Attempting to model NBA over/unders

[Buried_PIRATE]

Right now I'm using orating / drating / a proxy parameter for tempo (which I think needs improvement) and a variable for home court advantage (found this to have only a smallish effect +/- 2 points maybe).

I ran a regression using essentially the kitchen sink of possible variables and these were the ones I found that were pretty significant on a teams score.

I re-ran the regression using these core variables and took the coefficients and weight them accordingly for each match up. For orating/drating I have the ability to use a lot of "different" versions of these variables ie, BTB / over the last 5 games / year-to-date, etc. Right now I've been purely using YTD in the predictions that I have been posting. I have been toying around with using more "specific" inputs with some success I think.

I kind of am waiting to see when this model will actually fail. Right now, if you say picking "over/unders" is distributed like a binomial variable, and the probability of success is 50% the odds of getting AT LEAST 21 correct out of 32 is 5.5%

I would like to think that, that is something to be happy about, but right now I need a larger sample size.

[Buried_PIRATE]

Don't expect it to fail try to prevent it from failing by keeping up on the as much as possible. You won't have winning nights every night but if you can have more winning than losing your doing well. Good job.

Very good advice... (considering what I wrote RIGHT below you)

I think I am more motivated to try to improve my model

Can you elaborate on what you mean by filters as well?

[ManBearPig]

What I mean is if you develop a set of filters and apply them correctly you would hope to weed out the weaker plays and try and match the strongest ones. Being a programmer I'm always looking for as many cases as I can to weed out certain situations. For an 11-game slate do you really want to play all 11 games? I'd rather weed out the games that have the lower probability of hitting and try and find the stronger ones.

One filter I apply is I avoid games where my line is too close to the actual game line. To me I don't feel comfortable picking a game that is .4 or 1 pts higher or lower because that in reality could go either way. I feel that the more above or below the line I predict the better chance I have of hitting so I look for an overlay of at least 3 pts...although I am still playing around with this.

For example for Miami tonight I had a predicted line of 181.7 for a spread of 188. This would indicated that we have a good chance of going under, which it did by 10.

Like I said I'm still playing with it and haven't settled on this but that's just one that I user. I also use PR to predict the pace by looking at things like does the stronger team score more pts at home or away. There are tons more I'm sure but the beauty of this is you can do whatever you want.

[3-ball]

hi mate! just want to know if you can post your predictions early cause i really want to tail them! thanks again and goodluck!

[pattymayo]

hey guys-- I have virtually no background in statistics or anything like that but I love the ideas of modeling NBA, especially for totals. Could you reccommend me some beginner readings to get a better feel for the calculations that you use?

[monkeyking]

Interesting stuff guys...I'm wondering what your takes on garbage time/end of game fouling is. Obviously this kinda stuff is largely unpredictable, but those possessions still get jumbled into those pace/efficiency numbers along w/ the "real" data. Personally I like to stay away from unders for low spreads (4 or less) unless theres a big discrepancy because of those late game fouls and such. First half unders are also good for such scenarios. Overs are similarly less abundant in blowouts because you'll have bench players chucking up shots and the winning team running out the clock.

[Buried_PIRATE]

@Patty: I'll try to find some stuff and post it tonight

@Monkeyking: I agree, that late game component is pretty damn unpredictable. I think the HOU@DEN game should have gone under last night, but hit the over by +0.5. Sadly, I think that things like that happen often.. and I hope that a lot of the information I capture will try to point me toward that right side

Carrying a 21-11 record into tonight.

MIL @ MIA / Prediction: 184.5 / Line: 182.5 / Play: Over
SAS @ NYK / Prediction: 204 / Line: 209 / Play: Under
TOR @ CHI / Prediction: 191 / Line: 199 / Play: Under
OKC @ MEM / Prediction: 192 / Line: 199 / Play: Under
ATL @ SAC / Prediction: 188 / Line: 189 / Play: Under (corrected)
DET @ LAL / Prediction: 197 / Line: 194 / Play: Over

POR@DAL / Prediction: 191 / Line: 185 / Play: Over (Added now that line is up)

I will breakdown my model statistics a bit later today for the first 32 games I predicted on and see if I can find any patterns so far in the predictions.

