I’ve taken a statistical look at the Cheltenham Festival for the last three years, and each of those articles will still be relevant to this years festival. They are mostly based on eight years data, so rather than refresh the same angles each year, I try to find some new ones, or refresh ones from a few years ago. This year with the help of Proform software I am going to take another look at trainers, and with the help of a valuable new stat that software provides, I am going to delve a little deeper than I did in the past. You should also read the stats articles from 2018 and 2017 as they have different angles covered.
The problem with breaking data down to look at more specific cases, especially when we our dealing with just a yearly four day meeting, is the sample becomes too small, and often used stats like win percentages become almost meaningless, as they will have a big degree of noise in them. I normally do my research in excel and I import Betfair Place Sp’s to come up with Actual/Expected (AE) figures based on those, as although we are trying to predict winners at Cheltenham, historical Place AE is a much better predictor than Win AE or strike rates, when the samples are small. Percentage of rivals beaten (PRB) is another useful measure, which can show up luck in win strike rates, but it’s flaw is it regards the difference between first and fourth in a ten runner race, the same as the difference between seventh and tenth, were as in reality the places are both more valuable for punters, and also for jockeys, who normally won’t care if they came 15th or 20th, in a 20 runner race, but will class the difference between 1st and 6th in the same race as very significant, yet PRB will regard both as the same.
Proform has recently introduced a new stat, percentage of rivals beaten squared(PRB^2), and this is a better predictor of future events, it is still a robust measurement, but has the effect of weighting differences in placings in the first few positions higher than the last few positions. To be precise the name on the stat is slightly misleading as rather than squaring the percentage of rivals beaten, we are first converting it into a range between 0-1, so winning would be 1.0, beaten half of your rivals would be 0.5 and so on. It is the 1.0 or 0.5 that is squared, and the result turned back into a percentage. For an example if one trainer had a winner and a last in two eleven runner races, he will have a PRB^2 of 1.0+0=1/2=0.5 or a PRB^2 of 50%, were as if another trainer has a runner that comes 6th in both races, his PRB2 will be 0.5*0.5=0.25 for both runners, so his PRB^2 overall will be just 25%. I think this measure captures more information than just using PRB which will be 50% for both trainers. Since an awful lot of the trends analysis concentrates on just winners, finding trainers that have performed much better than the win rates alone would suggest, can lead you to find some profitable betting opportunities, and some trainers to avoid.
All Races
I’ve used the last 8 festivals for all the research, as going back any further could lead to the data being less relevant, but you need to go back that far to get a half decent sample. Most of the fields above are self explanatory. I’ve already explained what PRB and PRB^2 are. IV is Win Impact Value, and is more valuable that win strike rates as it accounts for field size. So if a trainer had 10 runners, all in 10 runner races, and won 2 of them, the IV would be 2.0, as he won twice as often as field size would dictate, as you would expect just 1 winner from the 10 runners. Win and Plc IV (Place IV from now on) is the same expect for this time it includes places. Ex Wins is the expected number of winners based on the horses Betfair SP, and A/E is the Actual amount of winners divided by expected, so anything over one would mean a profit if you blindly backed them to return a set amount. Chi Sc is a Chi square based statistic that measures the significance of the AE figure, and anything over 2.0 would be reasonably significant. The above data is for all Cheltenham races and I only include trainers with at least 8 places.
Sadly Dessie Hughes is no longer with us, but I left his data in as it is a very good example of the folly of using solely win rates when the sample size is small. With 1 winner from 28 runners, his AE figure is just 0.71, indicating you’d have made a loss backing all 28 horses, but his Place IV and in particular PRB^2 stats show that his horses were in the top echelons of the field quite often, and having just 1 winner was likely due to bad luck. Of the current trainers it’s notable just how much Gordon Elliot sticks out, he has had 22 winners, when only 11.61 would be expected given their prices, so you would have made a 90% ROI blindly backing all of them. As the Chi Sc stat shows this is hugely significant, and the win rates are backed up by a very high PRB^2 stat, as well as Place IV. It should be noted Place IV will rarely be as high as Win IV for very positive returns, as the higher the chances of something happening, the harder it is to double the frequency.
Noel Meade is another trainer who has done much better than a win strike rate of just 5.6% would suggest, as his PRB^2 is higher than Willie Mullins who has a win strike rate of 12.4%. Nicky Henderson and Jessica Harrington is also a good comparison, with their PRB^2 rates almost the same, but Jessie has almost double the win rate. I would expect them to have around the same win strike rates in the future. Both Jessie and Tony Martin have had much more winners than expected, although the PRB^2 stat would suggest they are unlikely to maintain the very high AE figures, but that’s not to say they won’t continue to be profitable to follow, just not quite as big a margin as in the past, were as although blindly backing Nicky Henderson horses in the past has led to a small loss, his overall distribution suggests you’ve probably got a bit unlucky with the amount of winners.
