- This topic has 31 replies, 9 voices, and was last updated 8 years, 5 months ago by
.
- Posts
Interesting discussion points from James Willoughby on handicapping today on Luck on Sunday. James was vehement in his opposition to lee Mottershead’s point that handicapping involved ‘opinion’ stating that opinion shouldn’t come into it. But he then rather let down his own argument by stating that once you had your mathematical rating you had to let ‘experience’ guide you as to how reliable the rating was (experience = opinion really in this case doesn’t it?).
I think I know what James was saying though. The compilation of a form figure CAN be model or algorithm based (he cited Google’s successful algorithms as evidence of what a good model can do for you) but will always, as he pointed out, be an estimate with an in-built error which will vary depending on the quality and quantity of the data. But I’d say any handicapping model, at least in UK or Irish racing, would surely need an opinion based input/modifier or it’d be spouting nonsense at times, would it not?
Personally, I just take whatever the rating is and adjust it myself for the weight difference if there is one.
The Racing Post add to the rating to reflect the weight carried, so if they have 9st on the flat, they add 14 lbs to the rating to take into account that it isn’t 10 stone they are carrying and that they should be capable of a stone higher performance. I think it’s cack handed and confusing, especially when the entire field is often carrying 9st, rendering any adjustment of the figures utterly irrelevant.
I have seen users on this forum, and other forums, get confused regarding ratings that they are not aware have been adjusted for weight. I recall someone crediting one of Willie Mullin’s novices last season with a huge rating because they had used the day of the race adjusted rating. Ultimately, for me anyway, it is easier to just compare that Thistlecrack is 4 lbs higher than the other horse and then allow for the 6 lbs, to get the theoretical 10 lb advantage, rather than ending up calling it 190 and 180
Anyway, I believe Colin Tizzard is full of magic and that is why I opposed Thistlecrack, once I added in my lack of confidence in the horse as a chaser, which posed the question as to whether he was actually up to his best last season, even before the injury.
Each to their own with the figures they choose to work with. I believe in keeping it as simple as possible and that means using a master rating and making my own adjustments for the relative weights if there is any difference to take into account.
There is no simpler way. With respect Steve, your way of doing things is primative. I can remember decades ago changing ratings in exactly that way when using the Timeform Black Book. Working some races out differently to others would be far more confusing to the reader/subscriber. ie When I am studying now for a particular race, I can see every race the horse has run in with both a performance rating for what the horse achieved and the adjusted figure instantly showing the subscriber how that performance rating relates to the race I am studying for. That would be impossible if races were rated differently to each other.
In their Race Passes/Race Card prior to the race, Timeform give the Race Rating up against the horse’s form. It would be pointless for them not to adjust the Master Rating to the race, if they didn’t then us subscribers would need to do it ourselves – including sometimes looking up how the month we’re in and horses age affect the calculation – wasting valuable time that could be spent studying.
Whole point of the Race Rating is so that punters realise immediately how much Timeform believe one horse has in hand or needs to find against its rivals at the weights. ie If the Master Rating had not been changed to a race rating then Thistlecrack would only be 4 lbs different to Unowhatimeanharry, instead of the rightful 10 lbs. It would give a totally distorted view of how much Timeform believe each horse has in hand/has to find with all its rivals.
It’s not their fault that some people mistake the Race Ratings with Master Ratings. Those that make that mistake only have themselves to blame.
Value Is EverythingCan you help me Mike or anybody? In compiling my Grand National website, foolhardily trying to compare performances over the years, I am working on the basis that 1 length = 8 feet and a staying chaser covers about 5 lengths per second. Am I a long way out? I’m finding that roughly 3 lengths per second makes more sense.
All the time I keep on hearing racing professionals quoting 1lb equals 1 length for national hunt racing which is totally wrong.
When Sheldon Kovitz a Harvard maths graduate taught Andy Beyer his concept of speed figures, on the lbs per length scale or points, using linear curve regression he had 3.3lbs per length for 5f, going through to 1lb per length at 2 miles.
Now here is the problem, linear curve regression doesn’t stop at 2 miles equals 1 lb per length, it continues all the way through to the Grand National distance of 4m2f74y(C), on my lbs per length scale I use 0.37 lbs per length, on another note I use a horse’s length as 10 feet which is 5 lengths per second, and makes the National scale 1.85 lengths per second.