[monkeyking]

If you do some kind of point differential analysis in that breakdown, it should be more accurate in projecting your edge rather than straight win/loss record. That way, stuff like those half-point wins and losses matter less in the long run.

[slimpickins]

please confirm your numbers on ATL / SAC ; see above

[pattymayo]

yeah you predicted 188, line is 189 and you selected Over?

[Buried_PIRATE]

Ill check it when I get home I know I make some mistakes when I transfer it over.

I was crunching some summary statistics. My model is 13-2 when I predicted the score within +/- 3 to 8 points, 4-6 when less than a 3 pt margin and 1-3 when over 8.

I still thank sample size is important but that 3-8 pt range is encouraging. Let's see how that range holds up tonight

[ManBearPig]

Another thing to consider, filter out games with high totals...say >210. I find it being harder to predict over's than under's anyways, and with a large line you're model will almost always come up with a number lower that will suggest the under. The amount of games in that range are so few it's not really important to track them anyways.

I mean do you want to try and predict that a game will have 230 total points? There's no way I would've got the game going over last night's HOU/DEN game in any scenario ( I predicted around 211) and although it barely went over, it still counts as a loss. I checked and my record in those high spread games is bad so why even bother. Those games are more the exception than the rule anyways and it's much easier to predict games that fall around the magic number 199, which is considered a key number for totals. On average you'll see more totals around that number anyways so you might as well put your focus there.

[Dark Horse]

Ill check it when I get home I know I make some mistakes when I transfer it over.

I was crunching some summary statistics. My model is 13-2 when I predicted the score within +/- 3 to 8 points, 4-6 when less than a 3 pt margin and 1-3 when over 8.

I still thank sample size is important but that 3-8 pt range is encouraging. Let's see how that range holds up tonight

Good initial model. I think you're right that there will be a correction. I'm getting similar results with another model, and they weren't as good before you started posting...

If you're not doing so already, you may find it useful to factor in the league average.

[bigjah15]

tailing you on all your picks. small bets. just wanna test the waters bol

[ManBearPig]

Well for the plays that I'm playing today it looks like we're on the same side so hopefully we're on the right side. I'm not on the DAL game but I think that game may go under from the numbers I'm seeing...it's too close to call and I'm staying away but I got this at 177 and 182 for the two lines I gen. Line is dropping so of course this means it will go over.

BOL tonight hoping for more W's than L's.

[Buried_PIRATE]

MIL @ MIA / Prediction: 184.5 / Line: 182.5 / Play: Over
TOTAL: 190
GRADE: WIN
SAS @ NYK / Prediction: 204 / Line: 209 / Play: Under
TOTAL: 243 (lol...)
GRADE: LOSS
TOR @ CHI / Prediction: 191 / Line: 199 / Play: Under
TOTAL: 202
GRADE: LOSS
OKC @ MEM / Prediction: 192 / Line: 199 / Play: Under
TOTAL: 215
GRADE: LOSS
ATL @ SAC / Prediction: 188 / Line: 189 / Play: Under (corrected)
TOTAL: 210
GRADE: LOSS
DET @ LAL / Prediction: 197 / Line: 194 / Play: Over
TOTAL: 191
GRADE: LOSS

I took out the POR@DAL I didn't realize that Butler got hurt and why the line was delayed

Tough today... bumps me to 22-16

I'm going to have some time tomorrow to tinker a bit with the model... disappointing day though

[RonDon]

Bad Day

I picked a bad night to follow your model...it is really interesting though. I will keep watching. GL.

[Buried_PIRATE]

I picked a bad night to follow your model...it is really interesting though. I will keep watching. GL.

Sorry man. I am very disappointed in the 1-5 day. I didn't have a lot of time to work on improving the model today, so instead I'm going to compare teams past 5 game predictions vs my full season predictions. Either way, carrying a 22-16 record into 11 games tonight.

TOR@CLE / Line: 205
5day: 185 / Under
YTD: 198.5 / Under

CHI@NJN / Line: 185
5day: 186 / Over
YTD: 187 / Over

MIL@ORL / Line: 187
5day: 178 / Under
YTD: 185 / Under

WAS@PHI / Line: 188
5day: 190 / Over
YTD: 192 / Over

SAS@BOS / Line: 191
5day: 180 / Under
YTD: 189 / Under

CHA@MIN / Line: 203
5day: 196 /Under
YTD: 195 / Under

GSW@NOH / Line: 198.5
5day: 193 / Under
YTD: 195 / Under

POR@HOU / Line: 197
5day: 197 / Push
YTD: 201 / Over

ATL@UTA / Line: 190
5day: 199 / Over
YTD: 197 / Over

LAL@PHX / Line: 209
5day: 193 / Under (LAL clearly slumping here...)
YTD: 206 / Under

DEN@LAC / Line: 211
5day: 196 / Under
YTD: 198 / Under


Interestingly... not too much discrepancy in the YTD / 5day decisions... here is hoping to a better day today. BOL if you happened to end up on one side or the other of these picks

Regardless, I am going to post here for the long haul... hopefully we keep making some money

[Buried_PIRATE]

TOR@CLE / Line: 205
5day: 185 / Under
YTD: 198.5 / Under
TOTAL: 225
GRADE: LOSS

CHI@NJN / Line: 185
5day: 186 / Over
YTD: 187 / Over
TOTAL: 190
GRADE: WIN

MIL@ORL / Line: 187
5day: 178 / Under
YTD: 185 / Under
TOTAL: 184
GRADE: WIN

WAS@PHI / Line: 188
5day: 190 / Over
YTD: 192 / Over
TOTAL: 206
GRADE: WIN

SAS@BOS / Line: 191
5day: 180 / Under
YTD: 189 / Under
TOTAL: 209
GRADE: LOSS

CHA@MIN / Line: 203
5day: 196 /Under
YTD: 195 / Under
TOTAL: 213 (in OT)
GRADE: LOSS

GSW@NOH / Line: 198.5
5day: 193 / Under
YTD: 195 / Under
TOTAL: 213
GRADE: LOSS

POR@HOU / Line: 197
5day: 197 / Push
YTD: 201 / Over
TOTAL: 203
GRADE: WIN

ATL@UTA / Line: 190
5day: 199 / Over
YTD: 197 / Over
TOTAL: > 201
GRADE: WIN

LAL@PHX / Line: 209
5day: 193 / Under (LAL clearly slumping here...)
YTD: 206 / Under
TOTAL: ~194
GRADE: WIN

DEN@LAC / Line: 211
5day: 196 / Under
YTD: 198 / Under
TOTAL: 199
GRADE: WIN

Ended today 7-4. After going 1-5 yesterday I won't complain that one game that went into OT was a lock for a win ...

Either way, going into today 22-16, adding 7 and 4 brings us to 29-20.

[ManBearPig]

That CHA game pisses me off...should've been an easy W...the way it rolls sometimes.

[Buried_PIRATE]

That CHA game pisses me off...should've been an easy W...the way it rolls sometimes.

Yup. Would've liked to have banked that win. Either way today was encouraging after getting trounced...

The day ended pretty sweet tho:
LAL@PHX / Line: 209
5day: 193 / Under (LAL clearly slumping here...)
TOTAL: 194
GRADE: WIN

DEN@LAC / Line: 211
YTD: 198 / Under
TOTAL: 199
GRADE: WIN

[ManBearPig]

Ya, you almost hit the Clip game on the button I had it at 202 so I was close...I was stupid though and bet the over for some reason so even though I won I lost cuz I failed to make the correct bet...no wonder I'm still a peon bettor.

Can't complain though cuz out my 9 plays I hit 6 of them including that stupid OT game. I need to work on getting the MM because it seems like I should be up more. If you can have more days like this than yesterday you will be doing well for yourself. Good job today

[Buried_PIRATE]

Only one play tonight, I don't know how to predict Dallas yet with 2 of their best players injured. I'll monitor the situation though...

DEN@SAC / LINE: 207
5day: 198 / Under
YTD: 189 / Under

29-20 going into 1 game today

EDIT:

I added the 5 day for Dallas..

