# NH Jockey Profiles: Creating Ratings

This is the last article in my series on jockeys and, to close, I have decided to do something slightly different, ** writes Dave Renham**. For this piece I have been number crunching using Excel and reviewing all UK National Hunt Racing results going back to the beginning of 2019.

My plan? To try to evaluate jockey performance in a different way compared with more standard horse racing approaches. The idea is relatively simple: I am going to compare a jockey’s finishing position with their market position. It will be easier to give an example so let us imagine the following set of ten results:

Using these ten results I add up both columns to compare the market position total to the finishing position total. Adding up the market positions (1 + 1 + 2 + 2 + 3 + 3 + 3 + 4 + 4 + 7) we get a total of 30. The finishing positions equal 50 (1 + 3 + 4 + 8 + 1 + 6 + 5 + 3 + 8 + 11) when added up.

Hence the jockey in this imaginary example has arguably performed below expectations as the total of finishing positions is higher than the total of market positions. Now my idea is to give this overall performance a numerical figure by dividing the market position total by finishing position total. In this case we would get a performance rating of 0.60 (30 divided by 50).

I soon realised, though, that I had an issue with non-completions: horses that fell, unseated, were brought down, or pulled up. I decided it made sense to group all such horses giving them a position in last place. Doing it this way each jockey would be affected in the same way creating as level a playing field as I could. Whether this is the ‘ideal’ I am not sure, but it made the most sense to me.

So let’s start by looking at the 20 jockeys with the highest performance ratings. To qualify each needed to have at least 300 rides during the five-year period.

In general, this list contains lesser-known jockeys most of which actually have quite a poor win record. This is illustrated in the table below which shows their overall record in terms of strike rate / returns:

The question that springs to mind, then, is why are these jockeys producing the highest ratings? I believe there are two things in play here. Firstly, if we look at the majority of their each way percentages, they are higher than one might expect given their respective win percentages. John Kington is an excellent example of this with his each way percentage roughly five times higher than his win percentage. As a general rule in racing, the each way percentage is around 2½ times bigger than the win percentage.

And secondly, most of these jockeys normally ride outsiders, and if you are riding the outsider in a field of 10, any result other than 10^{th} will give you a performance rating of greater than 1.00. Clearly, for shorter priced horses that are at the top end of the betting market, it is harder for those to beat their market rank with their finishing position. Indeed, favourites are unable to beat their finishing position – they can only match it should they win, or fall behind it.

I decided therefore that it would be a better idea to create jockey performance ratings within different Starting Price brackets, which would produce a more level playing field. The price brackets I chose to focus on, which granted were somewhat arbitrary, were: 3/1 or shorter, 100/30 to 5/1, and 11/2 to 8/1.

**3/1 or shorter** – let’s consider the shortest price bracket first. I have used a minimum of 50 rides and here are the top performing jockeys in terms of my performance ratings model. Overall, the ratings within this price bracket will look relatively low, for the reasons I mentioned earlier:

It is interesting (and pleasing) to see Charlotte Jones topping the list – she was discussed positively in the two most recent articles in the series. Her record reads a hugely impressive 28 winners from 51 (SR 54.9%) for an SP profit of £28.18 (ROI +55.3%). Also, as the graph clearly shows, she and Theo Gillard are well clear of the rest of the top ten.

I also want to share the ratings of the main jockeys who have appeared in this jockey series as well as some others I’ve mentioned along the way:

It is surprising perhaps to see de Boinville with the poorest rating with these well fancied runners. However, I did some digging, and he pulls up these shorter priced runners much more often than the average jockey (5.6% compared 3.5%). Hence, this looks the most likely reason why he is a significant amount below the rest.

**100/30 to 5/1** – now a look at the middle price bracket. Again, here are top ten jockeys in terms of my performance ratings:

Patrick Wadge tops the list and by a comfortable margin in relative terms. It should come as no surprise therefore that he has been profitable with these runners to the tune of 31 pence in the £ to SP, 45p in the £ to BSP. Specifically, he has had 57 runners of which 15 won (SR 26.3%). Fergus Gillard has also proved nicely profitable thanks to his 26 winners from 104 rides (SR 25%). Profits to SP stand at +£28.65 (ROI +27.6%); to BSP +£38.54 (ROI +37.1%).

