APBRmetrics

Two PG missed games today, Rose and Paul. I had the Bulls winning by 4.7 with Rose, and the Clippers by 19.2. Bulls lost by 16, the Clippers won "only" by 10.

Btw:
Including the 8 games played out so far tonight, my Power Ranking beat the spread 63 times out of the last 116 games. 90 times it predicted the correct winner. Vegas has 87 correct winners. The RMSE is 12.5 vs. 11.31 for Vegas.

The new "Vegas Ranking" I tested for the last 33 games has 10.97 RMSE, while Vegas has 11.26. It beat the spread 18 times. Interesting thing: The Power Ranking has 11.18 RMSE during that span while beating the spread 20 times. A blended version brings the RMSE down to 10.88 and also beat the spread 20 times. But well, that is obviously a rather small sample.

What about trying the least median square difference? Could that be a more robust measure for picking against the spread?

Well, I played around with a couple of different kinds of models. When I use a linear model in which I multiply the difference between AwayTeam and HomeTeam with 0.9 and substract 4.7 (instead of 3 as HCA), I bring the RMSE down to 11.95 while beating the spread 70 out of 119 times. But this model predicts the winner less likely than the original model based on the Power Ranking and with a HCA of 3 (88 vs 93). Vegas has 11.19 RMSE, only HCA of 3 has 13.39.

So far I would say, if I want to bet against the spread, use the linear model and if I want to predict the winner use just Power Ranking with HCA.

The new "Vegas Ranking" so far beats the old Power Ranking regarding the RMSE (10.59 vs. 10.84), but it is worse at beating the spread (21 vs. 19 out of 36 games). The linear model beat the spread 23 times out of the last 36. The best RMSE via least square I can get with the new "Vegas Ranking" brings the RMSE down to 10.42, BUT this model beat the spread only 17 times out of 36 games.

It also seems interesting that even with the small sample of 36 games, the Power Ranking model again predicted 78% of the times the correct winner (28 of 36). Vegas during that sample had 27 correct winners, while in the other it had 90 (75% vs. 75.6%).

The interesting thing about all that is that I only used all previously played out games, without any adjustment for recency or injuries. Implementing that would probably make it a better tool.

I think I said this earlier, but I would hypothesize that any rating system based on Vegas ratings will not beat Vegas spreads. That probably seems obvious but is worth repeating anyway. The value in having a Vegas rating is that it's: 1) easy for us to calculate 2) converges fast due to low noise and 3) going to be better than all but the best "professional" rating systems.

If you can beat it, that's great. But I think that will be very hard to do consistently. And btw, even if you can beat Vegas, have fun trying to bet large sums of money. I was pointed to and have been reading this book on the subject:

EvanZ wrote:If you can beat it, that's great. But I think that will be very hard to do consistently. And btw, even if you can beat Vegas, have fun trying to bet large sums of money.

What exactly is "large"? Can you sum it up, so not everyone has to buy/read the book?

EvanZ wrote:I think I said this earlier, but I would hypothesize that any rating system based on Vegas ratings will not beat Vegas spreads. That probably seems obvious but is worth repeating anyway. The value in having a Vegas rating is that it's: 1) easy for us to calculate 2) converges fast due to low noise and 3) going to be better than all but the best "professional" rating systems.

Maybe using "Vegas Ranking" is confusing, but when I say "Vegas Ranking" I relate to my own new calculation. I only compare the results to the Vegas spreads from this website: http://scores.goldsheet.com/merge/tsnfo ... slist.aspx
Those spreads have no influence on the results of my "Vegas Ranking" at all.

EvanZ wrote: If you can beat it, that's great. But I think that will be very hard to do consistently. And btw, even if you can beat Vegas, have fun trying to bet large sums of money. I was pointed to and have been reading this book on the subject:

Until now I have never bet any amount of money on anything except a couple of Poker rounds with some friends. And I have so far no intention on betting. My motivation for the Power Ranking was to find the best predictor for playoff success in order to find the a way to have the teams sorted by their "real" strength. As I pointed out before I'm more interested in the game of basketball than actual betting.

J.E. wrote:

EvanZ wrote:If you can beat it, that's great. But I think that will be very hard to do consistently. And btw, even if you can beat Vegas, have fun trying to bet large sums of money.

What exactly is "large"? Can you sum it up, so not everyone has to buy/read the book?