Mike.
T
I dont know if we’re on the verge of a data revolution in UK Horseracing, but its fair to say there has never been more data available to punters and that this trend will continue.Crunching evermore large and detailed datasets in more complex ways will of course increase the overall predictivity of horserace outcomes for those who use them, although for a myriad of reasons the largely static oddslines that are a feature of football betting are a long way off in our four legged pursuit imo.
Isn’t the ‘fundamental’ essentially the algorithm arrived at by data crunching and the ‘context’ essentially the tweaking of that algorithm using the intelligence and intuition of the bettor?
More and more and more data becoming freely available means that the myriad algorithms developed by a myriad of bettors will tend to spew out evermore similar oddslines. So applying context to these fundamentals will become evermore important…again…as this is really a turn of the full circle: once upon a time the few who had computer programs had an edge and the data-deprived race analyst of old with just his eyes and formbook was at a disadvantage; now that everyone has a plethora of data, the eyes might have it again
We hear much these days about the progress of ‘Artificial Intelligence’. It’s on its way but ‘Artificial Intuition’…when…perhaps never
I agree lbs per length shouldn’t be as simple as 1 lb per length.
I know Timeform are on record as saying 5f would be over 3 and sometimes 4 lbs per length. 2m+ Flat races 1 lb per length.
But 0.37 for the Grand National seems to be over-cooking it.
When it comes to lbs per length it doesn’t go up equally furlong by furlong. In my experience it goes up in a curve; going up faster at shorter distances and by the time you’ve got to staying distances it’s either levelled out or almost levelled out. ie 2m4f flat races are the same or almost the same lbs per length as a 2m race, where as there’s comparitively quite a big difference from 5f compared to 6f.
Similarly, when it comes to staying races over jumps once you get to 3m4f the curve has levelled out. ie 3m4f races the same or almost the same lbs per length as 4m4f.
2m NH races would be around 1.5 lbs per length, staying races around 0.7 or 0.67
We should also remember that the reason for the difference in lbs per length is stamina. Pace in the race and/or stamina sapping going should imo be allowed for in calculations. Do sometimes wonder whether ratings oranisations over-estimate performances on very soft ground. If dividing winning times with number of furlongs you’ll see that often horses in the Grand National (on goodish going) are running at a faster average speed than the Welsh Grand National (on softer). ie Usually needs more stamina to win the feature race at Chepstow than at Aintree.
Value Is EverythingCan you help me Mike or anybody? In compiling my Grand National website, foolhardily trying to compare performances over the years, I am working on the basis that 1 length = 8 feet and a staying chaser covers about 5 lengths per second. Am I a long way out? I’m finding that roughly 3 lengths per second makes more sense.
I’d say you were correct with 5.
2017 GN – 4m2f 74y = 22,662 feet = 2,833 lengths working on 8 feet per length. Time taken 543 seconds. Therefore 5.22 lengths per second if I’ve calculated correctly.
I calculate the 2001 National as 4.51 lengths per second if the 4m4f is dead on.
From the Betfair Chase thread I see Mike had the race as 0.51 lbs per length. Therefore instead of winning with 57 lbs in hand (1lb per length), it was more like 29 lbs. Therefore Cue Card ran to about a 144 rating given BDM’s Official Rating, going on Mike’s figures.
FWIW, I think there’s a considerable amount of guesswork in NH racing when it comes to trying to rate performances. James Willoughby was railing yesterday at those who said Bristol De Mai in the Betfair was mostly guesswork. He said all that should be allowed for is a greater margin of error. That sounds to me like statistician speak for guesswork.
When fitness, jumping ability, the path followed around the track, rhythm (rhythm rhythm rhythm!) can differ so much in one horse from race to race, there’s an awful lot more guesswork going on than many would admit (or accept).
I have very rarely used ratings as the deciding factor in a bet. I’ll glance at them sometimes if I’m not familiar with all runners in a race but that’s about it.
