Poll: RPM's degree of efficacy in sorting players

[schtevie]

Permaximum, it is your scheme to establish a relationship between "roster turnover" and "prediction accuracy". You defined the variables, so do it clean. For each metric, there are X number of games in your sample where roster turnover is between 0 and 10%, Y number of games where roster turnover is between 10 and 20%, Z between 20 and 30%, and so on up to 90 to 100%. This is how the X-axis of your graph should be defined (and the sample size of each roster turnover decile ought be shown in parentheses.)

As for the widget being too archaic, I do believe you are choosing to kid yourself. There is no reason to believe that this simple model doesn't provide estimates of "unexpected outcomes" that are very much in the ball park. HCA is but a second-order factor, as every team plays home and away, fundamentally canceling its influence, in terms of "unexpected" outcomes.

But never mind the upper bound, if having a number too close to actual RPM performance is too ungenial to your priors; that can be a discussion for another day.

Just present your data in graphical form, where a clear picture is available, showing the effect of roster turnover on predictive accuracy of game results. This is the basis of your arguments, and it ought be clearly presented.

What you're suggesting changes nothing. But I will do it for you. You want 10% of RT increments right?

That would be great, for RPM, with sample size in parentheses; and if I could be greedy, the same again for BPM? Thanks.

[permaximum]

Just to be clear you want it as;

0-10
10-20
20-30
30-40
or
0+
10+
20+
30+
40+
....

The first method would encounter sample problem very early. There are only limited amount of games in the 3pt. era of 1984-2016. RPM is not even doable because of the lack of data for it.

[shadow]

You could estimate the upper bound by looking at how well vegas does predicting the winner of games (e.g. http://www.oddsportal.com/basketball/usa/nba/results/).

As for errors, mathematically you should expect sqrt(p*(1-p)/n) error in the values you calculate. This means, for instance, a +/-5% error if you find 60% prediction success on a sample of 100 games, a +/- 2.2% error if the sample is 500 games, a 0.5% error if the sample is 10000 games.

According to Covers, the favorite based on the closing line has won 66.82% of games in the NBA this year. This would indicate a +/- 1.34% error for 1230 regular season games.

http://www.covers.com/pageLoader/pageLo ... eason.html

All of the public team based ratings tracked at ThePredictionTracker failed to pick straight up winners at a higher rate than the Opening Line. Most of those systems are essentially variations of SRS.

http://www.thepredictiontracker.com/nbaresults.php

[permaximum]

@schtevie

Beyond 70-80%, sample becomes too small for any comparison. RPM and BPM's on target prediction spread and success is extremely similar so you can take BPM for RPM comparisons. If you still insist I can make RPM chart with extremely small sample for such comparison which woulnt'be a valid approach but whatever. If I did this research after this year, 1-year of more RPM data would help but findings would be the same no matter what.

[Rd11490]

This is a much easier to consume visualization.

If I'm reading this correctly, MPG is the least successful prediction signal in the vast majority of cases, but outperforms other signals in the extreme RT scenarios. Please correct me if I'm wrong.

[DSMok1]

@schtevie

Beyond 70-80%, sample becomes too small for any comparison. RPM and BPM's on target prediction spread and success is extremely similar so you can take BPM for RPM comparisons. If you still insist I can make RPM chart with extremely small sample for such comparison which woulnt'be a valid approach but whatever. If I did this research after this year, 1-year of more RPM data would help but findings would be the same no matter what.

I agree, this is much easier for me to digest. Very useful data!

I wonder why the accuracy is going up as the roster turnover increases? That's just odd. I wonder what could be driving that?

[rlee]

Just to clear up a misconception that somehow persists even to this day: Vegas lines do not (and are not intended to) "predict" outcomes of games. They are set in an attempt to anticipate how the public will place its bets. (The ideal the house is aiming for is a 50-50 split by the betting public). That is why there are "opening" and "closing lines". The line "moves" (i.e. is adjusted ) in an attempt to maintain that 50-50 split as game day/time approaches. btw- A similar concept applies to the setting of over/under seasonal wins/losses. It has nothing to do with how many games "Vegas thinks" a team will win but, rather, O/Us are set in anticipation of (and to influence) betting behavior and, likewise, that's why they can change as it gets closer to opening day.

[Nathan]

Prediction accuracy would go up for high-RT games if those games are more frequently lopsided than low-RT games. This seems plausible. I could be wrong, but games with >50% RT probably disproportionately involve an awful rebuilding team fielding a motley crew of youngsters and journeymen. Those games are inherently easier to predict (bet against the awful rebuilding team).

RE: Vegas, I didn't mean to imply that vegas line accuracy will give the theoretical max prediction accuracy, just that it would set a decent lower bound on the theoretical max prediction accuracy.

[rlee]

The comment was not aimed at your statement but rather at the magical misleading persistence re: how lines are set. It is bad enough for individual games, but in a way even worse that virtually every reference to seasonal W/L O/Us gets it wrong. For a good laugh, just Google the topic. After all this time, I don't quite get why so many folks are still unaware of the purpose of O/Us and how they are set.

[sndesai1]

Just to clear up a misconception that somehow persists even to this day: Vegas lines do not (and are not intended to) "predict" outcomes of games. They are set in an attempt to anticipate how the public will place its bets. (The ideal the house is aiming for is a 50-50 split by the betting public). That is why there are "opening" and "closing lines". The line "moves" (i.e. is adjusted ) in an attempt to maintain that 50-50 split as game day/time approaches. btw- A similar concept applies to the setting of over/under seasonal wins/losses. It has nothing to do with how many games "Vegas thinks" a team will win but, rather, O/Us are set in anticipation of (and to influence) betting behavior and, likewise, that's why they can change as it gets closer to opening day.

that's only partially true, but regardless, the end result is that the closing line on high-limit popular events is likely more predictive than anything else you can find

season win totals are basically low-limit prop bets and require people to tie up their money for 6-7 months so it tends to attract casual bettors, hence the consistent success of pm metrics against those lines

[rlee]

As to season-long O/Us, virtually everyone who refers to the process gets it wrong. The %age of descriptions that state "vegas predicts", "vegas expects" or their equivalent is almost 100. (Google it). Very misleading.

