Australian
There is currently 1 person viewing this thread.
17 Apr 12
01:57
Joined:
Date Joined:
23 Jul 04
I am aware of Harville's work in this area,but find I am uncomfortable with the figures.
e.g 4/1defeats 9/1=39/1
whilst 9/1 defeats 4/1 = 44/1
seems the difference in 'real' terms is not great enough.
Does anyone have a table that skews the weight more in the favour of the more popular result?
e.g 4/1defeats 9/1=39/1
whilst 9/1 defeats 4/1 = 44/1
seems the difference in 'real' terms is not great enough.
Does anyone have a table that skews the weight more in the favour of the more popular result?
sort by:
Show
per page
Replies: 59
|
By:
Give me a couple minutes I have a few things to do.
but Harville is wrong because it biases favorites |
|
By:
ty,I thought his bias was the other way per my example
|
|
By:
http://forum.punterslounge.com/threads/23252-Working-out-betfair-place-odds
The Discounted Harville addresses the inadequacies of Harville. But unless you happen know the Lambda or Rho for your caper, it ain't going to help you much. |
|
By:
Didn't read his link but did that answer your question?
|
|
By:
sort of.i guess i'm looking for a table of probabilities that seems more realistic.
|
|
By:
PM
|
|
By:
No such table of probabilities exists.
Benter's HK lecture illustrated the problem he couldn't solve. The contenders have (possibly different) performance distributions. Even if you knew them there is still a massive computational effort required to get them. |
|
By:
ok thanks for your input.
|
|
By:
The guys in the link are using beta as 1 when that would just get back to ziggys original problem.. I dont think there is that correlation unless its maybe top 2 in the market and the rest of field with zero chance...Or is that correct
If people are using these models to trade then obviously it opens up the place market for scalping by form students |
|
By:
I thought he correct Harville?
Using MLE and he figured 0.82 was the constant p1 = 0.4 p2 = 0.3 p3 = 0.2 p4 = 0.1 p1= 0.4(0.82)/p1(0.82)+p2(0.82)+p3(0.82)+p4(0.82) = x1 x1 * x2 / 1 - x1 or something along the lines of this..any help JFC? |
|
By:
Ziggy you could do an excel VBA short course for 3-4 hundred and I'm sure you could just ask them how to do it or show you how to work it out
|
|
By:
I can send you a spreadsheet if you wish via email.
|
|
By:
If you are using betfair win prices as the basis of your calculations the alpha / beta need to be 1. These variables does the probability to account for any fav / longshot bias (which is almost non existent on betfair)
If you want to do it on Australian racing don't waste your time. Quinella / Exacta approximates even after the race has started aren't remotely close to the final dividends. |
|
By:
Googling for
discount harville formula should give you many hits. The first one for me was that of Alan Woods the deceased mentor of Zeljko Ranogajec. Who should make any number of shysters richer when chomping into his billion dollar estate, with dubious tax implications. Anyway if you trawl through the hits perhaps something might enlighten you. |
|
By:
If you want the spreadsheet using Normal Distribution to figure it out just PM me
|
|
By:
Dark Target,
even if you have Betfair Odds (very close to True Odds) your coefficients are not 1. Something like 0.89 and 0.69 from memory. But these vary depending on the underlying performance distributions. |
|
By:
Believe me, I've tested it with everything form .01 to 1 and the closer to 1 the better the results (for me).
|
|
By:
Either way, you won't have any joy doing it on our markets.
You could box the 2 most "profitable" runners at start time and get a return of about 9% without even worrying about the quinella odds. The quinella odds seem to be inextricably tied to the tab win odds which are horribly inaccurate... |
|
By:
I've also given this subject extensive attention.
When a favourite wins he uses up many above average performances. Therefore when it does not win it is competing with below average performances for a place. Making his place probabilities below the Harville estimates. To my knowledge for every continuous unimodal distribution in existence: The favourite will underperform for 2nd. |
|
By:
DT
LoL and we know why that is. I can see you shaking your head as you type. P.S.You getting itchy fingers yet ?? I suggest you've already made bookings for Dec. |
|
By:
ok i follow you dt
would building two models work? 1 for the actual probability based on win odds and 1 based on what tab punters are going to bet on, ie patterns of over betting a standout fav... rather then relying on projected dividend before the race? |
|
By:
Cant wait 8ball, can't come soon enough!
|
|
By:
i'm computer illiterate and mathmatically challenged..so most of what is being said is well over my head.
I do my own markets on greyhounds.Over a long period these markets have proven to be more accurate than other markets in the marketplace,hence I win regularly.My exotic betting is haphazard and based more on instinct ,than on a precise formula,hence my POT is nowhere near as healthy as my win POT.I would like to change this. It would appear that my humanities background has left me a long way behind the eightball,lol. Further the more random elements of greyhound racing (interference means that using win odds only to determine placegetters is fraught with even more danger than the thoroughbreds),insist that any plan must be copious and detailed by nature....perhaps its unachievable |
|
By:
Unclee that is something that interests me greatly, and something ill look into greatly if what I'm doing now ultimately fails.
Predicting where the late money will go and hence what will be over bet in the last seconds. |
|
By:
Ziggy if your markets are accurate, I would suggest you could box quinella / exacta the 2 most profitable dogs in the win market just prior to jump (providing their win expectancy is greater than 1) will be profitable on their own.
Horses / dogs that are underbet in the win market are generally underbet in exotic markets too |
|
By:
agreed DT,but i'm sure that an automatic device would find all overs and as such extract all the value from the market
|
|
By:
But the proportion of the pool on a lot of these QUIN/EXCT markets that gets dumped in the last 60-30-15 secs pre-jump, coupled with the post-jump money means said approxs are more likely than not to be wishy-washy. I think?
|
|
By:
The only problem with all of this is the approximates are really no guide to what the final dividend will pay. Something that looks am overlay quinella, exacta etc... will be overbet by every man and their dog that is using Harville, Henerey, Stern or whoever else's formula at the jump
|
|
By:
Correct balls
|
|
By:
Well I posted Benter's correction on Harvilles
although x1 * x2 / 1 - x1 is p1 * x2 / 1 - x1 Cannot get my ahead around these statistical programs jfc, 0.81 , 0.65 |
|
By:
yes it is wishy washy but that doesnt mean wishy washy is necessarily bad for you
something im sure you could do right away ziggy is have a spreadsheet work out the optimal amount you should invest in exotics before you start eroding your value, markets might move against your position i suppose though |
|
By:
Thats Harvilles nic, and its not corrected in any way
|
|
By:
0.81 and 0.65 were found by Benter
|
|
By:
You will find unclee the biggest overlays prior to jump end up being the biggest underway once final due to the reasons in my 3rd last post
|
|
By:
Yes that's right, they were the optimum coefficients for the Hong Kong tote.
The formula above is the standard harville formula |
|
By:
Uncleee
I have been taking trifecta, using a computer program and my own prices, for nearly 20 years. Over that time, I have found that once the amount you are investing. climbs over 5% of the total pool, you start to dilute your value. |
|
By:
Anyway I've said enough.
|
|
By:
Isn't Harville formula
(1-takeout)*(1-p1)/p1*p2 |
|
By:
Which is a straight formula..
0.4(0.82)/p1(0.82)+p2(0.82)+p3(0.82)+p4(0.82 While the above would require R or a program of the like would it not? |