Sports Forum   |   Scoreboard   |   Contact Us   |   WinPicks Software   |   Talkstock   |   Advertise   |   Follow Us on Twitter   |   Register  |   Login 
Register to post in the sports forumDownload the ScoresLine App for Android 

 
 
Forums   Register   Profile   Inbox   Address Book   Subscribe   My Forums   Search   FAQ   Login   Log Out

Another question for Bilbal.   Logged in as: Guest
Users viewing this topic: none
  Printable Version
All Forums >> [Sports Handicapping Resources] >> WinPicks Software Forum >> Another question for Bilbal. Page: [1]
Login
Message << Older Topic   Newer Topic >>
Another question for Bilbal. - 11/10/2005 7:33:20 PM   
Idlemind26@yahoo.com

 

Posts: 8
Joined: 10/27/2005
From: Oklahoma City
Status: offline
Since you're the pro... This might be a dumb question, but I'm new to Winpicks and was curious if you could clarify an issue for me. Is it possible to over-optimize a formula? The reason I ask is I also use technical analysis software for stock trading etc; and one of the big no-no's is to optimize formula's. It seems the more they are optimized there is a lot more chance of them losing their "robustness" and ability to work well in the future. I have been using formulas from the warehouse and I will run nightshift on them. They will typically be anywhere from 55-60% in the beginning which is realistic, but when nightshift is done they will be in the 80% range... I just don't see that lasting in the long run. I guess the basic question is if it's possible to get a worse results from tuning a formula even when it says it's a higher % after tuning. I hope I worded this right! Thanks for the help, it's much appreciated!

_____________________________

**Fear the Clown**
Post #: 1
RE: Another question for Bilbal. - 11/10/2005 11:20:48 PM   
farbror

 

Posts: 30
Joined: 10/17/2005
From: Uppsala
Status: offline

My gut feel is that you almost can see from an optimized formula that it has lost all its predictive power. If someone tells you to bet on teams with red socks if the total sum of their jersey numbers of the starting line-up is a prime number, then I think it is healthy to be a sceptic.

If the optimized formula gives you strange parameters, say an average Home Court advantage of -25 points then I would be a bit sceptic.

One healthy approach would be to use some version of the following validation scheme: If you are willing to believe that "things are stable" (Whatever that means? You need to believe that history repeats itself for formulas to be succesful) over the last four seasons then I would search for promising formulas in year one and two and then validate those formulas using data from year three and four.


Cheers, farbror
Sweden


_____________________________

www.farbrortheguru.com

(in reply to Idlemind26@yahoo.com)
Post #: 2
RE: Another question for Bilbal. - 11/11/2005 6:01:07 AM   
bilbal

 

Posts: 147
Joined: 11/24/2003
From: Harrisonburg, VA
Status: offline
you can over analyze anything and formulas are no exeception! I do believe you can overtune a formula and that in doing so, you will give yourself less reliable predictions.

Like I mentioned in another post, I don't test back more than 3 years in hoops because of the way the formulas are structured (NBA with weary/rest factor, and NCAA with conference factor). I believe that the longer back you test, the less reliable the power ratings are and power ratings are half of the equation for how results are predicted.

Like farbror mentioned, I would check the results of your formula and see what the parameters look like...or go back through the history of the formula and see how it performs. Whacky numbers usually mean whacky results.

One way to look at a formula that hits at 80% is this...Over time, all things statistically fall into a range of roughly 47-52%, as will the results with your formulas. Test out your formula over different date ranges and I'm sure it will look something like below:
3000 dates - 50%
2000 dates - 52%
1000 dates - 54%
500 dates - 58%
250 dates - 62%
100 dates - 65%
75 dates - 80%

In a sense, you know that your formula is statistically doomed to a range of 47-52% so you might want to "fade" the picks it makes!

I think the formulas that do best over the long run have small betting windows and smaller "point spread multipliers". I would also test the formula with different databases. If you see that the only database in the green is the one database you have it set at, it's a bad formula. If majority of the results are green, you're onto something good.

I hope this helps, good luck today!
Bill



(in reply to Idlemind26@yahoo.com)
Post #: 3
RE: Another question for Bilbal. - 11/11/2005 7:16:02 AM   
farbror

 

Posts: 30
Joined: 10/17/2005
From: Uppsala
Status: offline


<< I think the formulas that do best over the long run have small betting windows and smaller "point spread multipliers".
>>



Bill,
Thank you for interesting input. What do you consider as a small betting window? Small spread multiplier?

