Amongst all of yesterday's excitement, I picked Sunnyhillboy e/w only due to Peanuts. Thanks PM, a thoroughly captivating thread as always. I think that is 3 years in a row I have collected thanks to your amazing knowledge.
Too kind Chief - pleased to assist but you deserve the credit for picking the right one. Still can't believe he got collared....UGH!!
He was one of 3. Sadly I also picked Synchronised and Chicago Grey, purely because of the name.
re the worthiness of a stats model………..I get the feeling that this year was just an insanely competitive field. I presume that's why you had a number of horses in 'near miss' selections, I'll actually have to ask him if had the same. But such a field is always going to work against a detailed model due to reducing the margin of error. What do you reckon?...........
Sorry to be slow getting back Chief.
A few thoughts re the competitive nature of the race (posted in 2 parts).
PART 1
On the ground indicated by the time (Good or quicker), in addition to the 4 Win candidates, pre-race the model rated 6 to be in the “margin of error” (i.e. “good fit” with the profiles of horses finishing 4~20L behind the winners of the 1988~2011 GNs. The most Win & MoE candidates identified by the model in any of the previous four GNs was 6, so it was a good deal more competitive this year.
Higher-rated horses don’t necessarily make for Win or MoE candidates (3 of the 10 this year carried 11.00+) so, despite the higher “quality” of the field these days, next year’s GN may or may not be as competitive from the model’s perspective.
It is notable that Neptune Collonges and Cappa Bleu (1st and 4th) were in the MoE and it now transpires that Sunnyhillboy suffered a severe tendon injury during the race. So, despite NC carrying the heaviest weight by a GN winner since 1977, I’m totally confident that the result was capable of prediction by a points-based stat-model. The “failure” of mine to do so was down to me not getting the algorithm right, not a systemic failure of the approach.
So, the 2012 race has been analysed and the model has been tweaked and back-tested to verify that the changes hold up historically. Importantly, the tweaks required have been relatively few and affect only a handful of runners in previous GNs – none materially. Also, it has not been necessary to tweak the existing weight stat.
Analysing the first 6 finishers (winner and those <20L of him) and my other Win selections and using Cappa Bleu as the “anchor” point for the analysis, since he appears to be the one most obviously to have run (without significant interference and at a measured pace) to the model’s expectations and be best placed:
1st - Neptune Collonges (model predicted: “margin of error” [level with Cappa Bleu] – finished 12L ahead of CB)
One stat adjustment reconciles this outperformance. He was running in the GN off a handicap mark 15lbs lower than when recording his 7L 3rd to Denman in the 2008 Gold Cup. The relevant pre-existing stat was to record a positive if running off 10lbs+ lower mark than when achieving (less than 3 years prior) a win or <10L place in a Class 1 (or equiv) 3m+ chase with 8+ runners at a major track (viz. Cheltenham, Newbury, Chepstow, GN Course, Fairyhouse, Leopardstown or Punchestown). Party Politics in 1995 and Snowy Morning in 2008 were the only previous tickers of this box. The simple tweak for the model is to remove the time constraint, thereby elevating NC’s stat-score to an 8~12L superiority to CB.
Superficially SHB underperformed by 8L but, sadly, he suffered a severe tendon injury during the race. Given that he was also taken appreciably wide at times, it is reasonable to conclude that he ran more or less as expected. Conclusion: no tweak appropriate.
A very simple tweak to the stats is required to reconcile Seabass’ impressive performance. Pre-race the model penalised him for not having a win (or near-miss) in a 3m+ chase under rules (1 attempt). It’s not the first time since 1988 that a maiden at 3m+ has finished within 10L of the GN winner but it’s rare - Auntie Dot in 1991, Encore Un Peu in 1996 and King Johns Castle in 2008. Unlike Seabass, all had been placed. The tweak to this stat is simply to dis-apply it if there’s been only 1 attempt at a 3m+ chase, providing that a minimum Win Ratio at <3m is met, so as not to open the floodgates (none other affected).
I’d thought I’d have to dismiss IC’s 5th place as an “outlier”, that no stat-model could have predicted. That would have been a disappointment as there was nothing obviously false about his placing as there was with Cerium’s 19L 5th in 2009 when the pace had slowed appreciably during the race.
Amazingly, in reconciling Neptune Collonges’ win, it transpired that IC benefits from the same tweak, only significantly more so. Though without a win in a 3m+ chase in 5 attempts (for which he remains penalised), he recorded a 10L 3rd (spookily behind Neptune Collonges!) in a 3m1f Grade 1 at Punchestown 5 years ago, off a mark of 164. His mark for the GN was 137, 27lbs lower. Together with a minor tweakings elsewhere, including recording a positive for a good run with 11.05 in the Topham 2 years ago, it has been possible to reconcile his stats to this fine run but, importantly, the handicap drop is the key catalyst and, as back-testing has confirmed, it remains a very rarely achieved stat (now 4 among the 240 strong database).
