At Kellogg, we learned that people in aggregate tend to be quite correct (for example, say you have a random amount of jelly beans inside a big jar. Ask people to guess the amount of beans. When you average all the guesses, it will come out quite close to the real number, even if the real number is large and random, like 1,724).
According to How Microsoft got so good at predicting who will win NFL games, Microsoft Bing is an awesome prediction guru of human intelligence, machine learning, and big data:
Bing Predicts is run by a team of about a half dozen people out of Microsoft’s Redmond, Washington headquarters. It uses machine learning and analyses big data on the web to predict the outcomes of reality TV shows, elections, sporting events, and more.
In 2014, Bing was 67% accurate predicting NFL winners.
In all, the Bing Predicts model considers hundreds of these different signals, or data points, for each event, like an election or game, Sun said.
So far this year[2015 to game three], Bing is about 60% accurate in predicting NFL matchups.
Sounds great, right? However, my first thought was, who cares about winners? I can’t bet on winners, this is why the spread exists, to create (theoretical) 50/50 bets that bookies can make stable revenue from.
My next question is, in this awesome model built from millions of dollars in labor and computing power, are the prediction results better, hopefully at a statistically significant (p = .05) level, than information I could get free from a public resource? How little can I spend to get reasonably close results to aid in my for-profit wagering?
Let’s look at Las Vegas betting spreads.
From 1989 to 2013, Las Vegas favorites were correct 66.8% of the time. With a sample size of 15 years, and looking at the chart above, I can say that Vegas is pretty good.
1 signal – Vegas odds – versus hundreds of signals – Microsoft Bing = the same result.
Great work, Microsoft.