NBA 2H Totals Model Testing Issues
Anybody else run into an issue with the 2011-2012 NBA season? I am testing the 11-12 season and my results seem off compared to other seasons. Using data from 08-09 and earlier. Each season after 08-09 is showing promising results except 11-12. It just could have been one of those years, but I'm thinking the shortened season may have also affected it. Thoughts?
Never worked with NBA data before; where did you get it from? If you're feeling generous enough to share, I'd be happy to play around with it.
Matt
The short NBA 11/12 season put all models out of whack. I would not use it for either training or out of sample testing. Same goes with the strike affected NFL 11/12 season. The lack of pre-season changed the nature of the game and the predictive capacity of variables.
I'm not sold on this, but I'm intrigued. Hope someone is willing to post some relevant data; I'd love to dig into this more.Originally Posted by brettd
The short NBA 11/12 season put all models out of whack. I would not use it for either training or out of sample testing. Same goes with the strike affected NFL 11/12 season. The lack of pre-season changed the nature of the game and the predictive capacity of variables.
2nd halves last season were all out of whack because of the shortened season. No consistency whatsoever. I would just throw out that season all together from your model. 2H totals are my specialty and even I gave up on NBA last season after 4 or 5 weeks.
Words of wisdom...Unless you are looking for "blind" systems/trends involving 2nd halves. I wouldn't even waste your time modeling 2nd half totals. The only data that actually matters is the data from the 1st half of that specific game.
You're absolutely right. My database has 1st half stats and it's what i feed my model. Good to know that my 11-12 results match yours. Would you agree your results are more consistent this season to years prior to 11-12?2nd halves last season were all out of whack because of the shortened season. No consistency whatsoever. I would just throw out that season all together from your model. 2H totals are my specialty and even I gave up on NBA last season after 4 or 5 weeks.
Words of wisdom...Unless you are looking for "blind" systems/trends involving 2nd halves. I wouldn't even waste your time modeling 2nd half totals. The only data that actually matters is the data from the 1st half of that specific game.
Consider a game with a full game total of 200 -- if in the 1H both teams never miss a shot and the total is 140 going into halftime, you'd think O120 is a lock?The only data that actually matters is the data from the 1st half of that specific game.
Don't listen to these guys calling last year an anomaly, they are just using it as an excuse for why they were losing. Notice that they can't provide any specific evidence for why the shortened season affected their predictive method, just that last year "wasn't consistent" whatever that means.
Most likely they followed some ridiculous system and of course it did not continue out of sample.
This year is my first realtime attempt at NBA 2nd half totals so I won't claim anything. It's going well enough to continue for the season.
You should run some full season, basic stats and make your own decision about last year.
I compared numbers for 4 years combined, 07-08, 08-09, 09-10 and 10-11 vs. one year, 11-12. I looked at full game average score and full game median score for the two data sets. 11-12 is about 7 points less for both average and median. Interestingly the open and close total lines were also about 7 points less.
I don't know whether this means anything or not. It may be just random year to year variation. I chose to ignore last year.
Last year was a compressed schedule with more games in fewer days. It is reasonable that the players didn't play as hard sometimes. Maybe they were tired or were conserving energy for the more densely packed upcoming games and scoring was lower.
That is a fact but depending on how you model it, may or may not matter. Do you have data on closers? I bet they were lower as well, which suggests someone out there knows what they're doing.
Math made a good point. It does depend how you model it. Looking at the averages of the variables of my model and how they differ between my data set I used and 2011-2012....it was strikingly different. The rest of my testing data was closer to my training data.
I've run into something similar in 2005 MLB data. Run totals are completely out of whack as compared to the previous 8 years, as well as the following 8. It's the only year my model gives a negative ROI, and I have no good explanation for why it happens. The fact it was the Nats first season doesn't seem plausible, and the steroid era should have no effect, given how I model outcomes. But we're talking a whopping discrepancy, to the tune of 80+ units compared to the other seasons. Does anyone else know of major changes instituted in 05, aside from harsher drug testing and suspensions?
That is a dumb example, for one. And even if that hypothetical did occur, 2H totals are mostly generalized and almost never adjusted more than a few points off what they would were pre-game in correlation to the game total, so the 2H total wouldn't open any higher than 102, tops. Not penetrating 120.Consider a game with a full game total of 200 -- if in the 1H both teams never miss a shot and the total is 140 going into halftime, you'd think O120 is a lock?
There were a ton of in-game inconsistencies last season (and last season only) in pace, shooting, everything that never developed into any predictable pattern. That was noticeable right away to any experienced 2H total bettor. Who cares why exactly the shortened season had that significant of an affect on the gameplay. It did.
Don't listen to these guys calling last year an anomaly, they are just using it as an excuse for why they were losing. Notice that they can't provide any specific evidence for why the shortened season affected their predictive method, just that last year "wasn't consistent" whatever that means.
Most likely they followed some ridiculous system and of course it did not continue out of sample.
Monitor-Tan gave 70kgman 908 Betpoint(s) for this post.
Average & median game total, open total line and closing total line for 2011-2012 were all about 7 points lower than the previous years. Game totals were lower and open and closing total lines were lower by about the same amount.
In other words, the open and close totals were what they should have been, relative to the game totals, regardless of why the game totals were lower.
It certainly looks like whoever was setting the lines knew what they were doing. Maybe the linesmaker was carefully watching the early games and decided the totals were going to be lower for the year and started making adjustments.
I get a headache thinking about the possibilities and maybes. I just run the numbers and see what happens.
This is why no one should pay any attention to this poster. There is a reason the line isn't going to be any higher than 102, regardless of what happens in the 1H. You think they've been doing it that horribly wrong all these years? Or perhaps the (obvious) answer is that what happens in the 1H is not quite as important as you think.That is a dumb example, for one. And even if that hypothetical did occur, 2H totals are mostly generalized and almost never adjusted more than a few points off what they would were pre-game in correlation to the game total, so the 2H total wouldn't open any higher than 102, tops. Not penetrating 120.
Think what you want. All my 2H total plays (NCAA + NBA) the last 3 seasons have been posted publicly, and have won at a rate well over 60% combined last 3 years. I would like you to find someone better.This is why no one should pay any attention to this poster. There is a reason the line isn't going to be any higher than 102, regardless of what happens in the 1H. You think they've been doing it that horribly wrong all these years? Or perhaps the (obvious) answer is that what happens in the 1H is not quite as important as you think.
