this angle, as the effect isn't strong in any case, and my prior expectation is that professional football players wouldn't be extra-hyped up for a game just because they won or lost the last time. But maybe there is something to revenge after all.
Verdict: Probably fiction, but inconclusive
There is another interesting revenge system, which is based on looking at the total, not the point spread. In games where the home team won the last game between the two teams, the total points scored by both teams in the following game is 2.8 points lower than that of all other games, and this is a statistically significant impact, with only about a half percent probability of occurring due to chance alone. If the home team lost the last game, the total points scored is also lower, but only by 0.3 points. This angle is not incorporated in the market's estimation of the game totals at all.
Betting the under in all games where the home team won the first matchup would have profited 7.78 units over 141 games. But betting the under in all rematch games, not just games where the home team won the first leg, would have profited 12.35 units over 297 games. The average betting total in both types of rematch situations was the same, at 44.9 points - for comparison, the average total in all non-rematch games was 45.2 points.
My first instinct was that the lower point production in rematch games is probably due to some other factor, such as colder or windier weather, or the fact that meaningless week 17 games, which may be lower scoring, are always a rematch. However, the effect persisted even after I controlled for these factors.
While we could go even crazier and look at other variables, I can get behind the idea that the defense is advantaged in rematch games, as before I even looked at how teams did against past over/unders, I would have guessed that the experience of having prepared for the offense once before would favor the defense. In fact, that was why I bothered to look at totals for this specific angle. And I have read anecdotes of teams performing much better against quarterbacks who they have played several times in the past. However, it doesn't make much sense to me that the effect would be far stronger just because the home team won.
If I had to guess, I would say that the rematch effect is probably real, and something I am likely to at least consider when betting totals in the future, although I would never use it as the sole basis for a wager. But I would look to target the under on all rematches, not just rematches where the home team is coming off a win.
Verdict: Rematch under effect is probably fact, but sketchy