It’s all about judgement so therefore there is guesswork involved but you would hope through experience they would be there or thereabouts. You are right more guesswork involved in a performance like BDM’s. How are say the Timeform ratings worked out? Is it a few people who rate each performance and then take an average, discussing maybe those where there is a big discrepancy between the highest and lowest ratings? Is info fed into a computer which gives the rating? If I went in to ask would they tell me
There is no simpler way. With respect Steve, your way of doing things is primative. I can remember decades ago changing ratings in exactly that way when using the Timeform Black Book. Working some races out differently to others would be far more confusing to the reader/subscriber. ie When I am studying now for a particular race, I can see every race the horse has run in with both a performance rating for what the horse achieved and the adjusted figure instantly showing the subscriber how that performance rating relates to the race I am studying for. That would be impossible if races were rated differently to each other.
In their Race Passes/Race Card prior to the race, Timeform give the Race Rating up against the horse’s form. It would be pointless for them not to adjust the Master Rating to the race, if they didn’t then us subscribers would need to do it ourselves – including sometimes looking up how the month we’re in and horses age affect the calculation – wasting valuable time that could be spent studying.
Whole point of the Race Rating is so that punters realise immediately how much Timeform believe one horse has in hand or needs to find against its rivals at the weights. ie If the Master Rating had not been changed to a race rating then Thistlecrack would only be 4 lbs different to Unowhatimeanharry, instead of the rightful 10 lbs. It would give a totally distorted view of how much Timeform believe each horse has in hand/has to find with all its rivals.
It’s not their fault that some people mistake the Race Ratings with Master Ratings. Those that make that mistake only have themselves to blame.
I’ve always been a cheeky monkey
Thanks for the good crack. Time for me to move on. Be lucky.
Thanks for the input, folks. The general impression conveyed by this thread is that rating performances is very subjective and there are a multitude of factors to consider. Certainly, that’s the approach I’ve taken with my comparisons of Grand National performances. Its an aspect which forms only a part of the discussion of each GN and the discussion section is in turn merely part of the page for each year’s race. I have not long started the typing up of these pages (then have the indexes to compile, etc so probably a couple of years before launch). Based on your feedback, I’m inclined not to alter too much of what I’ve written by hand.
It appears 8ft is generally accepted to equal 1L. I had considered whether 1L should equate to less than 1lb, and you make some interesting points, but when I looked at how GN RPRs have been calculated from the mid-1980s on it seems generally (distances/weights carried) the RP has used 1=1 so I’ll largely stick to that. The distance of the GN has changed quite a few times but did not change with introduction of the new measuring system, merely the course was measured along an inner line instead of a central line.
The thing about the GN is its almost always been run at a decent pace and everything is trying. Its just a case of exactly how quick, and did the early pace collapse at any stage? The official goings are notoriously unreliable and, of course, no like for like comparison on the day can be made. It can be very tricky to determine the exact shade of ground or even what type of, say, Good to Soft prevailed. So the time of a GN, the actual going and the exact pace are like three separate elements blowing in the wind – none can be used as a foundation upon which to build another. Therefore, common sense has to be used and it must be employed in the context of any previous/future form between finishers in other races/GNs. Also, exaggerated distances due to Soft/Heavy must be taken into account along with ease/closeness of victory (& weight carried). Then, when comparing GNs some years/eras apart, there is what I call general athletic improvement to be considered. All athletes, human and equine, have become stronger, faster, fitter over the years (records are broken); training methods and jockey skills have evolved. How much quicker in, say, 2010 does Horse B have to have run than Horse A in, say, 1940 (similar going/weight carried/ease of victory) to have put up an equal performance? I just hope I’ve developed he correct feel for the merit of various GN performances by virtue of the study and research done for my project.
One thing I disagree with is how ratings services seem to pigeon hole races into certain ratings bands as if predetermined. They think the Gold Cup should produce higher ratings than the GN per se. This is not necessarily the reality. Who is to say the GC test is more worthy than the GN test? Why is it thought that 3 1/4m is on a higher plane (in conferring champion status) than 4 1/2m? Could the opposite be true or each be equal? A prime example is L’Escargot. A dual GC winner, he lost a bit of speed but developed more stamina as he aged. In a golden era, his GN success was roughly a stone better than his GC efforts. Great performances shouldn’t be undervalued because they come over much further than 3m or in a handicap!