[permaximum]

@schtevie

Beyond 70-80%, sample becomes too small for any comparison. RPM and BPM's on target prediction spread and success is extremely similar so you can take BPM for RPM comparisons. If you still insist I can make RPM chart with extremely small sample for such comparison which woulnt'be a valid approach but whatever. If I did this research after this year, 1-year of more RPM data would help but findings would be the same no matter what.

I agree, this is much easier for me to digest. Very useful data!

I wonder why the accuracy is going up as the roster turnover increases? That's just odd. I wonder what could be driving that?

Very good question. I answered this before but I didn't prove my answer. It's because games with higher roster turnover rates are found more in 2000s and 2010s and almost every metric's prediction success gets better when we get closer to our time. Since you asked so nicely here are the proofs:

[permaximum]

One interesting find is both BPM and the simple box score metric struggled constantly between 2002-2007. I'm not sure if it's related to the game that was played then.

[shadow]

Just to clear up a misconception that somehow persists even to this day: Vegas lines do not (and are not intended to) "predict" outcomes of games. They are set in an attempt to anticipate how the public will place its bets. (The ideal the house is aiming for is a 50-50 split by the betting public). That is why there are "opening" and "closing lines". The line "moves" (i.e. is adjusted ) in an attempt to maintain that 50-50 split as game day/time approaches. btw- A similar concept applies to the setting of over/under seasonal wins/losses. It has nothing to do with how many games "Vegas thinks" a team will win but, rather, O/Us are set in anticipation of (and to influence) betting behavior and, likewise, that's why they can change as it gets closer to opening day.

This isn't true in regards to point spreads on individual games where Pinnacle/CRIS are taking 10K+ limits. According to the article linked below, the closing point spread at CRIS explained 98.85% of the variance in margin of victory. For every one point a team is favored they won by 0.97 points on average. This was over a massive sample of games from 1995-2014.

http://fansided.com/2015/04/10/freelanc ... l-betting/

If books consistently posted lines that were significantly different from the true odds (or as close as you can get to approximating the true odds), then sharps could easily crush them by betting on the opposite side. They can certainly afford to shade point spreads by about 1 point without offering +EV bets on the opposite side, but the idea that they shade them any further than that is not true. TeamRankings has data on NBA line moves and based on about 2 seasons worth of games, the point spread only moved 2+ points from the opening line 9.4% of the time. So it's not as if oddsmakers are throwing up completely wild guesses for opening lines in the NBA since the opening lines are rarely moving a significant amount.

https://www.teamrankings.com/nba/odds-m ... /movement/

Sportsbooks like Pinnacle/CRIS profile their user's accounts and they will move the line more for bets made by an account that is profiled as sharp, even if that account is betting less per game than a lifetime loser. Hypothetically, if I opened a brand new account and Haralabob gave me his plays to bet on his behalf, my bets would start moving the line by increasing amounts as the system learns that my bets are likely +EV. A clueless whale who is a lifetime loser and betting more per game than me likely wouldn't make the line move at all. Pinnacle supposedly won't cut limits for its players, but it's believed they give slightly higher limits to accounts that are lifetime losers. Cantor Gaming in Vegas came out and said they'd take 1 million dollar bets, but when Haralabob asked their rep at Sloan if he could bet 1 million he was told he could only bet 5K per game. However, I'm willing to bet that if Haralabob bet 5K on an NBA game the line will move more than if Floyd Mayweather bets 100K. It's about respect as much as it is about money.

Apparently Cantor wouldn't even take bets from Haralabob though. They would reject his bets and then bet them at Pinnacle themselves. Haralabob posts as coltranedog at twoplustwo and made the following post in a thread there:

Posted by coltranedog
He wasn't anything special at all so there won't be any added edge.

I have no info on the arrests etc, but I will say that when I was betting Cantor, they'd often deny my bets and then the lines at pinnacle and a few other books would move instantaneously.

I was betting through their tablet and it was really frustrating, the little wheel on the tablet would spin, then the lines at books offshore would move and Cantor would refuse the bet.

Someone at Cantor was either betting these other books or giving info to others to go out and bet the games. I guess its more likely that Colbert was the guy doing that.

Full twoplustwothread: http://forumserver.twoplustwo.com/29/ne ... d-1260326/

This is why Haralabob's method of getting bets down is just as important as the system he's developed for modeling games. It becomes increasingly difficult to maintain accounts at most sportsbooks once they identify you as a potential +EV bettor. Which is usually done by tracking how favorably the line you bet compares to the closing line.

[rlee]

I guess I'm misunderstanding this excerpt from your linked story:

" Books are forced to take a side, as it’s impossible to have exactly the same dollar amount of both sides of a bet. They would like to be heavy the “right” side — the side they feel is better, although it’s questionable whether books know which side is the right side. Books also need to ensure they are not unnecessarily exposed on a single game — even if a bookmaker is confident on a side, risk management dictates the book is loathe to take a 90%-10% split in money wagered on a game, even if they are holding the “right” side. Hence, books move the lines in order to attract money on the other side."

Please explain what it means that may be different from what it seems to be saying.

Also, where did the "straw man" of "just throwing a number out there" come from?

Thanks