NFL 2013 Model Summary

Welcome to 2014 everybody!  WordPress does a nice thing in sending you a yearlong summary, and apparently I had about 30,000 views this year.  I took a class on science communication and the teacher said that any blog we had would get more readers than all of our actual science articles.  I think it’s fair to say that this year alone makes that likely to be true, and thanks to everyone who comes by the site.

The new year falls at a good time this year, between the end of the NFL regular season and the playoffs, which gives me an extra incentive to look back on how Luigi, Yoshi 1, and Yoshi 2 have done in 2013.  So this is just a rundown of how the predictions turned out this year.  The numbers from previous years come from last year’s summary.

I didn’t cover the model results from week 17, so we’ll start there in each case.  Picking against the spread on every game, Luigi would have been 7-9.  But three games were close to the line and the other picks would have been 6-7.  That puts Luigi’s record this year at 122-113-5/96-87-3.  Last year was a rough year, and Luigi’s record was 102-129-9/85-92-8, and in prior years was 246-222-12/197-176-9.  So in four years of picking every game Luigi is 52.4% ignoring ties, and 51.6% in games where the prediction is more than a point away from the line (ignoring pushes).  That latter number is due pretty much entirely to last year’s poor showing, as the this year came in at 52.5% and the other two years were at 52.8%.  I’m a little surprised that the numbers are so similar for every game and the reduced set, but it could be that drawing a line at 1 point just isn’t different enough.  We’ll see when I get down to the SuperContest record.

Yoshi 1 and 2 have only been going for two seasons now, so they don’t have as long a track record.  Yoshi 1 was 7-9/6-8 in week 17 for season totals of 114-121-5/95-105-5 compared to last year’s 124-107-9/100-86-7 for a two-year total of 238-228-14/195-191-12 or 51.1%/50.5%.  Yoshi 2 went 6-10/6-8 in week 17 for season totals of 116-119-5/95-96-3 compared to last year’s 105-126-9/85-96-8 for a two-year total of 221-245-14/180-192-11 or 47.4%/48.4%.  Given that last year was a poor year all around, I don’t want to draw too many conclusions from this set of games.  But it appears that Yoshi 2 is not doing as well as Yoshi 1, and neither is doing especially well.  As a reminder, Yoshi 2 is Luigi but using a regularized regression, and Yoshi 1 is also regularized but goes week-by-week instead of using the entire data set.  I had hoped that the regularization would help the predictions, but that isn’t true so far.

This year in consensus games, where all three models agree, the spread picks were only 3-6 for a total of 66-60-1 or 52.4%.  I don’t have those summarized from previous seasons, but if memory serves they’re only a slight improvement if any over Luigi’s performance.

Luigi’s SuperContest picks were 3-2; I had mentioned potentially switching out the Dallas and KC games, and that would have made the record 2-3.  But I didn’t get to switch out the Packers when Rodgers was injured, so we’ll stick with the 3-2 and give Luigi an overall record of 49-34-2 for 50 points.  I have about 920 people who made picks throughout the whole year, and they averaged a record of about 41-41-2 for 42 points.  The winner got 57 points and they paid the top 30 spots, which ended up covering 34 people with ties, down to 51 points.  So Luigi missed the money by a game, which I think I would have made up if I adjusted some picks over the course of the season.  But still not a bad showing overall.  Luigi had 45, 46, and 50 points across the last few years and is  183-141-11 across four years for 56.5%.  That’s pretty good.  The Yoshis were 2-3 and 3-2 for a season record of 40-43-2 for Yoshi 1 and 46-38-1 for Yoshi 2.  Last year they were 44-39-2 and 43-37-5, so Yoshi 1 has a two-year record of 84-82-4 and Yoshi 2 has a two-year record of 89-75-6.  Yoshi 1 is lagging behind Luigi and Yoshi 2, but it isn’t a big difference.  Of more note, those records as percents (again ignoring pushes) are the aforementioned 56.5%, 50.6%, and 54.2%.  Luigi and Yoshi 2’s SuperContest picks are definitely better than their all-game or most-game picks, and Yoshi 1 is about the same.  So for at least those two models, there is a benefit to limiting the picks to those most different from the line.

Luigi had a terrible week on the moneyline to turn the season into a net negative.  The same was true of the Yoshis, but they had it even worse on the season as a whole.  I haven’t tried to institute any kind of cut-off on the moneyline picks like I have on the spread, so it’s possible there’s a way to clean them up.  But so far Luigi has one winning year in four, and the Yoshis have been negative in both years.  I didn’t really track this year, but I typically also look after the outright winners.  Luigi was 151-98-1 for a four-year record of 605-363-2 or about 62.5%.  Yoshi 1 was 142-107-1 for a two-year record of 290-198-2 and Yoshi 2 was 151-98-1 and 295-193-2.

The last thing to check on is Bill Simmons, who’s been having a terrible year of picks.  His week 17 pick post has his record from previous years, so I can use that instead of my own records (which disagree a little).  In week 17 he was 6-10, Luigi was 6-10, Yoshi 1 was 5-11, and Yoshi 2 was 6-10.  That puts Bill at 98-122-6 for the year compared to 130-120-4 last year (I don’t think I included any week 1 picks), and has a four-year total of 479-488-25, or a bit below .500.  Luigi ended up 114-106-6 compared to 111-125-4 and a four-year total of 490-446-26; so it has a few more wins in fewer picks.  Yoshi 1 ended up 113-107-6 with a two year total of 234-222-10 and Yoshi 2 was 107-113-6 with a two-year total of 214-242-10.  Yoshi 1 has outperformed Luigi over the last two years and is doing well enough overall, while Yoshi 2 is doing poorly.  Hard to say why, given that they’re doing similarly against the SBR lines.  It could be something silly like Yoshi 2 has done absolutely terrible in the games that happen to be right at Bill’s line when that line is off from SBR’s (like SBR has a game at -3 and Bill has -3.5, and Yoshi 2 went the wrong way every time picking in the middle).  More importantly, Luigi appears to be beating the pants off of Bill.

So in 2013 I learned (or re-learned) that picking every game, or even most games, is silly; you can do better by limiting yourself to 5 or so games a week (following the SuperContest results).  Yoshi 1 seems to be doing ok with the spread although doesn’t benefit from limiting the game choices, and the models pick better than Bill Simmons does.  The moneyline picks either need some general work or some kind of cut-off, and the same goes for the over/under (which have historically been even but were a bit above this year).

I hope everyone had a good 2013, has a great 2014, and I’m sure we’re all looking forward to starting the new year with some bowl games, playoffs, and some outdoor hockey!


This entry was posted in Uncategorized and tagged , , . Bookmark the permalink.

One Response to NFL 2013 Model Summary

  1. Pingback: NFL Week 17 Results and Season Prediction Summary | Sport Skeptic

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s