Week 15 Analysis: Top Picks 5 of 7 ATS
I can’t believe it’s week 15 in the NFL.
I spent last weekend in a yurt with two of my best buddies. It was nice to get away from everything for a few days; no computers, no sports. Just way too many alcoholic beverages, gorgeous mountains, raging campfires and general insanity. We were in the Galena Yurts near Sun Valley Idaho – one of the most picturesque places on the planet. Even in this environment my mind couldn’t help but wonder to NFL football.
The machine had been setup to do its work while we stayed in the yurt so I could come back and enter the picks Saturday night before Sunday’s contests. By the end of the adventure I couldn’t believe how anxious I was to get back to my laptop so I could review the output, post the picks, enter them in various contests and set my fantasy lineup for The League – a fairly intense fantasy football league I’ve been in for the past 8 years. NFL football has kept me in close contact with so many friends I might otherwise have fallen out of touch with. I’ve spent so many Sundays with great friends freaking out on every big play, dancing around the living room watching NFL Red Zone. Without the NFL I might have just been sitting at home designing algorithms to pick stocks or, heaven forbid, putting in overtime at my actual job.
As the end of the season approaches I can’t help but get nostalgic over the last 15 weeks … is it too early to start getting excited for next year?
Week 15 was another week of solid performance for the machine. Let’s get into it.
Overall performance – week: 8 of 15 (53%), season: 35 of 70 (50%)
Here is the performance if we ignored the absolute size of the Cover.Index (SCI):
SCI greater than 1.0 – week: 8 of 15 (53%), season: 30 of 58 (52%)
Note: all games were predicted with confidence higher than 1.0. The graphic is the same as the overall performance.
SCI greater than 2.0 – week: 7 of 13 (54%), season: 18 of 38 (47%)
Here is the performance if we restrict our selections to those games in which the SCI is greater than 2.0:
SCI greater than 3.0 – week: 6 of 12 (50%), season: 14 of 25 (56%)
Here is the performance if we restrict our selections to those games in which the SCI is greater than 3.0:
SCI greater than 4.0 – week: 5 of 9 (56%), season: 8 of 14 (57%)
Here is the performance if we restrict our selections to those games in which the SCI is greater than 4.0:
SCI greater than 5.0 – week: 5 of 7 (71%), season: 7 of 10 (70%)
Here is the performance if we restrict our selections to those games in which the SCI is greater than 5.0:
The evidence seems to be mounting that above a 3.0 threshold the algorithm has some edge, and above the 5.0 threshold it is performing exceptionally well. It will be exciting to see if this performance holds up through the end of the season – particularly when the playoff picture causes such seemingly unpredictable behavior.
Week margin of victory statistics:
Sticking with the analysis first introduced last week, below is a table detailing by how many points the games were won or lost at each threshold level. A green row indicates that, on average, winning picks had a higher margin of victory than losing picks. An entirely green chart means… well – it’s all good! See the week 14 analysis for a more detailed explanation of this measure.
Season margin of victory statistics:
It’s a bit strange to me that, generally, as the SCI threshold increases the Margin.Difference seems to decrease and even go negative for the most successful 5.0 level. Part of this might be explained away as noise that should decrease as the sample size increases. With only 10 games at the 5.0 threshold, 7 of them being picked correctly (i.e. “wins”), the Loss.Margin is being calculated only off of three games: NE’s 1 point victory over CLE when they were favored by 10 (9 point Loss.Margin), NE’s 4-point loss this week to MIA when they were favored by 2.5 (6.5 point Loss.Margin), and NO’s incredible 11 point loss to STL this week when they were favored by 5.5 (16.5 point Loss.Margin). Fortunately the Win.Margin draws from a larger pool of games. Two of them were only half-point covers, which is the largest contributor to the discrepancy. As I said in the last post – the margin of loss or victory is only an indicator that a method is on the right track; that it’s tapping into something which has meaning in the real world. In the longer term wins and losses are all that really matter.
As a parting thought I’d like to note that 2 out of 3 of the losses chalked up by the machine above this 5.0 threshold come at the hands of the gosh-darn trucker-flappin’ New England Patriots. The gambling Gods obviously have a sense of humor. Each week my dad insists on pointing out that the machine has once again picked his beloved Patriots. Unfortunately the last two weeks I’ve had to point out to him that they are also the only team that is consistently really letting us down! If you take out the NE games the 5.0 threshold is an impressive 7 for 8 (88%). See the week 12 picks and analysis for all the gory details of why I hate the NE Patriots, and why my dad and the machine are (apparently) unabashed fanboys of Young Tom and the funky bunch.
That’s all for now – look for the QB analysis to be coming out shortly and the week 16 picks this weekend.