Synchronised (fell at 6th fence): no tweak to model. It would be handy if there were a stat that accounted for the fact that, despite no F or UR in 9 chases, there was particular reason to be concerned about a horse handling the GN fences. He hadn’t enough runs to record a positive for his jumping but, other than F/URs, there isn’t an objective way of reflecting that concern – let alone that he might dump his jockey before the off. For very sad reasons, we will never know whether, barring that fall, his extraordinary stamina would have enabled him to carry top-weight to victory to record a famous double but the fact that the winner carried only 4lbs less and the runner-up was a Festival winner suggests that he might have gone very close.
West End Rocker (fell at 2nd fence): no tweak to the model. If there was one runner that seemed a safe bet to jump round it was Becher Chase winner WER. Despite a scare at the first fence in the Becher Chase he had never fallen or unseated in 14 chases under rules and 6 hurdles. He was one of those that got very worked up with the delay and false starts – indeed he broke the tape and got entangled with it for the second one. This probably explains his early departure, thankfully apparently none the worse for it. You can’t account statistically for his fall and so no tweak is appropriate.
9th (28L) – Shakalakaboomboom (predicted: 10L ahead of CB – finished 16L behind: underperformed by 26L).
A minor tweak has been made as a result of Shaka evidently running out of gas. 8 year-olds continue to have a very patchy record in the GN, though better on good than soft. Those that win or make the places tend to have form at 3.5m+ and so I have introduced a small negative stat for an 8 y-o that doesn’t. Despite Shaka being 100% win & near-miss in 4 chases at 3m+, the furthest was 3.25m. So, retrospectively, he should have been rated “Big Run – No Cigar” rather than a Win candidate and he may well have finished closer if he’d not been ridden quite so close to the pace.
That said, this was a very creditable run indeed - up with the leaders virtually all the way and jumping beautifully (as he did in the Topham a year ago). For an 8y-o to run this well and finish within 30L of the winner is, in itself, a positive stat for a return. It’s reminiscent of Hedgehunter’s first GN in 2004 when, carrying 10.12 (like Shaka), he ran a blinder and was leading for more than a circuit until running out of gas 2 out and falling at the last - he came back a year later as a 9 y-o to carry 11.01 to victory. Assuming he’s sound, another year to mature for Shaka combined with, hopefully, both handicapper and market thinking he doesn’t get the trip could make him a very backable selection next year.
All of these tweaks have been back-tested against the winners and those placed <20L of them since 1988 and the entire fields of the 2008~2011 GNs (240 horses in all). Only 1 of the 80 winners & <20L in the database (Cerium in 2009) remains unaccounted for statistically by the model.
So, assuming that the race isn’t changed out of all recognition, the Winner and next 5 places are now guaranteed for next year ;o)
Have a great summer. Even if I haven’t earned it, Sir Chris and the boys have earned it for us all.
Comments
A few thoughts re the competitive nature of the race (posted in 2 parts).
PART 1
On the ground indicated by the time (Good or quicker), in addition to the 4 Win candidates, pre-race the model rated 6 to be in the “margin of error” (i.e. “good fit” with the profiles of horses finishing 4~20L behind the winners of the 1988~2011 GNs. The most Win & MoE candidates identified by the model in any of the previous four GNs was 6, so it was a good deal more competitive this year.
Higher-rated horses don’t necessarily make for Win or MoE candidates (3 of the 10 this year carried 11.00+) so, despite the higher “quality” of the field these days, next year’s GN may or may not be as competitive from the model’s perspective.
It is notable that Neptune Collonges and Cappa Bleu (1st and 4th) were in the MoE and it now transpires that Sunnyhillboy suffered a severe tendon injury during the race. So, despite NC carrying the heaviest weight by a GN winner since 1977, I’m totally confident that the result was capable of prediction by a points-based stat-model. The “failure” of mine to do so was down to me not getting the algorithm right, not a systemic failure of the approach.
So, the 2012 race has been analysed and the model has been tweaked and back-tested to verify that the changes hold up historically. Importantly, the tweaks required have been relatively few and affect only a handful of runners in previous GNs – none materially. Also, it has not been necessary to tweak the existing weight stat.
Analysing the first 6 finishers (winner and those <20L of him) and my other Win selections and using Cappa Bleu as the “anchor” point for the analysis, since he appears to be the one most obviously to have run (without significant interference and at a measured pace) to the model’s expectations and be best placed:
1st - Neptune Collonges (model predicted: “margin of error” [level with Cappa Bleu] – finished 12L ahead of CB)
One stat adjustment reconciles this outperformance. He was running in the GN off a handicap mark 15lbs lower than when recording his 7L 3rd to Denman in the 2008 Gold Cup. The relevant pre-existing stat was to record a positive if running off 10lbs+ lower mark than when achieving (less than 3 years prior) a win or <10L place in a Class 1 (or equiv) 3m+ chase with 8+ runners at a major track (viz. Cheltenham, Newbury, Chepstow, GN Course, Fairyhouse, Leopardstown or Punchestown). Party Politics in 1995 and Snowy Morning in 2008 were the only previous tickers of this box. The simple tweak for the model is to remove the time constraint, thereby elevating NC’s stat-score to an 8~12L superiority to CB.
