It has not been a good decade for the Derby. The last ten winners have included four horses who have met the race’s long-term historical status, but others who won ordinary renewals, benefitted from the run of the race or did not progress.
The TRC Race Ratings computer has no fear or favour, no bias to the past and does not get caught up in the hype as humans can. It merely operates a network of tens of millions of equations which the results of races describe. And, by finding the set of ratings that comes closest to a perfect solution, it assigns merit to the winners of every Group and Graded race run under IFHA standards, plus a select number of others that we deem worthy of a similar status.
Incidentally, it is our view at Thoroughbred Racing Commentary that the sport needs to do a better job at getting performance ratings out there. We live in a world where it is no good publishing figures days, weeks and months after the race – the moment has gone. Performance ratings need to accompany the official reportage of the race.
Computers can make the calculations seconds after a race is run, and even be relied on to flash the rating up on the television screen before the horses have even completed their run-out.
As the season rolls on and more evidence accumulates as to what form is worth, ratings should change to continually find the best solution to the gigantic set of equations: Each race impacts every other one to which it is connected by a chain of common horses. It is time for human handicappers to accept they cannot compete with machines at what is purely an optimisation exercise and to concentrate their efforts to building better algorithms to perform the calculations.
When the TRC computer performs this optimisation technique, here are the performance ratings of the last ten winners of the Derby on the big day at Epsom:
The box around the ratings achieved by Pour Moi in 2011 and Wings Of Eagles in 2017 describes the modern-day median of the Derby winner of 119/120. This would not have been good enough on average 30 years ago, but the race has fallen back a peg or two. Part of this is a random effect, but there is no doubt that some is down to the fact that top-tier middle-distance talent is in fewer hands, which is not good for competition.
This year there is a good chance of a winner to buck the trend.
Yes, trainers Aidan O’Brien and Charlie Appleby, who benefit from the powerful breeding empires of Coolmore and Godolphin, have a number of key runners each, but in opposition are just one or two more credible contenders than has been the case, headed by Mac Swiney.
The dual G1 winner heads TRC Computer Race Ratings on a figure of 119 for his Irish 2000 Guineas success last time out. The TRC algorithm doesn’t much care about the labels on the race (unless, in the very rare cases at G1 level there are sparse connections to other results in the database) but is impressed by what this race suggests is Mac Swiney’s place in the hierarchy of global talent: The Jim Bolger-trained colt beat his stablemate Poetic Flare, who was previously the winner of the 2000 Guineas at Newmarket. And that last race was replete with horses who represented the form of the major 2-year-old races, spreading its influence still further.
Mac Swiney has to prove he stays a mile and a half, but there is an interesting way to view his form. Here are his Group-race performances, arranged in descending order of merit:
Mac Swiney has put up his best efforts on a soft surface and his worse efforts on drier ground. It could be, therefore, that he needs soft conditions to perform. It is also possible, however, that it suits him when soft conditions slow down the pace of the race. He did not appear to be beaten because of stamina when 6¾ lengths fourth to Bolshoi Ballet in the G3 Derrinstown Stud Derby Trial at Leopardstown. Let’s see where the computer rates that winner and the other main Derby contenders in relation to Mac Swiney:
Bolshoi Ballet earned a TRC Computer Race Rating of 118 for his demolition of Mac Swiney and the others in the Derby Trial. He was rated TRC 115 for his previous victory in the G3 Ballysax Stakes. But there has to be a great deal of uncertainty surrounding this rating – on both sides.
Bolshoi Ballet has seemed to improve a good deal on his 2-year-old form, which included a comprehensive defeat by Gear Up and others in the G1 Criterium de Saint-Cloud in the mud at the end of his 2-year-old season. That anchors his numbers this year to some extent, otherwise the TRC Computer might have rated his form this year even higher. It is a matter of finding the sweet spot, which, technically speaking now, is the rating that maximises the probability of estimating his true performance merit in this piece of form.
Interestingly, Bolshoi Ballet’s rating of 118 is just 1pt less than the standard required to win the Derby in a year typical of the last decade – that 119/120 thing again. But there is a lot of variance around what the computer could have rated Bolshoi Ballet in the Derby Trial – and another question of how he will perform in the Derby.
Because the TRC computer uses what is called a Bayesian model, it doesn’t ‘think’ of performances as representing one single figure (though we express it that way for ease of understanding) but instead as a range of possible values. And, if you think of this array in the Derby as a probability distribution over how he will perform, then a lot of the mass is over ratings 119 and above. In other words, the computer is saying he is one of the more likely winners:
But, a similar graph for all the Derby runners would also have at least some mass to the right of the red, vertical line. For the runners more lightly raced than Bolshoi Ballet, the curve (which is ‘normal’ shaped because we normalise distances back from the winner before rating races) would be even wider and less peaked, indicating more uncertainty.
This way, the probability that, say, El Drama or Third Realm wins the Derby is seen by the computer as similar to the chance of Van Gogh winning, even though the latter is rated 4pts higher on TRC 115 (for his G1 Criterium International win) compared to their TRC 111 (for winning the Dee Stakes and the Lingfield Derby Trial respectively which were run as listed races but we graded G3). Van Gogh has simply had more opportunity to earn a higher rating than the other pair.
We would like to be able to specify different shaped curves for different trainers, for it seems likely that High Definition could improve a lot for his G2 Dante Stakes third to Hurricane Lane, if he follows the pattern of many an Aidan O’Brien-trained colt. But we are not at that level of sophistication yet and the exercise is left to the reader’s experience, for now.
Another possible big improver from the Dante is Gear Up, the aforementioned G1 Criterium de Saint-Cloud winner. He is best when forcing the pace and was ridden back at York before keeping on. It is a point beyond the scope of our data, but trainer Mark Johnston’s horses experience a variety of outcomes at York for whatever reason, especially when the ground is on the soft side. Ante-post odds of 33/1 about Gear Up are interesting.
Other big improvers are John Leeper and Mohaafeth. Both these horses qualified for a TRC rating because we upgraded listed races to G3 and included them in our database. John Leeper won the slowly-run Fairway Stakes at Newmarket and had scant opportunity to engage his big, powerful stride to the full. For his part, Mohaafeth hacked up in the Newmarket Stakes and could be anything.
Charlie Appleby trains two fringe contenders who add ballast to this fascinating renewal in One Ruler and Adayar. The former won last year’s G3 Autumn Stakes and was well backed before disappointing in the 2000 Guineas, while the latter is a tremendous physical specimen and may have got bogged down when second to Third Realm in the muddy Lingfield Derby Trial.
In summary, we cannot split Mac Swiney and Bolshoi Ballet in the Derby, as this week’s Horse Rankings aptly show:
Will another horse take a step forward and beat them both, exceeding the 119/120 level? The Derby deserves a good winner.