August 8, 2006 at 21:07 #2851alan1Member
- Total Posts 167
I don’t think there is any value in giving a horse a rating e.g.sayiing that you rate Horse A as an 80, as it all depends on the conditions of the particular race. The horse might be an 80 on soft ground, but a 90 on firm!
However rating individual past performances is a different matter. By doing this you can see under what circumstances the horse rated particularly well or badly, and then compare those circumstances to todays conditions to make your assessment of its chances.August 8, 2006 at 21:19 #75009
There would be a number of things I would want to know before deciding what’s value
- Are the ratings their best rating or a last time out rating. Eg a horse may be rated 80 on its debut run but on its four runs since it has run to 36, 38, 35, 31. I would be highly doubtful it will run to 80 again so its rating isn’t really helpful.<br>
- Is/ are the horses progressive or exposed. If the horses are all exposed and fairly consistent then the ratings are clearly very useful and the "value" is easy to spot. If the horses are progreesive you are guessing a little on how much improvement there is and the scope for improvement has to be factored in to the price, eg a horse rated 85 after 10 runs is 2/1, a horse rated 80 after 1 run is 10/1, I’d prefer to be on the latter.<br>
- Are conditions ideal? Has its best rating come over the distance and going or is it a soft ground specialist? Horses running under unsuitable conditions can usually be ruled out, or the price must be very big to arouse interest. And finally, is the horse stepping up in trip. Eg a Galileo- Imagine colt stepping up to 12f from 8f could really leave his old rating behind.<br>
<br>So imo ratings are a tool that can seek value and even make your selection for you, but particulary with 2yos a little caution is needed as sometimes they are rendered irrelevant. I’ll get shot down EC but that’s my personal experience.;)August 8, 2006 at 21:20 #75010
Oops a bit wierd with the bullet points there:oAugust 8, 2006 at 21:25 #75011stevedvgMember
- Total Posts 1137
However rating individual past performances is a different matter. By doing this you can see under what circumstances the horse rated particularly well or badly, and then compare those circumstances to todays conditions to make your assessment of its chances.
Yeah, you have to take into consideration today’s conditions.
(possible exception is with unexposed horses who haven’t had time to show their preferences – but even that is questionable)
Can an accurate forecast be made from ratings…and is there any point?
I read an article in Smartsig a couple of year’s ago which suggested a method of calculating probabilities from ratings using standard deviations.
It was interesting stuff though I seem to remember the calculation involved in even just a 10 horse race was pretty damn complicated.
SteveAugust 8, 2006 at 21:30 #75012
I developed a spreadsheet which converts ratings into prices and with reasonable results. Basically the way I use it is utilise the spreadsheet to arrive at a 100% book based on the ratings (I used RP ratings but Timeform or any other reasonable set would do) and then compare with the actual market. The credentials of those horses priced higher in the ‘real ‘ world Ãƒâ€šÃ‚Â are then examined in detail.<br>Most occasions there are good reasons for the discrepancy (horse out of form, wrong ground, etc) but it is surprising how often one turns up where it is a big price for more dubious reasons (e.g. talking horse in the race, poor recent runs but where they can be explained by fitness/ground, poor form figures but where the horse hasn’t run badly at all, other horses tipped/napped/pricewised, etc). <br>Not foolproof and needs a fair bit of work (at least half hour per race, more if a likely candidate turns up which needs detailed Ãƒâ€šÃ‚Â examination) but a good starting point. <br>Obviously working from ratings alone has many shortcomings but I use it simply as a basic tool to highlight potential value. After it’s done that I use other methods to examine whether the ‘potential’ value is ‘actual’ value.