2nd (nose) – Sunnyhillboy [WIN CANDIDATE] (predicted: 20L ahead of CB – finished 12L ahead)
Superficially SHB underperformed by 8L but, sadly, he suffered a severe tendon injury during the race. Given that he was also taken appreciably wide at times, it is reasonable to conclude that he ran more or less as expected. Conclusion: no tweak appropriate.
3rd (5L) – Seabass (predicted: 8L behind CB – finished 7L ahead)
A very simple tweak to the stats is required to reconcile Seabass’ impressive performance. Pre-race the model penalised him for not having a win (or near-miss) in a 3m+ chase under rules (1 attempt). It’s not the first time since 1988 that a maiden at 3m+ has finished within 10L of the GN winner but it’s rare - Auntie Dot in 1991, Encore Un Peu in 1996 and King Johns Castle in 2008. Unlike Seabass, all had been placed. The tweak to this stat is simply to dis-apply it if there’s been only 1 attempt at a 3m+ chase, providing that a minimum Win Ratio at <3m is met, so as not to open the floodgates (none other affected).
4th (12L) – Cappa Bleu – (model predicted: “margin of error” [10~20L behind winner] – finished 12L behind winner)
5th (15L) – In Compliance (predicted: 60L+ behind CB – finished 3L behind)
I’d thought I’d have to dismiss IC’s 5th place as an “outlier”, that no stat-model could have predicted. That would have been a disappointment as there was nothing obviously false about his placing as there was with Cerium’s 19L 5th in 2009 when the pace had slowed appreciably during the race.
Amazingly, in reconciling Neptune Collonges’ win, it transpired that IC benefits from the same tweak, only significantly more so. Though without a win in a 3m+ chase in 5 attempts (for which he remains penalised), he recorded a 10L 3rd (spookily behind Neptune Collonges!) in a 3m1f Grade 1 at Punchestown 5 years ago, off a mark of 164. His mark for the GN was 137, 27lbs lower. Together with a minor tweakings elsewhere, including recording a positive for a good run with 11.05 in the Topham 2 years ago, it has been possible to reconcile his stats to this fine run but, importantly, the handicap drop is the key catalyst and, as back-testing has confirmed, it remains a very rarely achieved stat (now 4 among the 240 strong database).
6th (17.5L) – Ballabriggs – (predicted: 4~6L behind CB – finished 5.5L behind)
Part 2 coming...
Of the model’s 3 other Win selections this year:
Synchronised (fell at 6th fence): no tweak to model. It would be handy if there were a stat that accounted for the fact that, despite no F or UR in 9 chases, there was particular reason to be concerned about a horse handling the GN fences. He hadn’t enough runs to record a positive for his jumping but, other than F/URs, there isn’t an objective way of reflecting that concern – let alone that he might dump his jockey before the off. For very sad reasons, we will never know whether, barring that fall, his extraordinary stamina would have enabled him to carry top-weight to victory to record a famous double but the fact that the winner carried only 4lbs less and the runner-up was a Festival winner suggests that he might have gone very close.
West End Rocker (fell at 2nd fence): no tweak to the model. If there was one runner that seemed a safe bet to jump round it was Becher Chase winner WER. Despite a scare at the first fence in the Becher Chase he had never fallen or unseated in 14 chases under rules and 6 hurdles. He was one of those that got very worked up with the delay and false starts – indeed he broke the tape and got entangled with it for the second one. This probably explains his early departure, thankfully apparently none the worse for it. You can’t account statistically for his fall and so no tweak is appropriate.
9th (28L) – Shakalakaboomboom (predicted: 10L ahead of CB – finished 16L behind: underperformed by 26L).
A minor tweak has been made as a result of Shaka evidently running out of gas. 8 year-olds continue to have a very patchy record in the GN, though better on good than soft. Those that win or make the places tend to have form at 3.5m+ and so I have introduced a small negative stat for an 8 y-o that doesn’t. Despite Shaka being 100% win & near-miss in 4 chases at 3m+, the furthest was 3.25m. So, retrospectively, he should have been rated “Big Run – No Cigar” rather than a Win candidate and he may well have finished closer if he’d not been ridden quite so close to the pace.
That said, this was a very creditable run indeed - up with the leaders virtually all the way and jumping beautifully (as he did in the Topham a year ago). For an 8y-o to run this well and finish within 30L of the winner is, in itself, a positive stat for a return. It’s reminiscent of Hedgehunter’s first GN in 2004 when, carrying 10.12 (like Shaka), he ran a blinder and was leading for more than a circuit until running out of gas 2 out and falling at the last - he came back a year later as a 9 y-o to carry 11.01 to victory. Assuming he’s sound, another year to mature for Shaka combined with, hopefully, both handicapper and market thinking he doesn’t get the trip could make him a very backable selection next year.
All of these tweaks have been back-tested against the winners and those placed <20L of them since 1988 and the entire fields of the 2008~2011 GNs (240 horses in all). Only 1 of the 80 winners & <20L in the database (Cerium in 2009) remains unaccounted for statistically by the model.
So, assuming that the race isn’t changed out of all recognition, the Winner and next 5 places are now guaranteed for next year ;o)
Have a great summer. Even if I haven’t earned it, Sir Chris and the boys have earned it for us all.
COYR