It’s all about judgement so therefore there is guesswork involved but you would hope through experience they would be there or thereabouts. You are right more guesswork involved in a performance like BDM’s. How are say the Timeform ratings worked out? Is it a few people who rate each performance and then take an average, discussing maybe those where there is a big discrepancy between the highest and lowest ratings? Is info fed into a computer which gives the rating? If I went in to ask would they tell me
Think I saw somewhere that with Timeform it’s up to one race reader to form the first opinion about the race ratings, Homer. Then (for the big races) those ratings might be taken to a panel for assessment to say yes or no (altering it) but I could be wrong.
From reading their Race Passes, I believe Timeform rated Bristol De Mai (which they admit was difficult to assess) based primarily on the time achieved in comparisson to other staying races at Haydock (no doubt also allowing for any varying pace differences in the sectionals – if there were any)… Especially in comparisson with the handicap chase over the same course, distance, ground and run at a similarly hot pace just half an hour later.
Value Is EverythingIsn’t the ‘fundamental’ essentially the algorithm arrived at by data crunching and the ‘context’ essentially the tweaking of that algorithm using the intelligence and intuition of the bettor?
More and more and more data becoming freely available means that the myriad algorithms developed by a myriad of bettors will tend to spew out evermore similar oddslines. So applying context to these fundamentals will become evermore important…again…as this is really a turn of the full circle: once upon a time the few who had computer programs had an edge and the data-deprived race analyst of old with just his eyes and formbook was at a disadvantage; now that everyone has a plethora of data, the eyes might have it again
We hear much these days about the progress of ‘Artificial Intelligence’. It’s on its way but ‘Artificial Intuition’…when…perhaps never
That’s the essence of what I’m trying to get across across really, Drone. A simple pen and paper model that properly takes account of fundamentals, context and smaller details can be more effective than a million rows of complex data. Warren Buffett swears by it!
Thanks for your earlier reply Ginger. Your correct, something 18 pounds clear on Timeform ratings in more normal circumstances would certainly be shorter than slightly odds against. So the market was “trying” to figure out Thistlecrack I suppose. Your post references knowledge of the trainers generally optimistic outlook regarding his horses, knowledge of how horses have returned from tendon injuries in the past, knowledge of individual horses fitness profiles, interpretation of market movement, a sound estimation of race tactics, etc. All factors very difficult if not impossible for a computer to quantify.
I’m satisfied with my fundamentals, my model’s output gives me a good entry point into a market, but I need to up my game in terms of context and details, because as mentioned earlier I’m presently stuck between hyper efficiency or almost zero liquidity.
I agree lbs per length shouldn’t be as simple as 1 lb per length.
I know Timeform are on record as saying 5f would be over 3 and sometimes 4 lbs per length. 2m+ Flat races 1 lb per length.
But 0.37 for the Grand National seems to be over-cooking it.To explain the maths behind how you calculate lbs per length figures over any distance using linear regression, I used the world record Dirt equation.
**************************************************************************************************************
Linear Regression (y=bx+a), where “Y” is the predicted score.
Correlation Coefficient = .9994
Equation for working out the linear dirt record times.
Y = -5.5846 + 12.292(x) (x = is where you plug in and multiple the distance in furlongs).
Where “Y” is the predicted score.
**********************************************************************************************
Calculate 5f dirt fastest time…
-5.5846 + (12.292*5f) = 55.88s
Calculate 5f lbs per length figure…
55.88s / 0.2 = 279.4 (0.2 = 1 length 10ft)
1/279.4 x 1000 = 3.58 (lbs per length) (1000 = decimal places).
**********************************************************************************************
Calculate 16f.
-5.5846 + (12.292*16f) = 191.1s
Calculate 16f lbs per length figure…
191.1 / 0.2 = 955.5
1/955.5 x 1000 = 1.04 (lbs per length)
***********************************************************************************************
Calculate the 34.3f Grand National distance on the dirt-5.5846 + (12.292*34.3f) = 416.0s
Calculate the 34.3f lbs per length figure…
416.0 / 0.2 = 2080
1/2080 x 1000 = 0.48 (lbs per length)
So the lengths per second over 34.3f on the dirt would be…
0.48 / 0.2 = 2.4 lengths.
*******************************************************************************************************
Please note I used linear regression in this sample, and not lego bricks.
Mike.
The outstanding Tom Kerr joins the debate!:
- You must be logged in to reply to this topic.