Super Bowl 54: Looking for Value Bets

I like the Bills

Bovada betting odds

The NFL playoffs start today, and as usual there are some interesting betting propositions. Here’s the current situation on Bovada:

Table 1. Earnings from $100 bet on Bovada (1:30 PM, Jan. 4, 2020).

The Ravens are favored, which is no surprise given their NFL leading record (14–2) and their QB, presumed league MVP Lamar Jackson. The Packers are the least respected team with a bye at +900, the defending champs are +1000, and the Bills pull up the rear at +6600.

Bovada vs. FiveThirtyEight

Let’s start by comparing these odds with those from FiveThirtyEight. That site lists probabilities of winning, with a floor of <1%. To compare Bovada vs. FiveThirtyEight, I’ll convert the Bovada numbers to implied probabilities (e.g. Ravens are +215, implying probability of 100/ (100 + 215) = 0.317), and I’ll treat <1% as 0.005.

Figure 1. Probability of winning Super Bowl 54 according to Bovada and FiveThirtyEight.

FiveThirtyEight assigns a much greater probability to the Ravens, and lower probabilities for the other 11 teams. If you trust FiveThirtyEight, the only good bet on Bovada is on the Ravens.

This is sort of a pointless exercise because we’re comparing two black-box algorithms. Bovada has their system to assign odds, and FiveThirtyEight has theirs. Well, there is one relevant distinction: Bovada’s odds are affected by betting action, while FiveThirtyEight’s presumably aren’t.

Looking at team metrics to find value

I’m sure Bovada’s algorithm is sophisticated and takes into account all sorts of metrics, momentum, expert sentiment, etc. I’m not delusional enough to think that I can build a better algorithm. But I think it’s fun and interesting to look at some simple team statistics alongside the Bovada odds, to see if any teams jump out as being potentially undervalued.

Pro Football Reference has fairly comprehensive offensive and defensive statistics for each NFL team in the 2019 season, so I loaded that data into R along with the Bovada odds.

Average yards and points differential, all 32 teams

These are two very simple team metrics. Average yards differential is calculated simply as the total offensive yards gained minus total defensive yards given up over the course of the season, divided by 16.

Figure 2. Average yards and points differential per game for each NFL team during the 2019 season. Green = made playoffs, red = missed playoffs.

First off, too bad for the Cowboys: #2 in yards differential and #6 in points differential, but missed the playoffs.

Of the playoff teams, I’d say the Patriots and 49ers look pretty good here relative to their Bovada odds, and the Packers and Texans look bad. The 49ers and Chiefs are both +400, but the 49ers have better statistics here. The Patriots (+1000) look much better than the Packers (+900), but they don’t have a bye so I guess that could explain their lower odds.

Now let’s look at the Bovada-assigned log-odds of winning Super Bowl 54 vs. these metrics:

Figure 3. Bovada log-odds of winning Super Bowl 54 vs. average yards and points differential. Green = first-round bye, red = no first-round bye.

Interesting. Naturally, Bovada gives a lot of weight to a first-round bye. I don’t see anything too significant among the teams with a bye; all are pretty close to the regression line.

For those without a bye, the Saints seem to have an unusually high log-odds. This could be because Drew Brees missed 6 (almost 7) games of the season, resulting in team statistics that underestimate the Saints’ true capabilities.

I like the Bills here. Bovada’s +6600 rating seems too high given their impressive yards per game differential.

A simple model

Following the results of Figure 3, I fit a linear regression model for Bovada’s log-odds of winning Super Bowl 54 vs. four predictors: first-round bye status, number of wins, yards differential, and points differential. Here’s the fitted model:

Table 2. Fitted linear regression of Bovada log-odds of winning Super Bowl (R-squared = 0.92).

This regression model explains an impressive 92% of the variability in Bovada’s log-odds for the 12 teams. It doesn’t give much weight to yards or points differential, which is interesting.

One way to spot potential value bets is to look for large residuals, i.e. teams with an unusually low log-odds given their values for these 4 predictors.

Figure 4. Bovada log-odds of winning Super Bowl 54 vs. log-odds predicted by model in Table 2.

We can also plot the residuals in decreasing order:

Figure 5. Residuals from the fitted regression model in Table 2.

For a good value bet, we’re looking for big negative residuals, i.e. teams for which Bovada’s assigned log-odds of winning are much lower than the log-odds predicted by our model. Taking this approach, the Bills are by far the best value bet, followed by the Packers and Vikings.

Go Bills!