[davopnz]

I figure with the low limits on player props, they've got to be beatable.

I've tried to build a model using Defense vs Position data but the results have been less than desirable.

I figure I need to go a step further and look at what's likely to happen based on matchups. For example the Wizards allow the fewest 3s in the league but they allow a lot of points in the paint and mid range therefore a team is likely to gameplan for that and feed their bigs more.

Is anyone running a successful model or handicapping player props successfully? If so what variables are you using?

Thanks!

[Gaze73]

I don't have one but I know a guy who has a model that calculates EV on prop bets and he's making loads of money so it can be done.

[davopnz]

I don't have one but I know a guy who has a model that calculates EV on prop bets and he's making loads of money so it can be done.

Yep, for sure it's possible - that's why the limits are so low. However, I can't for the life of me put together a working model unfortunately. It feels like the NBA has too much randomness, which of course averages out across the season but it doesn't seem to be quite as simple as betting overs in a good matchup and unders in a bad matchup. Perhaps things such as recent form and tendencies need to be accounted for more so than season averages but I do not know, I was hoping to get some input here.

[Waterstpub87]

Yep, for sure it's possible - that's why the limits are so low. However, I can't for the life of me put together a working model unfortunately. It feels like the NBA has too much randomness, which of course averages out across the season but it doesn't seem to be quite as simple as betting overs in a good matchup and unders in a bad matchup. Perhaps things such as recent form and tendencies need to be accounted for more so than season averages but I do not know, I was hoping to get some input here.

Haven't tested as much in basketball, but in baseball:

I used to weight past 2 seasons, current season to date and last ten games (so the current and 10 overlap). Roughly 40%/45%/15% at the mid point of the season, for things like strikeouts,total bases ect. Ran more data before last season, used a regression. The 10 games came up with almost no weight. It could be different in basketball, but my guess is that recent above average performance is mean-reverting, and too much weight to more recent events could be counter productive. Will likely look more closely at this during the summer to get ready for next fall.

[davopnz]

Haven't tested as much in basketball, but in baseball:

I used to weight past 2 seasons, current season to date and last ten games (so the current and 10 overlap). Roughly 40%/45%/15% at the mid point of the season, for things like strikeouts,total bases ect. Ran more data before last season, used a regression. The 10 games came up with almost no weight. It could be different in basketball, but my guess is that recent above average performance is mean-reverting, and too much weight to more recent events could be counter productive. Will likely look more closely at this during the summer to get ready for next fall.

Fantastic input, cheers. Definitely agree that recent form will be mean-reverting for the most part BUT it could also indicate a change in the players role which I definitely think is more likely to happen in the NBA than MLB. The same goes for team defense, if a team loses a player(s) to injury it could completely change their rating. This has happened to the Heat who were really good at defending bigs but have been getting obliterated since Bam Adebayo has been out.

Good example of a players role changing is Jarrett Allen, who is averaging 17.3 ppg for the season but if you scratch below the surface you'll see his usage rate has increased enormously as the season has progressed. In October his usage rate was just 13.7% and his PPG was 11.7 but since November his usage rate is up around 20% and he's now averaging 20 ppg. Last year with the Cavs he was averaging just 13.2 ppg as well. If you were using a mix of last years data and this years data without accounting for recent form you'd probably be betting unders and getting whacked because it appears his role has expanded and it's not just a flukey hot streak.

[Waterstpub87]

Fantastic input, cheers. Definitely agree that recent form will be mean-reverting for the most part BUT it could also indicate a change in the players role which I definitely think is more likely to happen in the NBA than MLB. The same goes for team defense, if a team loses a player(s) to injury it could completely change their rating. This has happened to the Heat who were really good at defending bigs but have been getting obliterated since Bam Adebayo has been out.

Good example of a players role changing is Jarrett Allen, who is averaging 17.3 ppg for the season but if you scratch below the surface you'll see his usage rate has increased enormously as the season has progressed. In October his usage rate was just 13.7% and his PPG was 11.7 but since November his usage rate is up around 20% and he's now averaging 20 ppg. Last year with the Cavs he was averaging just 13.2 ppg as well. If you were using a mix of last years data and this years data without accounting for recent form you'd probably be betting unders and getting whacked because it appears his role has expanded and it's not just a flukey hot streak.

Certainly good points. The balance really is whats key. NBA is difficult mathwise, because individual players are such a big part of it.

[WireWire]

I don't have one but I know a guy who has a model that calculates EV on prop bets and he's making loads of money so it can be done.

Must be the guy I know to, guys betting sometimes 20-50 players a night at around $1,200 per player.

[davopnz]

Certainly good points. The balance really is whats key. NBA is difficult mathwise, because individual players are such a big part of it.

I'm starting to think focusing solely on the player, regardless of the matchup is wherein success lies. The best and worst defense in the league only deviate 7-8% from the league average. If you apply that to a single player, it hardly moves the needle. I.e if player A averages 25 points a game and is playing Charlotte (8% worse than league avg D) 1.08 * 25 = 27 points. Given the variance in individual scoring, predicting a player to score 2 more than their average isn't significant.

DvP data shows bigger deviations from league avg but I really don't know how accurate this data is, considering how many players play multiple positions and even different styles of players within one position (i.e traditional big vs a shooting big) which teams may defend differently.

I think the key lies within the individual players stats and how the books are setting their line. I need to do more study but it looks to me like books are setting player lines using their previous 5-10 game averages where there is a discrepancy from season averages. These 5-10 game averages are often inflated from one or two big games in that stretch and therefore using the median could paint a different picture.

I still don't know how much data to use so that it's relevant to the players current role but not too skewed by streaky form.

Two good examples; James Harden and Julius Randle. Both well below last seasons scoring averages, can we determine that this seasons averages are now an accurate representation of their ability going forward? Or will they get back to last seasons averages?

Good discussion; keep it going.

[Waterstpub87]

I'm starting to think focusing solely on the player, regardless of the matchup is wherein success lies. The best and worst defense in the league only deviate 7-8% from the league average. If you apply that to a single player, it hardly moves the needle. I.e if player A averages 25 points a game and is playing Charlotte (8% worse than league avg D) 1.08 * 25 = 27 points. Given the variance in individual scoring, predicting a player to score 2 more than their average isn't significant.

DvP data shows bigger deviations from league avg but I really don't know how accurate this data is, considering how many players play multiple positions and even different styles of players within one position (i.e traditional big vs a shooting big) which teams may defend differently.

I think the key lies within the individual players stats and how the books are setting their line. I need to do more study but it looks to me like books are setting player lines using their previous 5-10 game averages where there is a discrepancy from season averages. These 5-10 game averages are often inflated from one or two big games in that stretch and therefore using the median could paint a different picture.

I still don't know how much data to use so that it's relevant to the players current role but not too skewed by streaky form.

Two good examples; James Harden and Julius Randle. Both well below last seasons scoring averages, can we determine that this seasons averages are now an accurate representation of their ability going forward? Or will they get back to last seasons averages?

Good discussion; keep it going.

Anything that is a pro-sport, I focus solely on players. This kind of breaks down a little with defense, because some players are better at covering others, IE a good center is good at covered other centers, preventing shots in the paint, but not protecting threes, whereas a good shooting guard would be the opposite.

I've read that some do things like weighting defensive types by player and using that. I haven't found a good way to do it. But conceptually, its an interesting topic. I use the NBA opponent shooting data, and aggregate by minutes. This is a rough way to calculate, but more research needed.

[Gaze73]

Must be the guy I know to, guys betting sometimes 20-50 players a night at around $1,200 per player.

The guy I know makes only 3-5 plays a night.