Steve – I’d be interested in looking at that Smartsig article if you still have it. I laboured long and hard to find a way of converting ratings to prices and eventually managed it using a constant in a formula. It took me ages and ages to discover my missing link (the constant) and it was a real eureka moment when I blundered upon the idea of using one. The basis for my theory is to assume that each horse will run to a certain level (it’s rating) plus or minus a certain constant. Once you know all the possible finishing positions (assuming each horse will have a finite number of possible finishing positions derived from it’s rating) it’s relatively easy then to establish a probability for each horses chance of winning (based on it’s rating relative to the others).<br>An inexact science but a useful tool nevertheless.August 8, 2006 at 21:39 #75013
OK EC – give me a few minutesAugust 8, 2006 at 22:04 #75014
80 BERTS MEMORY 4.41/1<br>76 GRAND ART 6.31/1<br>74 SEATON SNOOKS 7.4/1<br>73 AITUTAKI 7.98/1<br>72 TAWNYBRACK 8.6/1<br>71 BY THE EDGE 9.27/1<br>70 CEDARLEA 9.99/1<br>68 RONNIE HOWE 11.63/1<br>60 IRISH RELATIVE 25.31/1<br>60 FLYING VALENTINO 25.31/1
Sorry about the decimals.
Based on your ratings and the current RP forecast I’d be looking closely at the two Easterby runners (Aitutaki and Seaton Snooks) and also at both Cedarlea and By the Edge. On a statistical note the horse mathematically representing the best value based on your ratings and the RP forecast is By The Edge.
I note however that certain of these ratings are for horses who haven’t run so i guess this is all hypothetical!August 8, 2006 at 22:24 #75015
EC how do you rate Tawnybrack because on my unscientific method I have it well ahead?August 9, 2006 at 06:46 #75016Andrew HughesMember
- Total Posts 1904
Excellent thread, just the sort of thing that marks this forum out.August 9, 2006 at 07:20 #75017Andrew HughesMember
- Total Posts 1904
Theoretically, I suppose if you were interested in probability and therefore accurate prices, you would need to compare every horse with every other horse. So a five horse race would be reduced to a series of 10 2 horse races. Accurate ratings would give you a means of doing this, as Steve and Cormack have already alluded to.
Then again, given that anyone’s ratings are subjective, I suppose you would have to weigh up whether the extra time turning ratings into ‘accurate’ prices would be time well spent. I think there may be a danger in that expressing opinions numerically gives a false impression of statistical objectivity, when even the best ratings are no more than opinions. Is an opinion really expressable to two decimal places? Are we in wheels on a tomato territory here? If your ratings are good and you are satisfied they take into account all possibilities, a rough ranking of horses might well be enough to spot the clear value.
Ultimately though, only practice will tell. Maybe running the two approaches together over a period of time will give you some indication of which is the most effective. Perhaps other forumites who create ratings might like to do the same.
Without wishing to divert this into a ratings-compiling thread, or God forbid, a debate on the merits of ‘value’, I have found one of the problems with a ratings approach is that too little emphasis is given to the way the race might pan out. I am (very slowly) trying to devise a way of working out rough probability from trying to predict how the race will develop, what is the liklihood that each horse will take the early lead, for example. Others are much more advanced in their study of this and this is probably where sectional times come in. To be fair, EC has already alluded to the fact that such considerations are already built into his ratings to a certain extent.August 9, 2006 at 07:32 #75018dave jayMember
- Total Posts 3386
These are my prices (decimals) for these ratings. I think the trick to validating your ratings is that your favourite should not be too far away from the forecast favourite, even if its a different horse, or you are probably wrong. The reason for this is that races tend to be competitive or not. If favourites win 27% of all hcp races and 36% of non-hcp races then it stands to reason that your price (favourite) shouldn’t be too far away from that. As I havent read a paper yet and I’m not familiar with this race I have priced the favourite at 3.8 (26%) and calculated the others back from it.
BERTS MEMORY .. 3.80<br>GRAND ART .. 5.37<br>SEATON SNOOKS .. 7.09<br>AITUTAKI .. 9.15<br>TAWNYBRACK .. 9.54<br>BY THE EDGE .. 12.64<br>CEDARLEA .. 15.56<br>RONNIE HOWE .. 23.03<br>IRISH RELATIVE .. 228.48<br>FLYING VALENTINO .. 228.48
More generally, I think working out prices from ratings is a pointless exercise unless your ratings are good.Its simply rubbish in – rubbish out. Saying that, the spread of these ratings look okay to me, well done EC.August 9, 2006 at 08:43 #75019DroneParticipant
- Total Posts 5123
EC’s ratings illustrate how difficult it is to convert ‘RPR/Timeform type’ ratings into a meaningful tissue. The problem being a narrow band – 20 in this case – over a range of 60-80. As Steve and Cormack have mentioned determining a method/constant that results in a workable tissue requires an awful lot of blood, sweat and tears, and it’s something I gave up trying to master sometime ago. Didn’t spend that long trying to master it, truth be told.
