Is PDSS a good way to analyze the Raptors defense?

[Fairview4Life]

Fairview: I thought the bookies ask you to take an over/under, not just predict W/L? If they did W/L only, they'd need to make the money offered for betting on the Lakers in Lakers vs. Clippers pretty small.

Or alternatively, if you want to have equal money offered for both choices, adjust the over/under to where the chance of each event happening is 50% (e.g., Lakers favored by 7 points over Clippers)

I was talking about the spread. If you're talking strictly W/L, hitting at 74% won't really help you when gambling either. I thought you were talking about beating the spread 74% of the time.

[Ripp]

Can you define some of these terms in plain English (spread, "vig", etc)?

Do you have any sense of how well elite gamblers do at beating the spread? You said 53% is break even...is 54 or 55 considered very good? How much above break even is elite versus just good?

[notskeebs]

Predicting basketball outcomes with stats is like predicting the weather with a stone on a string.

ie.

Stone is wet, its raining
Stone casts a shadow, its sunny
Stone is bouncing, earthquake
Stone is covered in white stuff, its snowing

btw vegas makes bets based on how many wins a team will have. They give you the over under at the beginning of the year and you bet on if its over or under. So if you can predict a teams record 74% of the time you'd be a very very rich man.

ie. Vegas over under on the raps lets say 30 games. If your algorithim can predict the games 74% of the time then you find the numbers for each team (30 teams) and place a $1,000(or more) bet on each team and you'd be right 74% of the time.

In which case you wouldnt be wasting your time here trying to convince us Bargnanis sucks.

[ranger001]

Fairview: I thought the bookies ask you to take an over/under, not just predict W/L? If they did W/L only, they'd need to make the money offered for betting on the Lakers in Lakers vs. Clippers pretty small.

Or alternatively, if you want to have equal money offered for both choices, adjust the over/under to where the chance of each event happening is 50% (e.g., Lakers favored by 7 points over Clippers)

I was talking about the spread. If you're talking strictly W/L, hitting at 74% won't really help you when gambling either. I thought you were talking about beating the spread 74% of the time.

Depends on the matchup, if the Raptors play Miami 4 times then yeah we can expect Miami to beat the Raps at least 3 out of 4 and the bookies will price accordingly. If he can predict somewhat evenly matched teams 74% of the time then I think he is rich.

I have to say I'm skeptical, a lot of people with way more stats skills have tried and have been unable to overcome the random factor.

[Courtside]

Can you define some of these terms in plain English (spread, "vig", etc)?

Do you have any sense of how well elite gamblers do at beating the spread? You said 53% is break even...is 54 or 55 considered very good? How much above break even is elite versus just good?

Ripp's wheels are turning quickly now, figuring out a way to pay for the rest of his education by watching basketball even more than he already does.

[Ripp]

So let me see if I understand the comments here correctly, on how to translate from Vegas to a prediction system:

1) Vegas offers some sort of over/under on say a Raps vs. Miami game...Say Miami at home is favored by 15. This roughly represents 50% odds. So what you'd like is some system that can confidently predict that the odds of going above 15 are much higher than 50%.
2) Suppose you have a system that spits out predictions for the final margin of the game...say it predicts 18. Actually, it needs to spit out more than just a number...it would need to spit out a distribution. So for simplicity let's say it spits out gaussians, predicts a mean of 18 with a variance of V. So assuming I trust my system, I need to see with what probability a gaussian with mean 18 and variance V is larger than 15 (or equivalently, a mean 3 gaussian with variance V is larger than 0.) Call whatever number this is P.
3) If the variance V is huge (e.g., 1 trillion), then P is going to be barely over 50%, only slightly better than flipping a coin. Since Vegas is going to stack things in a way such that doing close to coin-flipping is going to be unprofitable.
4) On the other hand, if the variance V is small, you'll get something that might be well above 50%.

Interesting...so if you are trying to build a system for gambling, it is a bit more complicated than I thought...you need more than just a final predicted score, but also some understanding of the chance of deviating from that final predicted score.

Hrm, food for thought. Thanks for the interesting discussion.

[ranger001]

As notskeebs was saying the time to make your money if you can predict wins is at the start of the season when you can make bets on how many wins a team will have. For most players this years stats will be close to last year stats.

Run the numbers and come up with wins for each team. There's 30 teams and you win 70% of the time lets say, or 21 times. If you bet $1000 on each team you will walk away with $21,000 profit.

