2015-16 Team win projections

Home for all your discussion of basketball statistical analysis.
Mike G
Posts: 4425
Joined: Fri Apr 15, 2011 12:02 am
Location: Asheville, NC

Re: 2015-16 Team win projections

Post by Mike G » Thu Feb 25, 2016 11:30 am

Talk about a worst post!

What part of the post don't you like? All of it, or just the part you quoted?
Personally, I can't imagine Kardashian tracking.

You're replying to one person's post and addressing "you guys". Is this confusion deliberate?
And I agree with Barkley at least half the time. Should we know what you're referencing?

permaximum
Posts: 413
Joined: Tue Nov 27, 2012 7:04 pm

Re: 2015-16 Team win projections

Post by permaximum » Thu Feb 25, 2016 1:40 pm

Mike G wrote:Talk about a worst post!

What part of the post don't you like? All of it, or just the part you quoted?
Personally, I can't imagine Kardashian tracking.

You're replying to one person's post and addressing "you guys". Is this confusion deliberate?
And I agree with Barkley at least half the time. Should we know what you're referencing?
1. All of it. From subjective reasoning for the shortcomings of analytical player ratings to introducing manual intervention in order to correct those issues to suggesting the isolation of regression to the mean or any other prediction component for that matter.

2. It's not a confusion. By using the post I quoted as a strong supportive argument for my developing thoughts on this matter, I decided it was the right time the state them in a simple but effective sentence.

3. I know you know what I refferred to. Still to be more clear, I meant Barkley's view on sports analytics as a whole.

If you have more questions, I can elaborate on why I started to develop these thoughts lately. Simply put, the more knowledgable I've become on this era, the more I've started to doubt if there's any use for all-in-one metrics and analytical win, score predictions for the next game or the next year.

Crow
Posts: 6208
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2015-16 Team win projections

Post by Crow » Thu Feb 25, 2016 5:55 pm

What I see is at least 7-8 of the top 10 performers on team win projection by kmedved's two scoring measures use RPM / RAPM directly or indirectly or via aggregation and I am not sure about caliban and kmedved's methods. I assume they are one metric based and would thus make it a near sweep or complete sweep of top 10 by heavily influenced by one metric entries. RAPM influenced entries were 5 of top 7 performers last season and the other two were statistical plus minus based so this isn't just a unique at the moment performance level. A 1metric influenced projection set has always won the contest, unless Vegas was an exception and I doubt it was. I see only one entry that is known to use this metric to some degree is in bottom half and it was highly adjusted in a number of novel ways. Maybe a second, I don't know what everyone used. The bottom half seems full of entries that have substantial subjective influence or are not clearly all in one metric based or at least widely known and used ones.

Average projection was off by 8 wins or more for 9 teams. That distribution doesn't surprise me. I'd like to see how cumulative performance of players changed last season to this one for those collections of players on RPM, BPM and perhaps other metrics and if the biggest misses were due to team or role changes, injuries, coaching change or quality, off-court issues or something else known or unknown. Can't really support a dismissal of one metrics without seeing this and perhaps comparing it to predictions that were not one metric based. One metrics are far from perfect but so is inferring from eye test or un-compiled and formally weighted discrete stats. Right now it appears that RAPM based approach does better / best. Using something other than an explicit one metric is essentially aggregating stuff free-form or just leaping to an intuitive answer.

permaximum
Posts: 413
Joined: Tue Nov 27, 2012 7:04 pm

Re: 2015-16 Team win projections

Post by permaximum » Thu Feb 25, 2016 9:53 pm

1. Non all-in-one metrics and stats have specific uses. Although they can be used as parts of some kind of blend/aio metric to predict next year's team wins, it's not their real usage.

2. All-in-one metrics' prediction success compared to other analytical approaches means nothing to me if they cannot improve upon fan average and coach minute distrubution in any significant way. I wanted evidence for their true superiority over both of the things I mentioned and I haven't seen any yet. On the contrary I shared one of the posters' retrodiction test here as a proof for non-empirical PER's superiority over RPM or xRAPM in a true prediction test. Considering PER's inability to capture defense, it just simply shows all-in-one metrics' bad state. Actually, no. It shows analytics' bad state at predicting the outcome of future games.