Being a mathematical simpleton, nowadays I convert base ratings (Timeform) plus my own input into an ‘adjusted rating’ scale of 0-30 and employ an easy-peasy sum and divide to arrive at a base 100% tissue. This greater spread – percentage wise – allows an oddsline to be generated with a suitibly ‘sensible’ range of odds.
Applying such a simple/idle method to EC’s ratings results in a ludicrous tissue of 7.8/1 for the top-rated (80) and 10.7/1 for the bottom rated (60) i.e. the oddsline is far too constricted and patently unworkable.
Not saying for one moment that my method is best – I’m sure it isn’t – but it’s a KISS system that works for me, and of course it’s the development and evolution of the input, in addition to the Timeform ratings, in order to arrive at the adjusted ratings that has burnt the midnight oil, not the conversion into a tissue.
A simple rank-to-decimal odds ‘sum and divide’ system some may enjoy tweaking is the ‘Oz Tissue’ method:
BTW has Smartsig died? The magazine is no more and the new website seems virtually moribund. <br>
(Edited by Drone at 10:04 am on Aug. 9, 2006)August 9, 2006 at 08:58 #75020MauriceParticipant
- Total Posts 355
There is an Aussie website that automatically converts ratings into odds. However, it was sent to me in confidence therefore I’m not at liberty to pass it on. You might be able to Google for it.
(I don’t use it myself as I tend to concentrate on certain types within certain races. If they’re not favourite I assume they’re value. If they are favourite it’s a question of whether I take the price. My ‘types’ win about 40% of the time so if I can get better than 6/4 (it might involve more than one selection at combined odds) I’ll probably back it/them. Sometimes a single selection stands out at fantastic odds, eg Strategic Mount and Road To Love (twice) recently.)August 9, 2006 at 09:57 #75021stevedvgMember
- Total Posts 1137
I re-read that article last night.
The basic methodology is as follows:
(strap yourself in)
Imagine a 5 runner race, the runners are A, B, C, D, E.
You have a ratings scale that goes from 0 (dnf) – 100 (pegasus).
You take horse A. You work out, based in his proposed rating for the race and the statitical deviation for your ratings, the probability of him performing at:
0, 1, 2, 3 …. 98, 99, 100.
(this should be a bell curve that centres around his expected rating and the probabilities should add up to 1.000)
Then, for each of these potential ratings, you work out how likely he is to beat horse B.
e.g. if A was to run to a rating of exactly 46 and B’s projected rating is 51, what’s the likelihood of B running 5 (or more) pts below his projected rating? Ãƒâ€šÃ‚Â
(to keep things "simple", I’m going to ignore dead heats in this description of the methodology)
Then do the same with horse C, horse D & E.
You multiply the probabilities up and you now know the likelihood of A passing the post first if he runs a rating of 46 (for example).
You do this for all possible "A" performances from 0-100.
Then you multiply:
the probability A runs rating R Ãƒâ€šÃ‚Â Ãƒâ€šÃ‚Â
the probability that A wins if he runs rating R.
So, if he’s got an 8% chance of winning with a rating of 46 and he’s got a 4.0% chance of running to exactly 46, that value would be
0.08 x 0.04 = 0.0032
Add all the values up for R = 1 to R = 100 and you have the probability that horse A wins the race.
Then you have to do the same for each of the other horses.
You should, if you’ve done your sums right, have 5 probabilities that add up to 100%. Ãƒâ€šÃ‚Â
That’s the theory, anyway.
Hoep this made sense and was of some value.
(Edited by stevedvg at 1:47 pm on Aug. 9, 2006)August 9, 2006 at 10:00 #75022clivexMember
- Total Posts 3420
a new anorak to the first one that follows that :biggrin:
- You must be logged in to reply to this topic.