|
JCapper Message Board General Discussion
|
| Home |
Register
Log In |
| By | BETTING UNDER THE CURVE | |
| formula_2002 6/27/2008 8:06:39 AM | Originally Posted by formula_2002 ON the pa board I've tried the following on a data base of 100,000 + horses. 1. Determine the Z score for each entry in a race. 2. Find the representive area under the curve for each horse in the race. 3. Using the Z score tables, sum all the areas (it sums to about 35) 4. Divide the area found in item #2 by the sum found in item 3. 5. Normalize the value determined in item 4. The normalized value represents the horse’s probability of winning. ------------------ the above showns profits using the Jfactor, the bris prime fig, turn time and another bris factor. it may show profits for other factors. best results are attained when the value for the area under the curve is >=.95 only no maiden races. race booking percentage to be <=1.25 or there abouts. I would like to know how this works for anyone with a large data base. thanks Joe M
| |
| bettheoverlay 6/27/2008 11:54:18 AM | What is the Z score and where are the Z score tables?
| |
| jeff 6/27/2008 12:28:34 PM | There's a thread on the PA site where deriving an odds line using normalized ZScores was being discussed: http://www.paceadvantage.com/forum/showthread.php?t=41075 I read the thread (post #37) a few days ago and did some R&D. I developed an algorithm that takes a "number" for every horse in a given race and derives a probability using normalized ZScores. Here is the probability distribution created using the normalized ZScores of my own UPR for calendar year 2007: Data Window Settings: Divisor = 999 Odds Cap: None Surface: (ALL*) Distance: (All*) From Index File: C:\2008\pl_Complete_History_07_Z.txt)
to be continued... -jp . ~Edited by: jeff on: 6/27/2008 at: 12:28:34 PM~
| |
| jeff 6/27/2008 1:15:13 PM | The first question I had to ask was: How accurate is the probability expressed by normalized ZScores? Here, I've taken post time favorites, and broken them out by ZScore probability:
As you can see, actual win rate was quite different than expected win rate. It appears that ZScore probability can be used to do a very nice job of classifying favorites as either vulnerable or legit. But contrast that to something else I've been working on, a Tote Prob using a different type of probability distribution based on my own UPR: 2007 Post Time Favorites
I suspect that normalized ZScores can be used in a number of different ways to rethink (and improve) a lot of the things I currently do. But it's going to take some R&D to do that. -jp . ~Edited by: jeff on: 6/27/2008 at: 1:15:13 PM~
| |
| busseb 6/27/2008 4:34:44 PM | I wish I understood what the hell you guys are talking about. Oh well, just put it in an update and as long as I can figure out how to make money with it - who cares!!! ElPaso
| |
| formula_2002 6/27/2008 4:37:25 PM | Jeff,try your top pick only, and your probability is >=track probability*1.5 and the zscore is >=2 for non maiden races. I set the race book % at <=1.25. It's my way of not betting against low priced entries.
| |
| formula_2002 6/27/2008 4:38:00 PM | just google "z scores"
| |
| jeff 6/27/2008 4:55:24 PM | --quote: "Jeff,try your top pick only, and your probability is >=track probability*1.5 and the zscore is >=2 for non maiden races.--end quote Perhaps we should continue this discussion in the Private Section of this message board. -jp .
|
| Reply |
|
Copyright © 2018 JCapper Software back to the JCapper Message Board www.JCapper.com |