I have always multipliers far away from "1" with scepticism.

Cheers, farbror


_____________________________

www.farbrortheguru.com

(in reply to Idlemind26@yahoo.com)
Post #: 4
RE: Another question for Bilbal. - 11/11/2005 8:17:25 AM   
bilbal

 

Posts: 147
Joined: 11/24/2003
From: Harrisonburg, VA
Status: offline
I stick to mainly 1-3 points for a betting window, and usually less than 1 for a multiplier.

Let me do an example, maybe that will help expose what values do to predictions...try this:

create new formula with the following values:
HCA - +3
BOL - +20
PSM - +.5
LBL - +1
UBL - +2

leave all weariness/rest factors as 0. This effectively uses only power ratings to forecast games. (And note that if you use a Blowout limiter of 0 in a formula, that effectively uses only the weariness/rest factor for predicitions, something to play around with in your super secret spare time!)

use a rolling average of 1 game and analyze year to date... I get 10-2 using my own lines (which may differ than closing lines) 10-2, talk about not overanalyzing, right? If you look at games, the Winpicks lines are fairly tight with the betting lines. If you change the PSM higher and go back and look at the lines it's predicting, you can see they are much higher (and sometimes whacky, predicting 40-50 point victories), the smaller the PSM, the closer the lines will be "believable", and the final scores it predicts are a little more in line with the actual final scores.

Adjusting the window forces the program to pick only games that fall inside the program's predictions, so if you use a very low PSM, it's unlikely the software will find any reliable picks in the 10-20 point window, and I think the same would apply for using a high PSM and a 1-2 betting window. So you should probably adjust the PSM higher as you adjust your betting window higher to get the most accurate results. Anyone else have thoughts on that?

I hope this helps!



(in reply to Idlemind26@yahoo.com)
Post #: 5
RE: Another question for Bilbal. - 11/11/2005 3:26:27 PM   
Idlemind26@yahoo.com

 

Posts: 8
Joined: 10/27/2005
From: Oklahoma City
Status: offline
Thank you both for the help... that definitely sounds more reasonable! It's real hard to tell where that fine line is with optimization... Thanks again fellas!


_____________________________

**Fear the Clown**

(in reply to Idlemind26@yahoo.com)
Post #: 6
RE: Another question for Bilbal. - 11/11/2005 4:43:58 PM   
Idlemind26@yahoo.com

 

Posts: 8
Joined: 10/27/2005
From: Oklahoma City
Status: offline
Hey Bilbal,
I've been using quite a few of your formulas from the warehouse. Some good ones in there! At least they have your name in them, so I hope their yours. Anyways, I was curious if you think it would be ideal for me to never tune them to prevent over optimization. Looks like some have been tested over real long periods of time, so I would imagine they would be fine. Any advice? I feel more comfortable leaving them the way they are... thanks! Randy

_____________________________

**Fear the Clown**

(in reply to Idlemind26@yahoo.com)
Post #: 7
RE: Another question for Bilbal. - 11/12/2005 2:12:26 PM   
bilbal

 

Posts: 147
Joined: 11/24/2003
From: Harrisonburg, VA
Status: offline
glad I could help. yeah, I tag formulas with bilbalS (sides) and bilbalT (totals) so I know not to change/optimize them during the season. If you want to re-tune just rename it to something else so you don't lose the original settings. Can never hurt to tinker

good luck!

(in reply to Idlemind26@yahoo.com)
Post #: 8
RE: Another question for Bilbal. - 11/15/2005 11:14:56 AM   
kidnee34

 

Posts: 2
Joined: 11/15/2005
From:
Status: offline
hi bil how are you can you explain what the formula uses to make perdictions on nba games and football games if u know thanks

(in reply to Idlemind26@yahoo.com)
Post #: 9
Page:   [1]
All Forums >> [Sports Handicapping Resources] >> WinPicks Software Forum >> Another question for Bilbal. Page: [1]
Jump to:





New Messages No New Messages
Hot Topic w/ New Messages Hot Topic w/o New Messages
Locked w/ New Messages Locked w/o New Messages
 Post New Thread
 Reply to Message
 Post New Poll
 Submit Vote
 Delete My Own Post
 Delete My Own Thread
 Rate Posts


Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 Unicode

0.024