3. We have seen past prediction contests' results as RMSE. Have any of you there see any improvement?

It's not rocket science to predict teams' next year wins by an RMSE of ~5-10 when roster turnover rate for league average is somewhere around 35%. It would be rocket science if you could improve upon Vegas by such a margin that you're rich now.

kmedved
Posts: 82
Joined: Thu Jul 03, 2014 9:18 pm

Re: 2015-16 Team win projections

Post by kmedved » Fri Feb 26, 2016 12:24 am

Per that last comment, I've added a column for record vs. Vegas. For Vegas here, I used the 10/26 Westgate entry. Pushes are counted as pushes, as are times when you & Westgate picked the same win total.

Image

And yes Crow, my method is an RAPM hybrid. Differs mostly at the margins (in particular, with regression to the mean!).

Crow
Posts: 6208
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2015-16 Team win projections

Post by Crow » Fri Feb 26, 2016 1:31 am

So RAPM influenced predictions hold at least 8-9 of the top 10 right now.

A bit over half of the predictors would have had positive return if had bet with Vegas on all teams and equal amounts. Some with a pretty high rate of return, at least in theory, for this season.

bbstats
Posts: 224
Joined: Thu Apr 21, 2011 8:25 pm
Location: Boone, NC
Contact:

Re: 2015-16 Team win projections

Post by bbstats » Fri Feb 26, 2016 5:38 am

My RPM predictions vs Vegas:
2014: 20-9-1
2015: I don't know but I won the Avg Error thing at "6" so "very good"
2016 (projected): 22-8-0

Mike G
Posts: 4425
Joined: Fri Apr 15, 2011 12:02 am
Location: Asheville, NC

Re: 2015-16 Team win projections

Post by Mike G » Fri Feb 26, 2016 2:03 pm

Almost everyone here is looking better than Vegas?
Chart says I am 17-11 vs Vegas' picks. Yet Vegas ranks a couple spots ahead of me.

kmedved
Posts: 82
Joined: Thu Jul 03, 2014 9:18 pm

Re: 2015-16 Team win projections

Post by kmedved » Fri Feb 26, 2016 4:36 pm

Mike G wrote:Almost everyone here is looking better than Vegas?
As noted, there's a lot of "group-think" in these numbers, since so many people are using RAPM as the base. This isn't really a super diverse set of predictions. Also, while this is using the 10/26 Westgate numbers, I don't have the vigs for those. For some teams like the Celtics, the line was juiced like -135 or something.

I'll poke around why Vegas ranks ahead of so many teams beating it H2H however. Maybe I'm screwing something up with my code here.

permaximum
Posts: 413
Joined: Tue Nov 27, 2012 7:04 pm

Re: 2015-16 Team win projections

Post by permaximum » Fri Feb 26, 2016 5:43 pm

Go ahead gentlemen. Take the money. Noone is holding you.

I think it was about time calling it what it is.

Kevin Pelton
Site Admin
Posts: 102
Joined: Thu Apr 14, 2011 10:05 pm

Re: 2015-16 Team win projections

Post by Kevin Pelton » Fri Feb 26, 2016 10:27 pm

kmedved wrote:I'll poke around why Vegas ranks ahead of so many teams beating it H2H however. Maybe I'm screwing something up with my code here.
This year might be an outlier, but I think the explanation may be that Vegas is so well regressed that it almost never misses terribly on a team but misses slightly more on slightly more of them.

I know I've observed this trend in the past with SCHOENE, which was not a good predictor at all in terms of RMSE or even average error because of its tendency to be way off but still got more than half of the over/unders correct most years.

Dr Positivity
Posts: 298
Joined: Thu Sep 20, 2012 6:44 pm

Re: 2015-16 Team win projections

Post by Dr Positivity » Sat Feb 27, 2016 6:45 pm

Crow wrote:What I see is at least 7-8 of the top 10 performers on team win projection by kmedved's two scoring measures use RPM / RAPM directly or indirectly or via aggregation and I am not sure about caliban and kmedved's methods. I assume they are one metric based and would thus make it a near sweep or complete sweep of top 10 by heavily influenced by one metric entries. RAPM influenced entries were 5 of top 7 performers last season and the other two were statistical plus minus based so this isn't just a unique at the moment performance level. A 1metric influenced projection set has always won the contest, unless Vegas was an exception and I doubt it was. I see only one entry that is known to use this metric to some degree is in bottom half and it was highly adjusted in a number of novel ways. Maybe a second, I don't know what everyone used. The bottom half seems full of entries that have substantial subjective influence or are not clearly all in one metric based or at least widely known and used ones.

