Week 17: Spread Picks, Results and Analysis
You should always get nervous when you start betting anything with confidence. Gambling is, and always will be, gambling – there are no sure bets. At best we can shift the probabilities in our favor – but even if we shift them 99% in our favor there’s still the 1% chance that could wipe us out. So when I entered this week with an almost smug sense of confidence in the abilities of the algorithm to pick games against the spread I knew it had to be bad news. You can’t beat the odds makers week in and week out … it’s just not possible…
Or is it?! For the last 16 regular season games the machine only seemed to get better than it’s already solid performance! The picks this week were a great way to end the regular season and I can’t help but be excited for the potential next year. For now I’ll go over week 17’s performance as usual. I will be posting playoff picks as well moving forward, with the strong caveat that the playoff picks are not backtested; it’s not necessarily the case that the same math-a-magics that seem to predict regular season games will carry over into the playoffs. Finally – I plan to perform a more thorough statistical analysis of the picks sometime soon.
Without further ado – let’s get into it!
Overall performance – week: 11 of 16 (69%), season: 56 of 102 (55%)
Here is the performance if we ignored the absolute size of the Cover.Index (SCI), which is a proxy for the confidence of the pick, and simply picked every game:
SCI greater than 1.0 – week: 11 of 15 (73%), season: 50 of 86 (58%)
Here is the performance if we restrict our selections to those games in which the SCI is greater than 1.0:
SCI greater than 2.0 – week: 9 of 11 (82%), season: 33 of 58 (57%)
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: 9 of 11 (82%), season: 28 of 44 (64%)
NOTE: Graphic is the same as the SCI greater than 3.0 graphic; there were no games with a cover index between 2.0 and 3.0
SCI greater than 4.0 – week: 9 of 10 (90%), season: 21 of 31 (68%)
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: 9 of 10 (90%), season: 20 of 26 (78%)
NOTE: Graphic is the same as the SCI greater than 4.0 graphic; there were no games with a cover index between 4.0 and 5.0
If last week was encouraging, this week was shocking. To stick with our randomness-sanity check from when this thing started, the odds of flipping 9 or more heads in 10 tries using a fair coin (which are the odds of reproducing this kind of performance by picking games at random) are roughly 1%. That really is starting to be convincing.
The numbers for the whole season are even more impressive due to the larger sample size. The odds of flipping 20 or more heads in 26 tries are less than one half of one percent – 0.46%. Encouraging indeed. I think a 20 wins, 4 losses run against the spread qualifies as exceptional in any handicapper’s book.
Week margin of victory statistics:
Even with the higher volume of wins this week, the Win.Margin was greater than the Loss.Margin in every threshold category.
Season margin of victory statistics:
Again, the margin of victory trend for this week is just an extension of the season long trend. The highest SCI.Threshold picks retained their positive Margin.Difference – but just barely. The fact that this threshold category has the narrowest Margin.Difference makes sense – you don’t go 9 for 10 without winning some very close contests. The fact that the Margin.Difference is positive at all is pretty impressive.
One note – it’s been brought to my attention that some of the spreads are a bit off from what other people have been able to get at sports books. Most of them are within a normal tolerance of the kind of changes we expect from day to day; a half point here, a half point there. But some of the spreads this week, particularly the SF vs ARI game and the GB vs CHI game had very different spreads than what was available at most books. My apologies if this caused anyone issues betting the games last week – if you notice this kind of discrepancy the best policy is to avoid betting that game for the week. When I must pick games for which I’ve seen this type of spread disparity I typically take the SCI to be the amount of points the algorithm predicts the team to cover by, and then do the math to figure out which team is predicted to cover given the other spread. I’ll talk to my data provider to see if these issues can be resolved moving forward.
Keep an eye out for a more comprehensive analysis of the season long statistics. I’ll be posting picks throughout the playoffs – but please proceed with caution – the algorithm was designed to be fully valid between week 5 and week 17 – there’s no guarantee that it performs in a similar fashion through the playoffs.