Simon Rowlands / Simon Rowlands / 04 August 2010 / 9 Comments

Our 'god of odds', Simon Rowlands, shares his thoughts on judging a horse's price based on its rating

As promised in last week's blog, here follows an explanation of the processes involved in converting ratings into odds.

That there is a relationship between ratings, properly compiled and maintained, and odds is not only to be expected but can be readily demonstrated by taking a large enough sample of the former and comparing them with the latter.

And, as my other half is often keen to remind me, a relationship works both ways, so it follows that there is also a correlation between odds and ratings.

You do not need to use ratings to prompt odds, or odds to prompt ratings, but they both describe one phenomenon - a horse's chance in a race - to varying degrees of accuracy in one number, and therefore, where used, each implies the other.

I should say, however, that, whereas my approach to form handicapping owes a lot to my background at Timeform, I am self-taught where ratings to odds is concerned and would welcome any input from readers with experience in this subject.

I have looked into, or am aware of, a number of ways of converting ratings to odds, including by Monte Carlo simulation (all possible results are established from all horses' past performance ratings) and through Markov Chains (an array of past performance ratings can be expected to lead to other performance ratings with varying probabilities).

The approach described below comes as a result of one correspondent pointing out to me that the rather complicated process I was using was similar to a simpler one described by Don ("the Australian Phil Bull") Scott. This is something of a hybrid of the two approaches.

Create your own spreadsheet using the following instructions: Insert values in green-coloured fields, starting with ratings in A4:A10 (or further down if a bigger field is required).
Insert "power" value in Cell A2 (this will vary for different types of races).
[The default of 0.865 was the optimum least-squares value taken from comparing Timeform ratings with Betfair SPs for a representative sample of Flat racing from last week, but those values varied from 0.775 (handicap) to 0.9575 (Group 2)].
Cell B4 is: =(MAX($A$4:$A$10))-A4, then copy down
Cell C4 is: =(POWER($A$2, B4)), then copy down
Cell C2 is: SUM(C4:C10)
Cell D4 is: =(C4*100)/$C$2, then copy down
Cell E4 is: =100/E4
Insert "confidence" value by-horse, depending on the confidence you feel in that horse's rating (0.167 is a reasonable default).
Cell H4 (to-back price) is: =100/(D4/(1+F4)), then copy down
Cell I4 (to-lay price) is: =100/(D4*(1+F4)), then copy down
You can tie in Columns H and I to Betfair prices by inserting a look-up to the Betfair odds ladder on a separate tab.
The example used is the Guy Salmon Handicap at Sandown on 05/08/2010.

It must be immediately apparent that this process is very susceptible to the GIGO (garbage in, garbage out) principle. No manner of mathematical manipulation will make up for essential shortcomings in the ratings and in the confidence attributed to those ratings.

There is much to be said for using ratings-to-odds merely as a starting point, on top of which you can layer the various other factors which influence the outcome of a horse race. However, there is also much to be said for reducing as many of those factors as possible to numbers in the first place, so that subjectivity is minimised.

In last week's example, I used master ratings which were weighted according to a horse's consistency and the recentness and context of its best form. I also added an allowance for the draw and for the trainer's current form before coming up with a fair odds line.

A horse with an array of ratings (most recent effort first) of: 140/131/128/125/118 (let's call him "Harbinger") is different to one with an array of ratings of: 140/140/140/140/140 (let's call him "Super Harbinger"), and yet a rating system which takes the maximum value would rate both horses 140.

What I would term "confidence" could cover a whole host of factors, not least the identity of the horse's trainer. There are certain trainers whose horses can be priced up with confidence, and there are certain trainers whose horses, well, cannot be.

It is possible to factor in very much more than the above, and some do, but it is a question of how far down the road of automation you wish to go. Hopefully, this illustration has given food for thought if nothing else.

Comments (9)

1. Michael Turner | 04 August 2010

   Hello

   I have been a very enthusiastic follower of 'time' when studying horse racing for several years, and if the ratings-to-odds method increase the probability of gaining an advantage, then it can only be better to find value aswell. This is good news.

   Regards

   Michael Turner.

2. Simon Rowlands | 04 August 2010

   Hello Michael.

   Many thanks. I wrote a series about analysing overall times which is linked to in one of my recent blogs about handicapping. Ratings to odds could, theoretically, apply to time analysis as well as form analysis providing it is expressed in much the same way.

   I would, however, just stress that the accuracy of any conversion of 1 pt into an odds differential depends on the nature of the rating process. If using the above method, it would be best to see what sorts of outputs you get and whether they would be improved by a different "power" value or similar.

   Simon

3. Simon Rowlands | 05 August 2010

   The upshot of the race example above was as follows:

   1 JET AWAY (back greater than 4.03, lay less than 2.96) BSP 3.74: no bet

   2 CON ARTIST (back greater than 5.01, lay less than 3.68) BSP 4.00: no bet

   3 SENATE (back greater than 27.76, lay less than 20.38) BSP 12.5: lay bet

   also ran:

   MORTBET 7.15 BSP: lay bet
   RIGHT STEP 5.86 BSP: lay bet
   SHARAYEEN 12.4 BSP: back bet
   WHISTLEINTHEWIND 25.5 BSP: back bet

   Something of a score draw, perhaps!

4. Michael Turner | 06 August 2010

   Simon

   A draw perhaps, but if you follow individual horses and as a result of following your method above, you may find your selection to be a lay bet or a back bet & therefore bet accordingly. So if each box is ticked (and there are more boxes today than ever before) in your favour you are loaded with lots of ammunition?