Average projection was off by 8 wins or more for 9 teams. That distribution doesn't surprise me. I'd like to see how cumulative performance of players changed last season to this one for those collections of players on RPM, BPM and perhaps other metrics and if the biggest misses were due to team or role changes, injuries, coaching change or quality, off-court issues or something else known or unknown. Can't really support a dismissal of one metrics without seeing this and perhaps comparing it to predictions that were not one metric based. One metrics are far from perfect but so is inferring from eye test or un-compiled and formally weighted discrete stats. Right now it appears that RAPM based approach does better / best. Using something other than an explicit one metric is essentially aggregating stuff free-form or just leaping to an intuitive answer.
Hard to argue against the results of RAPM and RPM, though I'd also suggest the make-up of the predictor who uses advanced +/- first methods, is likely to be different than boxscore. When it comes to factors like using previous seasons predictions as data points, prior seasons for players, aging curves, regression to the mean, etc., or how one crosses the bridge from previous season results to this year's prediction, I assume a predictor who has been knee deep in APM and SPM for years, is more likely to carry over this high skill level and "advanced model inclinations" in the other parts of their prediction. If my boxscore driven prediction does worse than RPM/RAPM predictors, how does one separate how much of that is my stat being < RPM, and how much was a skill deficiency?

kmedved
Posts: 82
Joined: Thu Jul 03, 2014 9:18 pm

Re: 2015-16 Team win projections

Post by kmedved » Sun Feb 28, 2016 3:31 pm

Kevin Pelton wrote:
kmedved wrote:I'll poke around why Vegas ranks ahead of so many teams beating it H2H however. Maybe I'm screwing something up with my code here.
This year might be an outlier, but I think the explanation may be that Vegas is so well regressed that it almost never misses terribly on a team but misses slightly more on slightly more of them.

I know I've observed this trend in the past with SCHOENE, which was not a good predictor at all in terms of RMSE or even average error because of its tendency to be way off but still got more than half of the over/unders correct most years.
This makes sense to me. I think this and the missing vig element are what's going on here. The vig piece can be a pretty big deal obviously. It can swing team win projections a couple wins either way if you convert between the odds and the projected win total.

tarrazu
Posts: 72
Joined: Mon Aug 04, 2014 5:02 pm

Re: 2015-16 Team win projections

Post by tarrazu » Sun Feb 28, 2016 5:22 pm

kmedved wrote:
Kevin Pelton wrote:
kmedved wrote:I'll poke around why Vegas ranks ahead of so many teams beating it H2H however. Maybe I'm screwing something up with my code here.
This year might be an outlier, but I think the explanation may be that Vegas is so well regressed that it almost never misses terribly on a team but misses slightly more on slightly more of them.

I know I've observed this trend in the past with SCHOENE, which was not a good predictor at all in terms of RMSE or even average error because of its tendency to be way off but still got more than half of the over/unders correct most years.
This makes sense to me. I think this and the missing vig element are what's going on here. The vig piece can be a pretty big deal obviously. It can swing team win projections a couple wins either way if you convert between the odds and the projected win total.
Yes the vig. If your model projects 56 wins for the Krazy Koalas and the Vegas line is 55.5 (over -130), there is no bet to make. What is the threshold for difference from the Vegas line for your model to suggest a bet on the over/under?

Crow
Posts: 6208
Joined: Thu Apr 14, 2011 11:10 pm

Re: 2015-16 Team win projections

Post by Crow » Sun Feb 28, 2016 7:03 pm

If you are at least 80% confident of strong results to deal with vig and unexpected risks, justified in that view, trust where that money is going, don't mind it being tied up, don't have better uses, don't have legal or moral issues, then maybe you bet these. If you feel like it. Not doing so, is not necessarily a sign of doubt or weakness.

Post Reply