Now at look at the ‘main’ jockeys again – McMenamin featured in the top ten above so is not included again – he would have led this list and by a comfortable margin:

Again, we can see that Nico de Boinville is clear at the bottom. His strike rate with these runners has been 14.8%, with SP losses of over 28 pence in the £. Compare that with the average figures for all jockeys where the strike rate is 17.2% and losses are only at 12p in the £.

**11/2 to 8/1** – onto the final price bracket into which we will look in detail. Once more I have collated the jockeys with the top ten ratings:

Two ladies top the list, with Charlotte Jones appearing again, this time in second place, and once again she has proved profitable to the tune of 19p in the £. Tabitha Worsley tops the pile, though, with an impressive 1.04 figure.

The shame for Tabitha Worsley is that she gets limited opportunities on better horses. 65% of her rides in the past five seasons have been on horses priced 16/1 or bigger; 44% have been 33/1 or bigger. If we consider her overall record on horses from the top three in the betting, they have essentially broken even to SP and, to BSP, have seen returns of 12p in the £.

A look now at the ‘main’ jockeys with horses priced 11/2 to 8/1:

It’s a familiar story for Nico de Boinville, whose allegiance with Nicky Henderson means almost everything he rides is over-bet; while Danny McMenamin again tops the list, continuing his decent ratings performance across the board.

**Higher prices** - I did briefly look at other price brackets and here are a few headlines:

For horses priced **17/2 to 11/1**, once again two female jockeys had the highest ratings – Emma Smith-Chaston was top with 1.15, while Lilly Pinchin was second with 1.14. Danny McMenamin scored well again, with 1.02, topping the main group of jockeys. Nico de Boinville was not bottom of the main group this time with his score of 0.83; that dubious accolade went to Harry Cobden who was on 0.82.

Looking at the **12/1 to 18/1** bracket Paddy Brennan was third out of all the jockeys with an excellent 1.21 rating. Meanwhile, de Boinville was second worst out of ALL jockeys with a figure of 0.92. In the 20/1+ group, Tabitha Worsley and Danny McMenamin both appeared again in the top ten scoring 1.23 and 1.22 respectively.

It seems abundantly clear from these figures that Nico is hugely over-bet by the wagering public and Danny McMenamin is still vastly under-rated.

*

This article has hopefully fuelled some food for thought amongst readers. In order to make money at horse racing one needs to have an edge over the ‘crowd’. Ideas like these performance ratings have the potential to give us that edge.

While I am not remotely suggesting that these numbers are the holy grail of jockey ratings, what I am clear about is that if we don’t test out new theories or ideas, we are probably going to bet in very similar ways to everyone else. That is unlikely to give us the long-term edge most punters dream of.

Before I finish, this type of idea could be used to create horse performance ratings or trainer performances ratings, too. I might look into either or both in a future piece.

- **DR**

hello its many years since i found matt he is the best i ran my own tipping service a few years ago and always found matt excellent mel scott (itsbeat)

Many thanks Mel, that’s very kind indeed.

Matt

Hi Dave

Tried this process Dave,, it’s good thinking but the better jockeys are overbet when looked at comparative ranks like this so end up with lowly ratings. Two things contribute quite largely to a jockeys strike rate – jockeys on horses who have had > than one ride and if its placed last time or not (R2 0.778 – Linear regression) and an above average strike rate for first time debutantes (R2 0.694 – Linear regression) – the first one is just common sense as overall races since 2011 All codes UK & IRE there is a 9.36 delta being placed/not placed, that’s a significant difference – the LR placings using the first 3 (4th place usually consists of 16+ runner hcaps and the sample size is comparatively low if doing this seasonally) can be weighted by the over all strike rates of the placings , so if over ALL runs during the time frame 1st place was 18.96 that’s the weighting you would use , 2nd place would be around 17.94 and 3rd place would be around 14.52(since 2011). For first time starters take the difference between the OVERALL and the jockeys strike rate , so if all jockeys FTS strike rflat ate was 7.82 and a jockey’s FTS strike rate was 12.13 you would use 4.31 for all First Time Starters – Then each place category and FTS totals for each jockey get multiplied by the weights(overall strike rates) , then divided by total runs. Obviously i’m generalising here as Flat , AW and NH would all have different weightings for the placings & FTS but it works very well and can show up over or under rated jockeys as jockey A could have lesser placings and a lesser delta in FTS( as a % of total runs) but a higher rating than jockey B who could have more placings( as a % of total runs) and a higher delta in FTS (again as a % of total runs), it all comes down to the “quality of placings and the weightings , In the above example Jockey A could have more wins( as a % of total runs) hence a higher overall weighting so once the ratings are calculated he comes out on top – worth a look at least , remember though if your doing Flat/AW only , use that combined codes overall weightings. I go further and split the data by Hcaps or Non-Hcaps as the strike rate distribution is different for Non-Hcaps for placed horses between the two on the Flat/AW(combined) with LR 2nd place finishers having a higher strike rate than winners. I also usually use a minimum of a 100 rides for an overall rating. You could also use this method by course , although you would need to look at the distribution of runs first to find a decent prior (cut off point) within the sample size.