[Ripp]

It doesn't seem that simple, for the reason I said... Let's say i predict a team will win 40 games. Vegas puts the over/under at 39. If the team actually wins 38 games, that is not very far from the 40 I predicted...yet I'd lose money. So Vegas is a bit more complicated...effectively you are looking for cases where there is a big discrepancy between what Vegas thinks the 50/50 line is and what you think it is (e.g., Vegas predicts a 40 win team, I think instead a 60.) If the gap between where Vegas thinks the 50/50 line is and what I think it is is too small, then it doesn't make sense to bet.

At least, this is my understanding of the situation. So the 74% figure I mentioned earlier is not really relevant, if Vegas is managing to do 73.5...there is no money to be made.

Of course, feel free to correct me if I'm wrong, I'm just reasoning this stuff out, I don't really understand gambling perfectly well.

[notskeebs]

Thats the point. If you can figure these things out people would be making a killing, and vegas would not be allowing bets on such things.

Believe me smarter people than you and I have tried.

There is a reason vegas takes bets on basketball and not the exact second minute hour that haileys comment passes over altanta.

Stats can not be used to predict outcomes in such a way that it is viable to predict them to a certain degree.

Last year philly was predicted to be top 5 in the conference, they ended up with the 2nd pick overall. A lot of lucky gamblers who picked under are laughing all the way to the bank. And I can almost guarantee most of the people that won didnt look at statistics or algorithms.

Statstics are good for some things, like predicting GDP or figuring out demographics. But once you start trying to predict what 13 people do when pitched against 13 other people 82 times you're basically just guesstimating. Especially when you take 5 man units out of those 13 and try to assign blame to one player without taking into account the 95 million other variables in play.

[ranger001]

Ripp if you predicted 40 wins and the team won 38 then you're wrong. What exactly did you mean then by your 74% figure? 74% within 5 wins or so?

In any case though you could predict 40 on another team and they win 42, those things would balance out if you're usually correct.

[Ripp]

notskeebs: That isn't a good argument against the technique. In essence you are saying that Vegas does a pretty job of finding the 50% line, and it is hard to find other techniques that do substantially better in finding this 50% line.

If Vegas gets 80% of games and you can come up with a technique of your own that also gets 80% of games, you will on average lose money at Vegas (I guess due to fees.) But does that mean that your technique sucks? I don't see why this is true.

Ripp if you predicted 40 wins and the team won 38 then you're wrong. What exactly did you mean then by your 74% figure? 74% within 5 wins or so?

In any case though you could predict 40 on another team and they win 42, those things would balance out if you're usually correct.

We have a block of 410 games. 74% of them the winner is accurately predicted. On 26% (106 or 107), it fails. If Team X plays 15 games, what I do is predict which of those are wins and losses, not how many total W/Ls they got.
As I mention in a previous post, I wouldn't use it in this direct manner to predict "how many total W/Ls"...I'd instead rank teams by efficiency differential and use say Pythagorean wins (or whatever the equivalent tool is for eff diff).

Just in case the above didn't make sense, predicting that in 10 coin flips, 3 of them will come up heads is quite different from saying the 1st, 4th, and 5th will be heads, and the other seven tails.

[notskeebs]

What does the order of the wins matter? And if you can do it for 15 games why cant you do it for 82?

Also your coin flip analogy is a weird one to use considering each coin flip is independent of each other and any algorithm you use will be basically guessing between 50% each time.

I just dont think its possible for stats to quantify what is going on on the court. Unless you are talking about offence of course.

[Fairview4Life]

Vegas isn't really trying to predict wins and losses. They are trying to predict a number that will get 50% of the betting public to bet on each side, so they guarantee a 10% profit. The trick is to try and find some value to beat the public, not Vegas. You're really trying to outrace everyone else, not the book. In rare cases books even take a stand on one side and let those 10 sharp bettors even things up for them after seeing where the 1000 Joe Publics are going. If 1000 people are betting $10 the Heat will win over 70 games, go with the 10 guys betting $1000 they won't.

There are a few guys who do have systems that make them money against the spread, but they are professional, much better with stats than you are, and are rich. Truehoop interviewed one such bettor and had him go over the claims Donaghy made, for example (he said Donaghy was a liar, basically).

[Dr Positivity]

Vegas isn't really trying to predict wins and losses. They are trying to predict a number that will get 50% of the betting public to bet on each side, so they guarantee a 10% profit. The trick is to try and find some value to beat the public, not Vegas. You're really trying to outrace everyone else, not the book. In rare cases books even take a stand on one side and let those 10 sharp bettors even things up for them after seeing where the 1000 Joe Publics are going. If 1000 people are betting $10 the Heat will win over 70 games, go with the 10 guys betting $1000 they won't.