5. Stu | 06 August 2010

   Hi Simon,

   I do this intuitively and am interested in automating it. One pressing thought, though re. confidence factors: Why did you choose 1/6? Why is it 1/6 for all runners? Isn't this rather unscientific, i.e. I don't rate Mick Channon as a trainer (I don't like the way he campaigns his horses) so could I give him a lesser rating? But isn't this just my own ingrained prejudices (unbacked by statistical analyses - and therefore ng the quantitative approach?)

   I honestly think pinning this down and getting it right is the key to that over-used word "value"

   Cheers,

   Stu
   

6. Darth | 06 August 2010

   We were 0-0 at half time then they scored 6 or 7 lucky goals?

   Thanks for posting something of real substance on horseracing. If I might be critical but well meaning it is using handicaps for pricing with this method. You get a different answer depending on which ratings you use and any tweaks to the handicap rating get straight into the subjective. If the ratings were to be relied upon to be predictive then the power function should be more similar for each type of race. It is not, presumably, because of the allocated weight carried effects being better to equalise chances in some races than others.
   The BHA ratings for example would give these handicap horses odds of 5/1 each with 117% over-round. Any +/- poundage subjective tweaks have a huge effect on the answer. Particularly if any base rating may only be "accurate" to +/- 5 pounds at best.

   Listing your example the first 3 finishers would need ratings to be tweaked far higher to get nearer the market odds view. Two of the last 3 to finish would need to be tweaked much lower. A practical odds-line needs to get closer to the actual market to actually get bets placed which are more reliably value bets.

   Jet Away 2/1F (3.46 S Rowlands to 100%)
   Con Artist 3/1 (4.3)
   Senate 9/1 (23.79)
   Sharaayeen 10/1 (7.79)
   Whistleinthewind 18/1 (11.03)
   Right Step 9/2 (8.02)
   Mortbet (IRE) 5/1 (10.87)

   Total SP 117%
   

7. Simon Rowlands | 06 August 2010

   Thanks for all the feedback.

   Stu, I did say "insert confidence value by-horse" in the guidelines. I kept it to the default in the example above because in a public environment I did not wish to have to account for any implication that a certain trainer's horse can be priced up with less confidence than another's. And also because I did not want to make an already complex subject even more confusing (though it's possible I achieved the opposite!).

   The 1/6 default was off the cuff, but seemed to "work": you are right that it needs to be arrived at more scientifically if it is a default and more scientifically if (as I suggest it should be) it is specific to the horse/trainer in question.

   Darth also raises good points. You do indeed get a different answer for different ratings. That is why I emphasised the GIGO aspect of things and why I hope people can see the importance of sound methodology in compiling those ratings, which is something I addressed at much more length in my recent series on here about race standardisation.

   Every horse would indeed have the same price in a handicap if you took the BHA's single official assessment. But I recommend above, and elsewhere, considering the array of performance ratings which give rise to a master rating (these are not publicly available with the BHA but are with e.g. Timeform). Assuming those performance ratings vary - as they almost invariably would - then you would get different "weighted" ratings according to consistency and recentness (and perhaps more).

   I also recommend working in other factors where they can be pinned down to numbers, and these "tweaks" absolutely do not need to be subjective. In the previous week's example (linked to at the start of this week's blog) I worked in a strong draw bias and my assessment of the trainers' recent form, both arrived at as objectively as possible. You could do much more besides, though additional variables would need to be weighted appropriately.

   The variance in the "correct" power value is a function of not only the accuracy of the ratings but the types of races being dealt with. Form in handicaps SHOULD be more exposed/established and less open to sudden improvements than form in maidens. Horses in Group races tend to run closer to their ratings than horses lower down the scale.

   Even after taking such things into account, I would expect the power value to vary if the way of judging it was simply by its correlation with Betfair SP. While Betfair SP is remarkably accurate taken over a large sample, it is not remarkably accurate (in terms of horses always running the kinds of races their Betfair odds imply) on a micro basis.

   Ultimately, though, the only way to judge whether differences between your ratings-to-odds line and the market identify your ratings as accurate or inaccurate is to run the odds lines against events enough times.

   Simon
   

8. Wayne C | 31 March 2011

   Simon,

   How are you determining the value of POWER (in this case .865)?

   I have ratings I compile for myself. I have no idea how to determine that value.

   Also, do you know if Timeform is going to be selling ratings for US based stakes horses in the U.S.?

9. Simon Rowlands | 01 April 2011

   Hi Wayne.

   I determined the power as 0.865 through a best-fit calculation. Plot the odds and your ratings and then determine what "power" function would get closest to predicting the odds from the ratings (square root of the squared difference).

   Do this often enough and you should get a value that fits best overall.

   That said, the power function should vary according to race type. For instance, small differences count for more in a race like a Group 1 for older horses, in which all/most of the horses have been well tried and the likelihood of significant improvement is small, than in a race for lightly raced maidens.

   Regarding Timeform, the policy on ratings in the US has yet to be finalised and depends on various discussions we are having with partners in the US.

   I would hope that the ratings themselves (and not just the "star" ratings from 1 to 5 that are displayed for free through the Timeform website) will get a much wider airing, though it may be as a paid-for product. They have been performing very well so far, and these are still just early days.

   Regards

   Simon

Post a comment

Name:

Email Address:

URL:

Remember personal info?

Comment: