NFL Playoffs — FiveThirtyEight vs. the Bookies

January 10, 2020
Posted by Jay Livingston

The 49ers are a 7-point favorite tomorrow over the Vikings. According to FiveThirtyEight’s Elo rating, they’re only five points better.

Elo — named for Arpad Elo, a physicist who used it to predict chess matches — is a “power rating,” which takes into account a team’s record, the records of the teams it played, the margin of victory, and some other team variables. Power ratings are common in sports. Elo, as you would expect from something at FiveThirtyEight, is a bit more statistically complicated. For example:

We created a multiplier that gives teams (ever-diminishing) credit for blowout wins by taking the natural logarithm of their point differential plus 1 point. This factor also carries an additional adjustment for autocorrelation, . . .the tendency of a time series to be correlated with its past and future values. [Source]

Can you use Elo to make money betting on the NFL? Is that two-point difference tomorrow enough to warrant a bet on Vikings getting seven points? Below is a chart showing wins and losses for each level of difference between the Elo spread and the actual betting line. I set the minimum difference at 1½ points.

For example, in the last week of the regular season, the bookies had Broncos as 4-point favorites over the Raiders. Elo rated them 5½ points better. So according to my system, that 1½-difference means bet the favorite. As it turned out, the Broncos won, but by only one point. So we Eloists and other Broncos bettors lost.

So far this season, there has been an Elo-vs-bookmakers difference of exactly 1½ points in 37 games. If you had bet accordingly, you would have won 19 bets and lost 18. Unfortunately, most bets require the bettor to give 11-10 odds. You bet $110 to win $100. So on these thirty-seven 1½-point games, you would have lost a little (4%).

(Click for a slightly larger view.)

As the chart shows, Elo as a betting guide does not improve as the rating differential increases. If you had bet $110 on each game where the Elo rating differed from the betting line by 1½ points or more, you would have wound up winning 70, losing 81 — a net loss of $1100. Increasing the size of your bet as the differential increased might have made a small improvement.

In any case, here’s how Elo and Las Vegas see the games this weekend.

Las Vegas Line
Elo likes the Vikings as underdogs. In the other three games, it thinks that the betting line is underestimating the strength of the favorite. That’s not an accident. As their methodological post at FiveThirtyEight says, “We found that, in the NFL playoffs, favorites tend to outplay underdogs by a wider margin than we’d expect from their regular-season ratings alone.”

UPDATE: Here are the Elo picks and their outcomes in the actual games.

1. Elo liked the Vikings + 7. They lost by 17 (27 - 10). A loss for Elo.

2. Elo liked the Ravens - 9½. They lost on the field. Badly (12-28). Nobody saw that one coming. A loss for Elo.

3. Elo liked the Chiefs - 11½. They won and covered (51-31). A win for Elo.

4. Elo liked the Packers - 4½. They won by 5 (28-23). A win for Elo.

Two up, two down. With Elo as a guide, if you had bet of $110 to win $100 on each game, your net would be -$20.

No comments: