Is the NBA player prop of points per game a derivative of team total? Is it profitable to take, for example:

Durant's average points per game / OKC average points per game = x

y = friday night game durant's prop line / friday night game OKC team total


test to see is y different from x by .5 points or more and bet the side which x predicts

No both the team total and the player points per game are derivatives are the game spread and total and the player points are what I call a partial derivative since it cannot be fully explained by the full game numbers whereas the team total can be.

It helps me to think of it as the "game within the game."

i wish it were that easy

my brother actually thought of that last nba season, he looked at it more as a total rebs/under total assists/over, but also total points/over. we really had no way to backtest but tracked it for a couple weeks and looked like a 50/50 lose the juice type of thing.

basically this.

it was just a drunken kicking ideas around thing, but we did track it for a while. probably have the data on my other computer.

that said, i wish my technical/math abilities matched my desire to be strong in technical/math areas.

to beat NBA player props you need to be able to project minutes to be played by the player in that particular game.

wiper why doesn't your brother post on 2+2? guy sounds like a good dude

might get the 2p2 points you just transferred me stolen.

he has an account, mostly lurks.

that's important and hard to do, esp this year. I can't tell you how many overs I was 3/4 or more of the way there by halftime, but the guy doesn't play the 4th qtr.

I will say I've had good success taking a player's averages for reb or asst and then adjusting for how many his opponent gives up. Esp successful against SB's props, but lol SB.

pretty much. but like freedom fries said you need to be able to project minutes, which really isn't that hard since coaches follow pretty straightforward patterns with their rotations and there's so much edge in prop betting you always have a huge margin for error.

you also can't take points per minute entirely at face value given lineup changes etc (i.e. if a team recently had a high usage player get hurt obviously everyone else gets a bump). but it's pretty easy to ballpark this stuff and a lot of these aren't particularly matchup dependent- i.e. derrick rose vs steve nash would be a favorable matchup for rose but the suns aren't going to let him go 1 on 1 too much. the only matchups that significantly impact totals are if a big man who gets a lot of points at the rim goes against an elite big like dwight howard.

and if you're just not good at handicapping this stuff you can always just use the pinny markets as a baseline and look for props that get way out of line, it happens pretty often.

also you don't want to pay -115 juice for durant u28.5 if EV is 28, it's slightly -EV

thanks. my brother who watches basketball suggested handicapping it on matchups but what you say makes more sense to me.

why wouldn't that durant bet be profitable? i have been basing my calculations on a poisson distribution which seems ROUGHLY reasonable. this distrubtion says it will land exactly on 28 7.5% of the time so we are gonna get 28 or under 55% of the time which would make u28.5 profitable.

thanks for the post your_mom i will see if i am smart enough to figure out how to work that.

You shouldn't use Poisson for this.

^^^^ +1

Poisson is supposed to be used for events that happen 1 at a time.

Is there a probability distribution to model non-individuals events like this, run scoring in cricket etc? Or would it have to be modelled by Monte Carlo for example?

Monte Carlo or use Poisson on something related that does happen one at time such as attempts

I've read a couple times people say poisson is bad for this. Based on what I see it is reasonable, I hope you can tell me where my thinking is wrong.

True, poisson is supposed to model events one at a time so it would be fine to use for something like assists. But for player points a person will score points either one at a time (on free throw line), two at a time, or 3 at a time with three point shots. So this will average out to about 2 at a time. Going with a poisson expectency of 14, the average outcome of this event is 10.5% so betting U 28.5 is even more profitable than I showed above. (I think the basic idea here is that since points are scored 2 at a time instead of 1 at a time, in general, the distribution will center more closely on 28.5 since there are less events that need to take place so less chance for variance)

This distribution shows that 28 points has 10% expectency while 40 points has a 3% expectency and 14 points has about 2% expectency. This might not be exactly right but seems reasonable to get a rough estimate and seems accurate enough to show that U28.5 is profitable.

Please start betting these at Pinnacle.

here's a dumb example of why poisson just doesn't work for pts. Look at Kevin Love, who avg 24.5 pts/game. Poisson says that if you set the over/under at 28.5, a fair price for the over would be +385. However, Love has gone over 28.5 13 out of 33 games. Clearly +385 is just way too high of a price. In other words, poisson puts way too much value on an additional pt or two. I think it works pretty well for Reb and Asst, but I just don't see any value whatsoever for pt distribution.

You should have seen the chart TomG sent me 2 years ago when he used poisson when he tried to model these props.

It was off the charts laughable.

the game not within the game?

Probably explains why he sucked wang at total bases last year.

Actually I should probably not speak so assertively because my methodologies are chock full of leaks. It's not really a big deal though because I'm not trying to make $ at Pinny, I was mostly just betting at places like bovada and sportsbook.ag as long as they would have me. But I'm assuming you're not trying to break in at Pinny, and if you're going to start at retail sites you can find boatloads of edge regardless of how fuzzy your math is.

FWIW I use the normal distribution which is flawed for a number of reasons, but the lines at the aforementioned books are just so so bad that it's sufficient if you want to just make a few K until you get limited.

poisson is good for reb and asst but not for points.

even tomg 2 years younger should have known that

Hi Poogs my new friend.

I don't think I've ever placed a bet that requires the assumption of Poisson as the underlying distribution.

Two years ago B00T gave me some NBA data that I was playing around with and I created an Excel spreadsheet that used the data and modeled the half points using Poisson. It was clear that Poisson doesn't work very well and I abandoned the project (for other reasons--I would have adjusted the Poisson assumption to make that not an issue). I guess I should have known better but I don't really regret anything since it didn't cost me any money and was a good learning exercise both in setting up the Excel spreadsheet using the Poisson formulas and seeing how poorly Poisson estimates certain things.