The overall rating for the code of racing (Flat, AW, NH) is one of the inputs i use to construct Jockey Ratings using a combination of a jockey’s (or trainer or sire’s) PRB2, WPIV(WinPlace Impact Value), mean BFSP(p) (p=Probability) and the above Weighted LR Placings/FTS Rating.

The formula is (PRB2/Sample mean of PRB2)/(WinPlace Impact Value/Sample mean of WinPlace Impact Value)*(Exponent^0.5*Constant(1.20)*(mean of BFSP(p)/Sample mean of BFSP(p)/LR Placings/FTS Rating/Sample mean LR placings/FTS Rating*(Exponent^0.5*Constant(1.20) – obviously you would have to find the means of the metrics of the dataset you are using depending on timeframe and cutoff point in sample size. If you wanted to rescale the ratings to a certain scale you can use this formula where NR = NEW RATING and OR = ORIGINAL RATING and you use Min-Max – NEW RATING = OR*(NR Max-NR Min)/(OR Max-OR Min+(NR Min*OR Max-NR Max*OR Min)/(OR Max – OR Min) – so if the original jockey ratings came out at 365.15 Max to 35.43 Min and you wanted to scale them to say 125 to 20 , i’ll use a selection of original ratings to show how the formula works

Original Rating New Rating

365.15 125.00

286.10 99.83

134.44 51.53

110.99 44.06

86.77 36.35

68.98 30.68

35.43 20.00

Btw , you can experiment with the Exponents and constants in the combi-metric formula – you’ll find Exponent on Excel or any of its free derivatives like Open Office, Libre Office and Google sheets as EXP – you can also experiment with the metrics – as a Race Rating for LR in Handicaps you could use PRB2 , Field size , Mean OR & Max OR(all LR) , so you find the race means first then use the individual horses metrics , experiment with exponents and constants until your happy with the scale of ratings or re-scale them using the re-scaling formula i put up.

Rob.

PS

If there was an upload function i could upload an example from last season’s UK Flat Turf & AW

PPS – one metric i have found particularly predictive is a horses “Last Race Collective PRB2” – obviously you would need to keep a database for this but say you had Horse A – all your doing is calculating the mean of all the horses PRB2’s that ran in it its last race – you could use career, LR , L3R , L5R or L10R (<=) then for its last race calculate the races mean collective PRB2 – (btw if a horse has not had a run use a PRB2 of 22.25) – so obviously you will have current race PRB2 and the formula for combining the two is SQRT(current race PRB2*0.50)*SQRT(LR collective PRB2*0.35) , so your still weighing the current race PRB2 more but adding a "last race strength" function – so for example if your current pre-race PRB2 (however you calculate it determined by racecount , career , LR , L3R, L10R etc) was 46.77 but its LR collective race PRB2 was a lowly 33.66 , then the formula will spit out a PRB2 of 40.13 – You can carry this forward to make a new distribution of PRB2's that incorporates LR "race strength".

One metric i have found quite predictive is BFSP(p)/FS(p) where (p) again =Probability (just 1/BFSP and 1/FS) – the result is a ratio where the baseline is 1.00 (obviously a 6-1 decimal in a 6 runner race = in probability terms is 0.166666667/0.1666666667=1.00 – around 78% of all races are won by horses with a ratio of >=1.00

But it’s predictiveness is in past races – over a years results and just using the values from that year 2022 Flat Turf/AW – the top rated ratio won around 24% of the time. It’s really a measure of expectancy taking into account Price and Field Size. A lot of punters put a price limit on their data studies (for example <=10-1) , well this is a more accurate way of doing that where you can split a jockeys runs by = 1.00. When your calculating it say over a season use the aggregated totals of BFSP(p) and FS(p) and divide instead of the mean of the individual race values. Its another metric i use when calculating say jockey ratings.