DAL@OKC / Line: 192.5
5day: 185.2 / Under
YTD(with Dirk and Butler in the line up): 196 / Over

^^ clearly do not know how to adjust for injured players yet :X

[ugabooga]

On this too. BOL.

[Buried_PIRATE]

den@sac / line: 207
5day: 198 / under
ytd: 189 / under
total: 224
grade: LOSS

29-21 since 12/28

[pattymayo]

Edit: was looking at the wrong thing. Nothing to see here...

[ManBearPig]

Don't take this one hard because although I had the over the way it worked out was pretty bizarre. I wouldn't worry about accounting for injured players because the line should already reflect it. Dirk and Butler didn't just go out tonight so there were no adjustments that needed to be made. Under was clearly the play here and I just got lucky it hit.

[Buried_PIRATE]

Don't take this one hard because although I had the over the way it worked out was pretty bizarre. I wouldn't worry about accounting for injured players because the line should already reflect it. Dirk and Butler didn't just go out tonight so there were no adjustments that needed to be made. Under was clearly the play here and I just got lucky it hit.

Lmao yeah. I was watching the game on gamecast and thought the under had hit, and then I was like WTF?

Fortunately I didn't wager any money on the game and I'm glad that it worked out for you. I wasn't confident predicting that game because I wasn't sure how the injuries would impact the result.

The game that really surprised me was the Kings lol...

Either way, another day tomorrow with more games. Looking forward to it

[Buried_PIRATE]

I'm going to try to rank some of the games today. I will still post the predictions for all the games, but I am trying to get a better feel of where I feel like my predictions are stronger than others. I'll post the plays I think are better than average first, then the rest.

Better than avg (ranked best to worst)
MIA@MIL / Line: 183
5Day: 190.5 / Over
YTD: 187 / Over

TOR@BOS / Line: 199
5Day: 178 / Under
YTD: 195 / Under

---- guess I'll see what happens with those ... I'll still try to rank these, but these would be more leans

UTA@MEM / Line: 195
5Day: 197.5 / Over
YTD: 196.5 / Over

POR@MIN / Line 200
5Day: 193.7 / Under
YTD: 198.6 / Under (Going under here... tighter than I would like)

SAS@IND / Line: 200.5
5Day: 178.1 / Under
YTD: 192 / Under

NOH@LAL / Line: 190
5Day: 185.5 / Under
YTD: 193.5 / Over (Taking the Over here)

CHI@PHI / Line 191
5Day: 190.3 / Under
YTD: 188.3 / Under

HOU@ORL / Line 207
5Day: 197 / Under
YTD: 198 / Under

CLE@GSW / Line 209
5Day: 198 / Under
YTD: 199 / Under

NYK@PHX / Line 218.5 (My model is 0-4 predicting lines above 215... so as MBP says in the post below OVER or no play might be the best play here)
5Day: 202.3 / Under
YTD: 210 / Under (No play here)

NJN@WAS / Line 189 (No lean / hunch on this game... )
5Day: 187 / Under
YTD: 191 / Over

[ManBearPig]

Don't try and model games that high because you can't. You'll always have an average a lot lower than the total. There is a trend going back to 2006 or 07 where when two teams from the opposite conference have a line of 220+ they cover around 65% of the time. This year there have been two games that fit this trend. NY/GS and NY/DEN those games had lines of 221 and each covered going for a total of 239 and 254, you just can't predict scores that high.

This line is currently at 219.5 for so it has a chance. I think it's over or no play here. Remember SAN got in a track meet with NY the other night and they have traditionally been a team that doesn't like to get up and down the floor that much. PHO, although isn't as pure of a running team these days, no one will be mistaking them for a defensive minded team anytime soon.

