BroBible
Sports betting is a multi-billion dollar industry—and that’s just the legal kind. Of all sports betting, football’s is the biggest. Accurately predicting which teams will win is a dark art that’s equal parts statistics analysis and magic. One diehard football fan may have cracked the code. He’s built a computer model that thus far has an 88% success rate. That’s 15 right and just two wrong. And he’s giving the predictions away for free.
Faro (@data_by_faro) has been building these models for a while. The NFL prediction model has taken his TikTok account from the fringe to the rarefied air of an influencer seemingly overnight. As of this writing, Faro’s up to roughly 120,000 followers.
Just a few weeks ago, most of his posts were getting roughly 1,000 views. Now some of his most popular ones have hundreds of thousands of views.
There are innumerable metrics that go into choosing a winning team. Some are semi-joking in nature, such as predicting the winner will be the team whose mascot would theoretically win a fight against the other’s. Choosing winners based on whose uniform you like better is another tongue-in-cheek method.
Possibly the worst but most common method is to just choose your favorite of the two teams.
Those in the know use multiple metrics to predict winners. Common ones include the yards allowed per game in running and passing plays, offensive yards per game in the same, points per game, points allowed per game, whether the team’s playing at home or away, and injuries.
Faro has created an NFL win-predicting code using what’s called predictive modeling. It uses historical data and statistics like the ones listed above to predict the outcome.
Some other examples, which he’s shown in scatter plot charts, are win percentage vs turnover margin and time of possession rate vs overall success rate running plays. He is continually tinkering with the model to improve it.
Faro says that he began by “thinking about upsets.”
“If the team with the better stats always won, there would never be an upset. So what is this underlying factor causing upsets, or set of factors?” he says he wondered.
This question led to others, which is key to his process.
“You don’t need to be great at math to build these models; you just need to ask yourself the right questions,” he says.
He’s trying to figure out how much can possibly be predicted about a game in advance.
“If we had the absolute best prediction model, how much of a game could it actually predict?” he says, then acknowledges, “It’s not going to be 100% no matter what. We want the prediction accuracy to be as close to the amount of the game that’s predictable as possible.”
His model might not be perfect, but an 88% accuracy rate, even with a dataset of only 17 games, is exceedingly rare.
It has people clamoring for...
