Finally it’s NFL time again! I spent some time earlier today getting my fantasy line-ups all set and over the next week I’ll get to see what kind of silly changes Yahoo has made that I’ll have to deal with for my data scraper. But in the meantime, I’ve got some picks for the week 1 games and a plan for game predictions below.
First, the plan. In the banner above there’s a link giving a rough description of my NFL model. Over the course of that year, I decided that I liked my Luigi model better than my Mario model, so this year I’ll be sticking with Luigi. Luigi is an inherently predictive model; I take the stats accumulated by teams through week X and use them to predict the outcome of games in week X+1, not week X. But in the interest of trying to improve, and since predictions are all about emphasizing out-of-sample and avoiding possible overfitting, this season I’m going to try to put together a new model using a regularized regression. This model will tentatively be called Yoshi. Both models need at least one game to start chugging on, so these predictions will start appearing before the week 2 games start.
And now predictions for week 1. Since I can’t use my typical model, I use a different system. I’ve used it twice before, so you can check those posts for a more thorough description, but essentially I use the teams’ previous season’s results to predict the winner of each game. I just pick the winner, nothing about the spread, but you can pick the winner to cover if you have the gambling itch already. It’s a pretty simple system, but it’s 23-9 in two seasons at picking the winner and 16-13-3 against the spread. Not great, but not shabby. Here are the picks for week 1, with the home team’s probability of winning listed for each match-up:
|Green Bay||San Fran||.71|
As in previous years, I’ll compare my models’ performance to a couple of benchmarks. One will be against the odds provided by online casinos I can find; I’ll either use Bovada (formerly Bodog) or a consensus from SBR. For the more ‘mainstream’ comparison I’ll use Bill Simmons, assuming he publishes his picks somewhere again. This post has a summary of Luigi versus Bill over the past two years. And I’ll also check in on the Hilton SuperContest to see how my ‘best’ picks do. This week my top three picks would be Cleveland +8.5 (since I have it as about a toss-up), Tennessee +5.5 (since I have them as a slight favorite), and Arizona +2.5 (I have them as a favorite). Basically there seems to be a good-sized home field advantage in week 1 and these are home underdogs that the model thinks are in striking distance. The other two picks are hard to pin down exactly since I’m comparing a win probability to a point spread, but I think I would go with Carolina -2.5 (Carolina has a 61% chance of winning) and the Jets -2.5 (64% chance of winning).
Here’s to a good season of football!