August 14, 2019 at 23:16 #1451037LostSoldier3Blocked
- Total Posts 1894
It’s absolutely fascinating and I’m a little bit obsessed with them at the moment. There was a decent introductory piece by James Willoughby in the RP the other day about Machine Learning but he stopped short of saying whether or not they are in the market already. They are!
Quite who ‘they’ are, what data they are crunching, how they have such huge backing and how they carve out a really small ROI over such massive turnover are all hard to grasp. Whenever I feel I’ve got them sussed they do something unexpected. They’re also active at early prices and, from the patterns we see, they have huge influence on the eventual ISP and BSP.
What I would say is that their edge (so far) seems absolutely tiny and out of proportion with the major market shifts they’re creating on a day to day basis. Hence my opinion about SP beat % no longer being a valid way for ‘client profiling’ departments at bookmakers to judge customer accounts.August 15, 2019 at 12:05 #1451051KevMcParticipant
- Total Posts 1157
Very interesting indeed, cheers for elaborating mate.August 15, 2019 at 13:34 #1451054
The R value returned when running a linear regression on all UK racing BFSP’s since 2011 makes for interesting analysis, and backs up LS3’s observations imo.
The R value basically measures the strength of the relationship between a winner and its BFSP. As can be seen here, that relationship is inflating as time passes, and increasingly so since 2017. Also the average returned BFSP of favourites in 2019 so far, is the lowest its been since BFSP was introduced in 2008.August 15, 2019 at 16:48 #1451059Marginal ValueParticipant
- Total Posts 639
Hi Cav. I see that your figures show a slightly increasing trend. But I am struggling to undertstand what it is that is increasing. In a typical linear regression scenario the model usually looks like:
D = C + F(1)*V(1) + F(2)*V(2) + F(3)*(3) etc
D is the dependent variable
C is a constant
F(n) is the regression coefficient of each independent variable
V(n) is the value of each independent variable
and R² is the explained variance
So, for example, in analysing one race where you are trying to assess each horse’s likely performance on the day, for each horse use the following formula:
Horse score = 0.42 * current RPR + 0.31 * best RPR + 0.50 * Trainer’s current RP RTF%
and the data that produced this model also produced an explained variance ( R²) of 80%, then you could know that 80% of its performance is explained by the three independent variables in the formula and 20% is explained by other factors.
So, in producing your list, is the BFSP a single independent variable, with a winner accorded a score of 1, and a loser accorded a score of 0 to indicate the dependent variable? And is your R value a percentage? I should be grateful if you could put me straight on this.August 15, 2019 at 19:20 #1451064GingertipsterParticipant
- Total Posts 26356
Sorry it’s taken me a while to respond, Colin; haven’t been on here for a few days.
Yes, Value is in the eye of the beholder… Value punters can come to different opinions as to what the best value bet in any race is… Although if a value punter believes a horse is particularly good value, it is highly probable a good number of other value punters will also identify it as value. So tbh I didn’t find much to disagree with in that article.
The author makes a good point that risk and reward will be different depending on the individual punter’s outlook. So it’s not all about finding value, at least in that although finding value is necessary the punter’s thinking/mental abilities come in to it. I can understand this completely. Some will be able to just bet on the biggest value, but this does mean a punter will have some long losing runs (which happens to even the most successful punter). I myself have main bets on the horse/s I think are the best value, but also saver bets on what I consider lesser value selections. These (as much as possible) keeps me having winners which is important for my mental state. ie Although cutting out savers and concentrating on the very best value may in theory improve overall profits… In practice savers (plenty of winners) keeps me looking at races in the same profitable manner I’ve always done.
As others have said: Kelly has nothing to do with whether an actual bet is or is not a value/good bet. So this should probably be in another discussion. Think what the author is suggesting – which has some truth – is in order to get the most out of finding value – a good staking plan can maximise returns/profit/value. However, tbh I gave the Kelly stuff a miss.
Kev mentioned my staking plan; which is not Kelly but has some similarities in its objective:
My own idea of the horse’s (percentage) chance…
Minus the available odds (in percentage terms)…
+ My idea of the horse’s (percentage) chance…
So something I believe has a 3/1 (fair 25%) chance available @ 4/1 (20%) is 25 – 20 = 5.
5 X 8 = 40
40 + 25 = 65
Stakes 65 points @ 4/1 = potential profit of 260 points
It would be 72 points @ 9/2 = 324 or 91 points @ 5/1 = 455 etc.
Stakes go up with how much chance the horse has of winning and how much value is in the bet.
