# Are Moneylines Useful In Predicting NFL Games?

Today’s post is over at Edgehogs.

I wanted to know how good moneylines were at predicting wins and loses. Since Las Vegas is still in business, it was a good bet (first joke of the day!) that they were. But I wanted to quantify by how much.

All the moneylines flowing from bookies have a built-in juice or vig, in other words a transaction cost or brokerage fee. Friction, I think they call it. These are easily removed—well, easy to remove when analyzing the data. You’d have no luck removing them when making a bet (second joke of the day, even lamer than the first!).

At Edgehogs there is no juice in our X-Bucks bets. That means you could use our system to try out your secret betting scheme using virtual money. Go get ’em!

Turns out to be an easy calculation to transform (juice-free) moneylines into probabilities of teams winning. I did that for over 2,700 NFL games and looked at a calibration plot, which is the one given above. Full details about what it means are on the blog post.

A hint of how moneyline odds are related to spread odds is also covered (need I say it? ha ha!).

Click on over at see the rest.

1. DAV

I was wondering if you were aware that the juice inflates (or deflates, depending on perspective) the odds.

I suspect the spread is largely the basis for the odds and is one of the places where juice can be applied. If the odds were perfect betting would be completely pointless as it would then be a guaranteed money drain. The prospects look dismal already. The only way to beat them would be to predict the outcome more accurately.

Oddly (no pun), at the horses where the bettors set the odds, the win favorite is underbet while the underdogs are overbet. This “edge” isn’t worthwhile because it’s not large enough to overcome the track take. There may be some finagling on sports odds to encourage betting. More betting means more juice will flow.

2. mt

Am I reading the calibration graph and the red/green graph correctly? You seem to have found a significant home field disadvantage in the data, both in the odds and the results. In both graphs, the home team are expected to lose more often, and end up losing more often.

3. Briggs

mt,

Perhaps not. If I were to chose different buckets the circles would change positions, but not by much. The forecasts/moneyline predictions appear reasonably calibrated. Except for the possible quirk when the moneylines are near +100/-100 (probabilities near 50/50). I don’t think the bucketing matters here because of what we saw on the color plot: a significant dearth of probabilities right near 50/50.

Maybe DAV can help us. My only guess is that bookies are reluctant issue 50/50 bets because it doesn’t appear to the bettor that the bookie is offering any added value. If you knew nothing except that two teams were playing, that evidence gives a 50/50 probability. So there may be a temptation to move from that mark that is sometimes, but rarely, unwarranted.

4. mt

Just looking at the calibration graph, there are more games where the home team is expected to lose (probability of win < 0.5) and more games where the home team loses (proportion of wins 0.5 on both axes.

5. Briggs

mt,

Aha! Thanks! Goes to show you what cutting and pasting can do. I had the home and away inversed. I’ve fixed it. Thanks very much.

6. DAV

It may be because 50/50 mark is the hardest to predict. In info theory that’s the place with the most uncertainty. Also, as you pointed out, who wants to bet on a push? Most bettors don’t really look at the bets as my predicted probability vs. yours — at least not explicitly. It would seem a waste of time. Not that the other bets aren’t but at 50/50 most bettors would consider it obvious.