One more decent metric is a jockey (or trainer , sire’s) individual Public R2 again using BFSP(p) and Field Size (p) – here the resultant value , the higher the better. For example over the complete 2023 of UK Flat Turf/AW races the public R2 was 0.16518, you then look at individual jockeys and trainers , so William Buick with a total of 166 wins from 733 rides had an R2 of 0.35706, Rossa Ryan had an R2 of 0.25292 , on the other extreme Luke Morris had an R2 of 0.13892 , Tom Eaves had an R2 of 0.0502 – its very easy to set up on an excel sheet – just enter all a variables WINNERS over a season with the following data

Column A = actual Field size (as an integer)

Column B = Decimal BFSP as odds

Column C = 1/Decimal BFSP odds (converts to a Probability)

Column D = 1/Field size (converts to a Probability)

Column E = LOG(Value in Column D)/LOG((EXP(1))

Column F = LOG(Value in Column C)/LOG((EXP(1))

Column G(ROW 1) Enter “R2″

Column H = 1-SUMIF(B:B,”>0″,F:F)/SUMIF(A:A,”>0″,E:E)

Just make sure that if your doing a variable such as a jockey with a higher amount of wins then doing a jockey with a lesser amount of wins , delete the previous variables extra wins or you will get a false reading – take the variables R2 and subtract by the baseline (All wins for the year , 0.16518)

– that’s how much better the betting public “value” him – at least the participants and players on Betfair

Note: RE: BFSP(P)/FS(p) – i call this P/p Ratio – the Big P being the Betfair Probability and the small p being the FS probability also in the above i meant “you can split a jockeys runs by =1.00” – i have found that jockeys who are still profitable to a reasonable degree after commission (say 5%) with their >=1.00 runners are doing very well indeed and i should have added the publics R2 will self calculate in row 1, column H next to column G where you should have the heading “R2” – Make sure you calculate the baseline first , using ALL winners and ALL Field Sizes for the timeframe analysed.

And in this day and age i should have added “her” as well as him. There are some great female riders. Young Charlotte Jones for example – her numbers are fantastic

Charlotte Jones 2016 to current – UK NH Races

Runs 382

PRB2 39.24

WIV 1.59

Win% 19.11

WPIV 1.12

WAx 29.08

WAx500 38.06

mean BFSP(p) 0.11497

mean BFSP(odds) 8.70

Individ. Horse Count 86

Individ Horse Wins 35

WTR% 40.70

Hcap PRB2 35.66

Mean OR hcaps 109 (rounded)

Chase PRB2 50.85

Hurdles PRB2 36.18

NHF PRB2 50.41

Level Stake Return £1005.97

Level Stake P/L £623.97

Level Stake ROI 2.634(US Style)

Level Stake ROI 163.34% (UK Style)

P/p Ratio 0.9587

CHI 21.62

*Market metrics include 2% commission/std 40% Disc.Rate to BFSP

For a jockey whose P/p is below 1.00 she is certainly outriding her odds, her A/E is around 1.66 which is a big advantage – would love to see her ride for some bigger volume trainers though but she certainly knows James Moffatt’s horses.

Dave

Don’t know if you have read it but C.X. Wong’s book “Precision” has a decent method for rating jockeys which is known as F(x). Its ideal for use with BFSP as there is virtually no overround but he takes the aggregated BFSP’s as probabilities (1/BFSP) but also calculates the variance of each ride (so if the decimal odds were 8 , that would be a prob of 1/8=0.12500, he then uses the formula of (Prob*(1-Prob) so ((0.12500*(1-0.12500) which gives a variance of 0.10938. He aggregates the probs (as Expected Wins) , aggregates the variance and also takes a count of actual wins. He then uses the “NORMDIST” function found on any Excel derivative, which just signifies the “normal distribution” and uses the formula NORMDIST(Actual Wins, x Wins, Sum of Total Variance^0.50,TRUE ) – it produces values between 1.00 and 0.00 and does a decent job of pushing the better jockeys to the top.

Matt

any chance you could follow me on Twitter , its Rob (was 2DeadStattos for years) if you remember

Took a break and went past the 30 day “get your account back” process

I’m currently on there as rob mac123 @Mac123Rob, i’ve followed you and would appreciate it if you could follow me back

Got something i would like to discuss with you.

Rob