Last year I spent some time posting predictions from two different models (Mario and Luigi), which were really a set of regression models trying to guess the point differential, total points, and winner for NFL games. It turned out that Mario, which used a large set of predictors, didn’t do nearly as well as Luigi, which used a subset similar to, but different from, what Brian Burke uses at Advanced NFL Stats. For a summary of how the models did, here’s the last round-up from the regular season last year. In general, they didn’t do great against Bodog.com over/unders, Luigi was a little ahead on winner moneylines, and Luigi was a little ahead on spreads but way ahead using Bill Simmons’ spreads (I still wish he would say where they come from). This year the predictions return, but I’ll only be posting Luigi’s predictions.
The over/under: The over/under is just a guess as to how many total points will be scored by both teams (you can see the link above on ‘how to bet’ for more info). As I said, the models didn’t do very well here. Last year is the first time I looked at them, so I don’t have any insight at where to place any cut-offs about when to avoid a bet or not. I’ll just be posting the straight predictions again this year.
The moneyline: The moneyline is a guess as to who will win the game outright. The team with a negative sign is the favorite; you would have to bet the number after the negative sign to win $100 if the team actually does win. The other team is the underdog and you would win that number if you bet $100 on them and they won. You can use the moneyline to figure out what probability a team would have to win to be profitable (again, check the link above for more info). I use that to make the chart below. Some games didn’t have a moneyline on Bodog yet, so they 0s below. Other times the model just won’t think that either team is getting the right odds so it won’t bet, but this week it happens to have a pick in every game. Last year the model seemed to prefer going with underdogs; it wasn’t correct particularly often, but it came out ahead on money because you win more when an underdog wins.
The spread: The spread is a guess as to how much a team will win by. For example, if a team is -3, they are predicted to win by 3. You can take the number as a guess as to what you would have to adjust the home team’s score by to get a tie game. Home teams usually win by about 3, so if two teams are equally matched the spread should be -3 to counter the home field advantage. If you saw that the home team was +3, that would mean the away team is actually about 6 points better (from the -3 home field advantage to +3). The column below with Luigi’s prediction is made as home minus away, so it’s opposite the spread (sorry for the confusion). For example, Luigi thinks the Lions will win by 20, but the spread says that they’ll win by 8.5. Thus you should take the Lions to cover the spread. In general, I would advise staying away from games where the prediction is close to the spread, like Giants-Rams or even Tennessee-Baltimore.
Bill Simmons: Bill posts his picks every week and over the last three years at least has been fairly successful. My model did much better with his spreads than the Bodog ones; I don’t know why. I recommend staying away from games where the prediction is too close to the actual spread but Bill picks every game. So it could be that Luigi did very well on games that it would have otherwise ignored, or perhaps Bill just makes up lines he thinks he can crush and they’re fundamentally easier than Bodog’s. Either way, here are his picks as well as Luigi’s against his lines.
As a note, the predictions in week 2 can be pretty wild because there’s only one week of data. For example, both Buffalo and Detroit are projected as virtual locks to win and win big; I don’t think the model ever predicts a 20 point victory by mid-season. However, I tried re-predicting last season after regressing early season data to the mean and it didn’t really improve overall performance (although I can think of some other things to do as well). Since it’s easier to not regress, that’s what I’m going to keep doing. So these predictions might be taken with a grain of salt, but I don’t think they’re necessarily worse than the predictions that happen later in the year.