After finishing up showing value in unders in windy games, I wanted to get more out of the weather data by looking at precipitation. Everyone likes a snow game — it’s pretty on HD TV, and it tends to create slightly wacky things. Find me someone who didn’t like that Buffalo and Indy snow game last year. You can’t.
Rain is more “eh.” It’s ugly, it’s sloppy. At its worst, teams completely abandon any identity (except stubborn-ass Brian Kelly in a hurricane) and you’re reduced to watching 22 brown uniforms slip and slide in a mud field for four hours. It eliminates talent gaps and style differences. It kind of sucks.
If it’s going to suck to watch, we might as well see if we can get some value on the line.
- Does precipitation depress scoring? Of course it does.
- Do sportsbooks account for the depressed scoring? Yes. I won’t list it here, but when comparing scoring models that don’t include weather versus books’ total lines the answer is clear.
Now, do the sportsbooks adjust the line enough? Well, maybe not. Of the 7,300 games I have data for, only 702 of them (9.6%) had any precipitation (almost all of which is rain). The table below indicates that maybe bookmakers aren’t moving total lines enough.
Rain vs. the spread
Instead of using the binary response of over or under, let’s look at how total precipitation affects the difference between actual points and the total line (negative values mean the total was under).
(If you’re curious about those three wettest games, they were Louisville at Southern Miss in 2012, Notre Dame at NC State in 2016, and Kansas State at Baylor in 2010.)
Okay. Certainly some value here, but let’s not go running to the books yet. The linear model (blue line) above shows about one point of value for every quarter-inch of precipitation during the game (with lots of noise). However, as with windy games, there are some caveats.
One caveat is accurate weather forecasts. Another is that, if it rains hard enough, there might be some lightning involved, and the game will be suspended until the lightning is out of the area. My analysis is looking at precipitation during the game; often, waiting out the lightning also involves waiting out the rain.
I have another qualm as well. The American Meteorological Society defines “heavy” rain between 0.3 and two inches per hour (the only thing higher is “violent” rainfall). A single inch of rain during a college football game would be heavy rain for 3-4 hours straight. There are only 46 games in the data (0.6%) with more than an inch of precipitation.
I’m typically against removing naturally-occurring “outliers,” but in this instance I’m looking for consistent betting value. By all means, if a game is for sure to be played in a hurricane, then go bet that under before it’s off the board. But in those other cases, where we just have “regular” rain, what happens?
Below I re-graphed and modeled, excluding games with more than an inch of rain.
LOTS of rain vs. the spread
|Less than 1"||359-286-11||55.7%|
|More than 1"||31-13-2||70.5%|
Well, my qualm is unfounded. We’re seeing similar results when looking at all games under an inch of precipitation. The percentage of unders is the statistically the same, and there’s even more value looking at the actual point differential from total line with about two points of value per quarter inch of precipitation.
To wrap all this up: on games with any level of precipitation the total is under 55-56% of the time, with a point or two of value against the line per quarter inch of precipitation.
There are an average of 54 games per season with precipitation. If you took the under on all of those, you’d be expected to hit 30-23-1. Pretty good for just opening a weather app on Saturday morning.