It seems like betting more than $20,000 on games was very difficult. And that was for football. I would think for NBA games, it will be quite a bit lower than that.

mystic wrote:

EvanZ wrote:I think I said this earlier, but I would hypothesize that any rating system based on Vegas ratings will not beat Vegas spreads. That probably seems obvious but is worth repeating anyway. The value in having a Vegas rating is that it's: 1) easy for us to calculate 2) converges fast due to low noise and 3) going to be better than all but the best "professional" rating systems.

Maybe using "Vegas Ranking" is confusing, but when I say "Vegas Ranking" I relate to my own new calculation. I only compare the results to the Vegas spreads from this website: http://scores.goldsheet.com/merge/tsnfo ... slist.aspx
Those spreads have no influence on the results of my "Vegas Ranking" at all.

Well, that's confusing. You might want to come up with a new name.

I realize that you are not a bettor, and neither am I. But money is a great motivator. A lot of people come out with ratings, but how many of those people actually put money where their mouth is? By keeping in mind the challenge of beating Vegas (even if only hypothetical), I think it encourages better model building. For example, do we just focus on RMSE? Or should beating the spread be the ultimate goal? I actually prefer to focus on the latter, because I think it might be more challenging. Going back to the Prediction Tracker (for NFL games), the correlation between beating the spread and RMSE does not appear to be very good at all. The rating system with the lowest RMSE beat the spread 52.5% of the time. Meanwhile, the rating system with the best record against the spread (56.9%) was in the middle of the pack as far as RMSE (and absolute error).

Here are the current results for NBA at Prediction Tracker:

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

"Compughter Ratings" has the best record ATS right now (59%). The ratings look a little odd to me, though, with LAC at #3. Also the magnitude of the ratings are smaller than any I've seen:

http://www.compughterratings.com/NBA

EvanZ wrote: Well, that's confusing. You might want to come up with a new name.

Yeah, I will.

EvanZ wrote: Here are the current results for NBA at Prediction Tracker:

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

"Compughter Ratings" has the best record ATS right now (59%). The ratings look a little odd to me, though, with LAC at #3.

Well, but he only predicts the correct winner in 71% of the cases. And I'm a little curious: Why is the number of games for the spread different than for the record? Does that mean that a game is discounted, if the prediction and Vegas are the same? If that is the case, I would go up to 59% too with the linear model.

EvanZ wrote: Also the magnitude of the ratings are smaller than any I've seen:

http://www.compughterratings.com/NBA

Which raises a question: How do I get the predicted scoring margin out of that? Because I doubt that this system beat the spread that often, if we use those numbers only adjusted by HCA.

Btw: I had also the Clippers as 3rd before last night.

mystic wrote: Which raises a question: How do I get the predicted scoring margin out of that? Because I doubt that this system beat the spread that often, if we use those numbers only adjusted by HCA.

??? That's why I thought the ratings look odd. Of course, none of these people tell you how they come up with those ratings, so they are practically useless for us.

As to your other question, sometimes the line is off for certain games. I don't exactly know why. It might be due to injury sometimes, or sometimes the game is predicted to be such a blowout, that they don't want to put a line on it. It's something to look into, though.

BTW, mystic, if you e-mail "Todd" at the Prediction Tracker, he'd probably be happy to add your ratings to the list. I send him my NFL ratings every week this season (they're listed under "Ridge Regression").

EvanZ wrote: ??? That's why I thought the ratings look odd. Of course, none of these people tell you how they come up with those ratings, so they are practically useless for us.

I actually don't really care how he comes up with it, I'm more interested in how to interpret those numbers. What does a rating of 2 mean in that ranking, when this is the best or -2?

A simple subtraction obviously doesn't give us the scoring margin.

EvanZ wrote: As to your other question, sometimes the line is off for certain games. I don't exactly know why. It might be due to injury sometimes, or sometimes the game is predicted to be such a blowout, that they don't want to put a line on it. It's something to look into, though.

No idea whether that explains it, because the line has 161 games, and Doctor Entropy for example has all 161 games in the record, but only 159 are counted for the spread. But well ...

mystic wrote:

No idea whether that explains it, because the line has 161 games, and Doctor Entropy for example has all 161 games in the record, but only 159 are counted for the spread. But well ...

Ah, I remember now, because it's happened to me. That's when it's a "push", the line and the result are the same.

EvanZ wrote: Ah, I remember now, because it's happened to me. That's when it's a "push", the line and the result are the same.

Interesting, but I don't understand why I would discount that. That happens twice this season and both times I lost obviously. Of 119 games I have 70 times beat the spread, once it was the same, twice a "push" and 46 times a loss.

What do you mean you "lost"? If it's a push, neither side wins.