[Buried_PIRATE]

MIA@MIL / Line: 183
5Day: 190.5 / Over
YTD: 187 / Over
TOTAL: 196
GRADE: WIN

TOR@BOS / Line: 199
5Day: 178 / Under
YTD: 195 / Under
TOTA: 224
GRADE: LOSS

UTA@MEM / Line: 195
5Day: 197.5 / Over
YTD: 196.5 / Over
TOTAL: 209
GRADE: WIN

POR@MIN / Line 200
5Day: 193.7 / Under
YTD: 198.6 / Under
TOTAL: 206
GRADE: LOSS

SAS@IND / Line: 200.5
5Day: 178.1 / Under
YTD: 192 / Under
TOTAL: 177
GRADE: WIN

NOH@LAL / Line: 190
5Day: 185.5 / Under
YTD: 193.5 / Over (Taking the Over here)
TOTAL: 198
GRADE: WIN

CHI@PHI / Line 191
5Day: 190.3 / Under
YTD: 188.3 / Under
TOTAL: 204
GRADE: LOSS

HOU@ORL / Line 207
5Day: 197 / Under
YTD: 198 / Under
TOTAL: 205
GRADE: WIN

CLE@GSW / Line 209
5Day: 198 / Under
YTD: 199 / Under
TOTAL: 214
GRADE: LOSS

I'm most surprised about TOR@BOS... I'm not too surprised losing CLE@GSW (I do worse when the line is around 210+ and CHI@PHI had a very tight line around where I predicted the score.

MIA@MIL, which I felt strongly would go over, would have ended at 182 if not for OT.

5-4... goes to 34-25

[Buried_PIRATE]

34-25 since 12/28...

I like these first 3 more than the others...

IND@ATL / LINE: 192.5
5DAY: 183 / Under
YTD: 189 / Under

MIL@NJN / LINE 178.5
5DAY: 183 / Over
YTD: 185 / Over

PHI@DET / Line: 193
5Day: 203 / Over
YTD: 196 / Over
-----
BOS@CHI / Line: 188
5DAY: 178 / Under
YTD: 185 / Under

ORL@DAL / Line: 187
5DAY: 182 / Under
YTD: 191 / Over (***Took Over here***)

WAS@CHAR / Line: 187
5DAY: 184 / Under
YTD: 191 / Over (***Took Over***)

MEM@OKC / Line: 203
5DAY: 187 / Under
YTD: 196 / Under

UTA@HOU / Line: 207.5 [***NO PLAY***]
5Day: 208 / Over
YTD: 203 / Under

[jeepsguy004]

Hey guys I just wanted to jump in on this conversation. I use a model pretty similiar to what you guys are doing to make my own lines on the game. I do this for College Basketball and NBA. Now these two are a little different. In College because their is so many teams I do factor in SOS and PR. In the NBA I do use PR. I use Home and away numbers in both of these models due to the drastic difference for some teams from home and away. Just a few things I would recommend is actually looking at what the public is doing only because you can get more value in your play if they are betting the opposite and the line is going in your favor. I also stay away from those games that have totals over 210. I usually narrow my plays down to about 3 per day even less. In the NBA alot of games are about situations that teams are in. I am nto saying to look at trends but what the situation is. Example. The under is 5-2 in Hornets games this year after allowing 100 plus points. So if you have a game where your model says under in a hornets game and they are in that situation you should feel a little bit more confident. These are just things that I do when I am making my picks. Hope maybe they can help some guys out there.

[Buried_PIRATE]

Hey jeeps thanks for the post.

I definitely have started to take some of your advice and I can tell now, that's pretty good. I'm pretty new at this (only been trying for about 10 days) and I think I am starting to be able to see which games are in my comfort zone. Some good ideas there

Either way, I have been posting all my results just in case it might help someone .. but in the future as I get a bit better at this, I will start to cut down.

[jeepsguy004]

Buried.....

For being someone that just started out you are on the right page with this kind of model. As you get further in this you will see what games to eliminate and what games to look further into. I have been at this about 8 years now and I have learned so much in those 8 years and I am still learning. This is one business that you continue to learn in. When you start to narrow games down I know this might sound stupid but you really want to look at games with popular teams involved. NBA (Boston, Lakers, Orlando, Miami, Dallas,) teams like this see alot of public money bet on them due to their name and how good they are. But the thing is I do not care how good any team is I am more worried about if they are going to cover or not. In College you want to look at games where teams in the top 25 are invloved due to public money that comes in on them and you get value in these type of games. Some handicappers feel the books slack a little bit when it comes to games in college that involve smaller schools due to them being more focused on the big schools but with those big schools come great value due to the public money like I said.