Jonjo O’Neill and Colin Tizzard both have a positive AE the past eight festivals, but their relatively low PRB^2 figures would suggest they may not maintain that in the future, while both Donald McCain and Venetia Williams do poorly on almost all metrics.
All Chase races
The above screenshot is for chases only this time, and I include any trainer with at least 15 runners over the last 8 festivals. Once again I sort by PRB^2 and Noel Meade comes out on top. Given so many people concentrate just on winners, it’s reasonable to assume his horses could be under-bet next week, as the more robust measures clearly indicate they have outperformed market expectations, just not in the win/lose category. Gordon, Tony Martin, and Henderson again do very well in the chases, while Gary Moore is another very interesting one, as while you’d have made a small loss backing his chasers to win, his PRB^2 and Place IV stats suggest they have performed much better than that flimsy stat alone would indicate.
No matter what metric you use, Paul Nicholls chasers have performed poorly in the past eight years at Cheltenham, and while Rebecca Curtis has had 3 winners when only 0.72 would be expected given the market price, her very positive AE figure doesn’t look very solid when you look at the more robust measures. Charlie Longsdon, Venetia Williams, Donald McCain and Sue Smith are other trainers whose chasers just haven’t performed over fences at the meeting.
All Hurdles races
This time we look at just hurdles, again only including trainers with at least 15 runners in the last 8 years. Gordon again comes out very well on all metrics, while despite a negative AE figure, it’s clear that Harry Fry hurdlers have done very well, and you can expect him to improve that win strike rate in the coming years. In contrast to his record with chasers in recent years Paul Nicholls has done very well with his hurdlers, which is surprising given he is best known for training top chasers like Kauto Star and Denman. Neither Dan Skelton’s win or place IV is overly impressive, but he still shows a decent profit with 2 winners when only 1 was expected, and his high PRB^2 figure relative to his Win IV suggests there’s no fluke about it. Alan King has a very low AE and Win IV figures, but his Place IV and PRB^2 figures show this to be more of an unusual distribution issue, rather than his hurdlers under-performing at the meeting. David Pipe is another with a poor AE, but in his case the other measures back this up, while Venetia Williams, Warren Greatrex, Charlie Longsdon, Evan Williams and Tim Vaughan all do poorly with their hurdlers.
Novice Chasers
The requirement to have at least 15 runners mean only 4 trainers qualify in the novice chase department. Willie Mullins has a much higher win strike rate, and AE than Nicky Henderson, but their PRB^2 figures suggest they will be much closer together in the future, with Nicky’s novice chasers running better than his 9% strike rate would suggest, and Willie’s not as dominate overall as a 21% win rate would make you believe. Paul Nicholls again does poorly on all fronts.
Handicap Chases
Gordon Elliot’s AE for handicap chases is 1.12, indicating a small profit blindly backing all of them, but the Place IV and PRB^2 figures suggest his win strike rate would likely increase if he maintained those. David Pipe has done very well with his handicap chasers, while Tom George’s have run much better than a 0 from 19 return would suggest. You’d be well down backing Nicky Henderson’s chasers in Cheltenham Handicaps but his PRB^2 and Place IV stats suggest that may not continue. Charlie Longsdon, Venetia Williams and Paul Nicholls horses perform poorly on all metrics, while a profitable AE figure for Colin Tizzard camouflages the overall poor performance of his handicap chasers.
Novice Hurdles
Gordon seems to do well in every type of race, with the above showing novice hurdlers. Once more Nicky Henderson’s have performed better than his win strike rate suggests, and despite the fact his Win IV is much lower than Willie Mullins, the PRB^2 figure is a better predictor of future win rates, which is useful as many punters will use the historical win rates instead.
Handicap Hurdles
Dan Skelton has had 2 winning handicap hurdlers, when only 0.79 was expected, in the past 8 festivals, and while the Place IV figure doesn’t back that up, his high PRB^2 does, indicating he had a good few runners who ran very well, and just missed the places. Gordon again does extremely well on all measurements. Paul Nicholls handicap hurdlers do an awful lot better than his handicap chasers, with his 46% PRB^2 figure double what the chasers achieve, and while Alan King comes out poorly on win stats, his other ones suggest this is just an anomaly. You would need a good reason to back Charlie Longsdon, Venetia Williams or Evan Williams in a handicap hurdle next week though, as their runners in the past eight years haven’t troubled the judge much at all.
I hope you found something of value in this article. It’s pretty clear Gordon Elliot comes out extremely well on nearly all metrics, and while the market will likely continue to cotton on to that, with the result that his AE figures will get squeezed, you should always take each horse on it’s merits anyway, and it’s clear from the very valuable PRB^2 information above that Gordon’s horses, as well as plenty of others I’ve noted, improve a good bit on what they’ve done before in the Cheltenham handicaps. You can use this to help you gauge what price a horse should be, and if there is any value in backing it.