[JVP3122]

Has anyone fully built the model for MLB that is described in Conquering Risk? My first steps are to simply recreate the sort of results described, obtaining line agreement more than 60% of the time. I was hoping to use the model as a starting point but I ran it for the entire 2010 season and got agreement close to 37% of the time. I'm wondering if someone's achieved Justin7's results and is willing to take a look at what I've done or if someone would be willing to trade models. I'm at a loss as to what I'm not doing right.

[DEP78]

Has anyone fully built the model for MLB that is described in Conquering Risk? My first steps are to simply recreate the sort of results described, obtaining line agreement more than 60% of the time. I was hoping to use the model as a starting point but I ran it for the entire 2010 season and got agreement close to 37% of the time. I'm wondering if someone's achieved Justin7's results and is willing to take a look at what I've done or if someone would be willing to trade models. I'm at a loss as to what I'm not doing right.

Ask Justin7. He seems like a nice guy. The worst he can say is no.

Good Luck.

[JVP3122]

Ask Justin7. He seems like a nice guy. The worst he can say is no.

Good Luck.

He's actually a very nice guy and helped with some mundane beginner problems when I was starting. If he happens to see this and wants to help then that's great. I just thought I'd open my inquiry to the think tank. The fact that my model is only having 36% agreement tells me I might be 'calculating backwards'

[Justin7]

He's actually a very nice guy and helped with some mundane beginner problems when I was starting. If he happens to see this and wants to help then that's great. I just thought I'd open my inquiry to the think tank. The fact that my model is only having 36% agreement tells me I might be 'calculating backwards'

When you tested the model, did you use preseason projections or live stats? The approach I suggest works very well for about 6 weeks, but then falls off. How well did it do during that period?

[JVP3122]

When you tested the model, did you use preseason projections or live stats? The approach I suggest works very well for about 6 weeks, but then falls off. How well did it do during that period?

Right now I'm just using preseason. I tried it for the first two months as well with not even similar results. In thinking about it more I think I really need to check and make sure I'm calculating everything properly. I have live stats at the ready to try and implement but first I would like to try and somewhat replicate the results. We'll see how it goes.

[Justin7]

Send me a copy of your spreadsheet to Justin@sportsbookreview.com. I'll take a look at it. Include stats from any point in May.

[mebaran]

Love how that pic just calmed the storm. Babies can prevent wars!

At any rate, JVP, J7 brings up a good point...that model wasn't meant to be used after a month or so into the season (the preseason projections are pretty much useless at that point). Also, notice how I say "wasn't" instead of "isn't". odds are that the approach he used in creating that model has been re-created numerous times, and any value that it previously had, is now also almost worthless because it is public knowledge (and I use the term "public" very loosely here, as most people who read his book didn't try attempting to build the model).

The only people making money using predictive modeling are the ones who keep the models (the good ones anyways) to themselves. Basically, if you have an out who offers you some correlated plays, or a consistently agreeing model.. KEEP IT TO YOURSELF...or just share it with me.

Side note: Take Justin7's model as what it is...a learning tool.

[JVP3122]

Love how that pic just calmed the storm. Babies can prevent wars!

At any rate, JVP, J7 brings up a good point...that model wasn't meant to be used after a month or so into the season (the preseason projections are pretty much useless at that point). Also, notice how I say "wasn't" instead of "isn't". odds are that the approach he used in creating that model has been re-created numerous times, and any value that it previously had, is now also almost worthless because it is public knowledge (and I use the term "public" very loosely here, as most people who read his book didn't try attempting to build the model).

The only people making money using predictive modeling are the ones who keep the models (the good ones anyways) to themselves. Basically, if you have an out who offers you some correlated plays, or a consistently agreeing model.. KEEP IT TO YOURSELF...or just share it with me.

Side note: Take Justin7's model as what it is...a learning tool.