(how much a point is worth depends on how much I’ve got in the bank/to bet with)
The author also mentioned racegoers getting value from racing without attempting to find value and having no chance of making an overall profit. But whether a punter is getting value from his/her Racing experience is a totally different subject to betting for value (profit). To get value from the Racing experience a punter does not need to win at all. A Racing pal of mine did not study form at all – or no more than the basic couple of minutes in the race card. Indeed, sometimes selections had nothing to do with the horse at all. One such method was “tottie-form”. His selection being the horse led up by the most attractive stable lass… Don’t get me wrong. Each to his own, not all punters will enjoy “value” punting, there is absolutely nothing wrong with his way of punting/going racing. Compare his activities with someone who loves going to the pub, having a drink and putting the World to rights talking to his mates. The money on a pint is equivelent to my mate’s stake money on the horse. He went racing, made some selections/bets and – unlike the pub or any other leisure activity – actually had a chance of making a profit on the day’s events. He probably got just as much if not more value from his Racing experience as I did, he just was not interested in betting value/for profit…
Yes, Potato is right too “true value lies in the results column”… With the proviso number of wagers involved is important in that statement. Even mug punters go on winning runs now and again and an overall profit can sometimes be just a coincidence. eg If a punter has only made a profit because of one or two multiples or one or two much bigger than their average sized bets and/or at much bigger odds. If profit is only due to winning an accumulator or the punter has bet for a year using £10 stakes and then suddenly has a couple of £1000 stakes bets and one just happens to win @ 10/1. One £10,000 win pays for many loing £10 wagers, so a profit of £6000 would in all probability be down to luck rather than “value”.value is everythingAugust 15, 2019 at 22:00 #1451077
That’s more than a slightly increasing trend, MV.
When you consider the RP Forecast over the same period has an R of 30.7, which increases to just R 31.3 when the Timeform forecast prices are added as a second independent variable, it demonstrates how much additional information is required to increase it up to R 37.7, where it is today.
BFSP converted to its equivalent percentage chance is the single independent variable used in the example. The conversion to percentage is required as the calculated predicted probabilities from a standard linear regression do not deal adequately with non normally distributed data very well in my experience. Non normally distributed is the order of the day when working with the rankings, ratings and odds, found in the vast majority of horse racing data files. The dependent variable is as you mentioned in your opening post.
In relation to the percentage; R squared for 2019 BFSP is 14.2August 18, 2019 at 22:41 #1451332tbracingParticipant
- Total Posts 1342
Value argument stills rages on I see in the absence I have had from the forum. For me value is absolutely essential, I say to any punter starting out learning to price up a race is the most important skill you can learn. But I think in simple terms how Veitch describes in his book as looking for underestimated horses is a way the casual punter can identify with. As small example is I last week tipped a horse to some people I know Diocletian. I recommended him at 3/1 (I put him on a blog too) as I thought he was a 2/1 shot. He went off 6/5, he definitely represented value, unfortunately he didn’t win but was happy I had the value in him. A couple of days later I backed Emenem, he was 7/2 early shows and I didn’t think he represented value at that price, but he drifted to 8/1 by the time of the race. There was 2 gambles in the race and the early favourite took plenty of support still so Emenem was friendless and drifted. I still thought Battle Of Wills was the most likely winner but couldn’t see value in him. I backed Emenem at 8/1 purely because I thought he was value not because I thought he was most likely winner. Luckily this time he obliged all be it in a dead dheat but the concept of betting to value applied. I think you still have to find you have a significant edge to back a runner, I work to a value concept but I am only prepared to back certain horses in a race from my price line where I feel they have a significant edge. The result is you will often watch horses win you feel are too short but applying value you take the patient approach over time. Of course horse racing is subjective and everyone’s opinions differ so where one punter sees value another may not. Lastly I refer to a horse I backed a long time ago (and blogged it) it was a 20/1 shot Trois Vallees at Southwell. A lot of 20/1 shots I wouldn’t be prepared to back as I wouldn’t find much significant about them. On this day though his trainer had strong statistics at the track as well as strong first time out statistics at the track but also significantly he had the USA pedigree which excelled at the track and he was by Elusive Quality who had an outstanding first time record at the track and is generally one of the best sires there. All that pointed as very significant and he wasn’t your run of the mill 20/1 shot I priced him a single figure price but his angles were not factored in and he went off at 16/1 and won. Obviously others didn’t deem he was value but as they say value is in the eye of the beholder and for me he was outstanding value.August 18, 2019 at 23:39 #1451342Red Rum 77Participant
- Total Posts 1257
As I stated before I’m no expert in finding value, but believe that win or lose value must always be value.
The was one time were I believe to this day I think I found the value, and quite proud of it. When BEECH ROAD won the 1989 Champion Hurdle. I backed him at 50/1. Which was also his SP. However he beat at least twice a more fancied horse for that race, and in my opinion should have started at lower odds than the horse he beat.August 25, 2019 at 09:49 #1452156DroneParticipant
- Total Posts 5014
When you consider the RP Forecast over the same period has an R of 30.7, which increases to just R 31.3 when the Timeform forecast prices are added as a second independent variable, it demonstrates how much additional information is required to increase it up to R 37.7, where it is today
I’m assuming that the R value is essentially a measure of Regression-to-the-Mean
So does this imply that SP is becoming more efficient?
That is, it is regressing inexorably towards a zero-sum market with SP ever more accurately reflecting true chance?August 25, 2019 at 20:48 #1452188
R is a measure of strength in the relationship between the independent variable (BFSP), and the dependent variable (1/0), with the R value increasing as the correlation between both variables increases. A linear regression model with an R of 1 (1 is the maximum value that R can be) would indicate perfect correlation between both variables, and would predict the winner in every race.
Yes to the latter imo. As tracking equipment is rolled out at all tracks, and its spin off data becomes increasingly available in the mainstream, its inevitable really.
- You must be logged in to reply to this topic.