There are a few guys who do have systems that make them money against the spread, but they are professional, much better with stats than you are, and are rich. Truehoop interviewed one such bettor and had him go over the claims Donaghy made, for example (he said Donaghy was a liar, basically).

Bingo. That's also why they adjust the line when it comes heavily on one side.

[Ripp]

Vegas isn't really trying to predict wins and losses. They are trying to predict a number that will get 50% of the betting public to bet on each side, so they guarantee a 10% profit. The trick is to try and find some value to beat the public, not Vegas. You're really trying to outrace everyone else, not the book. In rare cases books even take a stand on one side and let those 10 sharp bettors even things up for them after seeing where the 1000 Joe Publics are going. If 1000 people are betting $10 the Heat will win over 70 games, go with the 10 guys betting $1000 they won't.

Indeed...hence my comments at the top of this page saying the task is really to figure out if the "over" is going to occur with probability significantly different from 50% or not, not really predicting what the final value is (though of course having a good estimate of the final value helps.) If you believe that the over occurs with say 60% chance, then it makes sense to bet on the over. If 40% or less, bet the under. If the spread is too close to what you believe the final outcome will be, then unless you have exceedingly great confidence in your prediction (e.g., you think the variance around your prediction is very small), then you probably shouldn't be betting. So like I said earlier, the trick seems to be finding cases where there is a big gap between the spread and your own prediction...those cases are likely to be the most profitable.

There are a few guys who do have systems that make them money against the spread, but they are professional, much better with stats than you are, and are rich. Truehoop interviewed one such bettor and had him go over the claims Donaghy made, for example (he said Donaghy was a liar, basically).

Yeah, I saw that TrueHoop interview of Haralobos Voulgaris. Dude is actually from Toronto. He has a blog too...it is pretty interesting reading.

[ranger001]

There are a few guys who do have systems that make them money against the spread, but they are professional, much better with stats than you are, and are rich. Truehoop interviewed one such bettor and had him go over the claims Donaghy made, for example (he said Donaghy was a liar, basically).

No he didn't he said Donaghy fixed the games.

http://espn.go.com/blog/truehoop/post/_ ... hy-scandal

Most of the information I have about Donaghy is from the 2006-2007 season and its plain as day to me that Donaghy did change the outcome of the games, I don't see how any rational human being could argue otherwise.

[Ripp]

^--- I found it interesting that Voulgaris seems to believe in APM himself for measuring defense:

http://aloneinthecorner.com/post/527082 ... nsive-team

I really wonder how much he uses it to make gambling decisions...

[Dr Positivity]

Fairview: I thought the bookies ask you to take an over/under, not just predict W/L? If they did W/L only, they'd need to make the money offered for betting on the Lakers in Lakers vs. Clippers pretty small.

Or alternatively, if you want to have equal money offered for both choices, adjust the over/under to where the chance of each event happening is 50% (e.g., Lakers favored by 7 points over Clippers)

There are moneylines and point spreads. Moneylines operate on W/L, but are weighted

eg. Toronto @ Miami

Point spread = -11.5, roughly even odds on both sides. Bet on Miami, they will by more than 11 you win, less than 12 you lose

Moneyline = -500 for Miami = You have to bet $500 to win $100, +500 for Toronto = You can win $500 with a $100 bet

A lot of the moneyline action is on parlays however. Meaning you bet two moneylines and need both to win the bet - you get a much higher return, but of course at a higher risk. Two -110 bets can be worth as much as something like a +150 one

[Dr Positivity]

And really dumb people make 5 bet parlays for some "bet $3, win $175!" action. I work retail and every point spread I sell is $2 massive parlays to win $80-$175, lol

[Ripp]

Thanks for actually naming these things...I ended up googling and finding a pretty nice intro on about.com. All of these bets (the simple ones like point spread/totals/moneyline and the complex ones like parlays/teasers) seem equivalent...your task is to estimate the probability of each outcome and thus what your estimated payoff will be for different actions, versus what the sports books offer.

So due to those dumb people, won't the most profitable bets be available in parlays and teasers? Since they will have the wrong intuition about what the likelihood of different events are, and thus badly estimate the expected payoff of different choices? And of course, the more mistakes they make, the more potential for profit there is.