Yeah I know what you mean. The thing is, though, in modeling I feel like the basic idea is valid. You want to have an estimate for given stats to take place in a game and then calculate runs scored off of that. That's why I at least want to try and reproduce his model and then improve on it myself by determining which parameters to vary to improve the model. If I'm way off base I'd appreciate someone letting me know but as a mathematician I know modeling in math has the same basic premise. I'm not expecting his model to have any value as it is but it seems like a good starting point.

[Justin7]

Some posts have been moved. Stay on topic here. The issue is modeling. If there are any more insults or personal attacks, I'll cut the poster's kneecaps off.

[MonkeyF0cker]

LOL. An excuse to remove digs at your book. Huh, Justin?

Unreal.

[durito]

you should just delete this sub forum it has no purpose anymore.

[MonkeyF0cker]

Ganch created it. Justin destroyed it.

[pedro803]

in the interest of sticking to the subject of modeling originally intended for this thread....

JVP and Justin, were you guys able to ascertain whether JVP executed the model correctly? Or perhaps as he suspected he coded something in a way that 'inverted' the results. Please share what you found out as it may be of interest to some of us reading this thread.

As for the sidebar, my two cents:

This subforum has lots of great threads in the archives, so no -- please don't delete it. I will admit that useful discussion has slowed down, but it is still nice to have a place that you can come to for discussion of this. I don't know of any other place like it on the net. By far it was the think tank that initially attracted me to SBR.

I wish everybody could just relax a little bit, I for one enjoy hearing from anyone willing to weigh in on the subjects at hand. If you deem somebody to be clueless, you can do your best to shed some light for them, but no need to attack -- just chuckle to yourself that you are ahead of at least some of the competition.

But also try to be sensitive to the reality that we are all coming to this not only with different levels of math expertise but also many different approaches to modeling and handicapping in general. For example, Ganchrow's posts were obviously beyond reproach for mathmatical/statistical expertise, but most of it was simply over my head. And not only that but also most of it did not address the things I am more interested in reading about -- and actually for me Justin's book fell into a similar category in that it was a lot more about probability theory than it was about how to actually build elaborate predictive models.

And for me Ganchrow's posts were often the same way (I suspect that this was the part he was willing to share, and the actual factors of his own models he kept closer to his vest) -- so for players of this ilk, it is about finding even a very small edge with probably a very rudimentary model. And then using a big bank roll to sit there and pound it, so lots of turn over and grinding but winning at a strong enough clip to overcome the vig (coupled with line shopping and all the other little things that make it work but take a lot of time).

I am more interested in finding a more powerful model so that profit can be earned with a much smaller bank roll, not saying the other way is wrong, I am sure it is a lot more tried and true. I could be wrong but I think I can pursue my endeavor with a small fraction of the math chops that Ganchrow has. I am going off topic here but......

What I am trying to illustrate is that a lot of the stuff that we sometimes get so hung up on is not so important in the bigger picture. Because the stats and the probability side of it is available in lots of dry textbooks. So if you are right you can easily verify it for yourself, and I suspect many times we argue about apples and oranges since we don't have the other poster's work right in front of us and they write something that we interpret to mean something different than they intended and wind up in a squabble where we aren't even talking about the same thing.

There are relatively few of us think tankers around, and I value everybody's input, so I for one wish that we could put a little bit of the snippiness aside and try to keep a good thing going.

Everybody agrees that nobody is going to share their trade secrets -- so part of the difficulty lies in the fact that it is hard to know what to start a thread about in here, and a big part of what moves us forward is our responses to the odd internet traveler that stumbles in here and poses a question. I don't have a solution to this problem -- with everybody protecting their own work maybe there isn't a solution.

Anyway, best regards everyone and happy think tanking! Don't give up on us just yet!

[Justin7]

pedro,

I am discussing his implementation of the model. The devil is always in the details... We were using different sources for projected stats, and different methods for predicting daily lineups. I haven't reviewed his spreadsheet, but having a few incorrect batters in the lineup or the wrong bullpen available could easily eat up your edge.