Cappa Bleu - Won Foxhunters at Cheltenham two years ago and fancied to run a big race for that years Hennessey when falling. Not many many miles on the clock following lay off and back to form this season with a win at Haydock followed by a great trial yesterday when finishing third in the Welsh National.
50/1 currently available with Stan James.
0
Comments
You're too kind, as always. Need to see the weights (I think 14 Feb they're announced).
Cappa Bleu certainly looks interesting (available at 40 on Betfair if you want to take the plunge) but until the weights come out I wouldn't like to say whether it quite makes a winning stat score in my model, which has undergone the customary review and tweaking since Ballabriggs pooped last year's party.
Will be crunching the numbers and posting the selections as soon as poss after 14 Feb.
Did you just watch the Lexus? Synchronised, off the bridle a long way, hung in with them gutsily (under AP of course) on ground that didn;t suit him and simply ground them down to win it. They should have run him in the last GN (not least coz I'd backed him ante post!) (Edit - though to be fair AP couldn't have deserted Dont Push It and on quick ground he definitely needs an "AP" on board)
He was a bit unlucky as the winner was thrown in at the weights. He has to get some form at 3 miles+ but when he romps home at AIntree in 2013, I won't forget I read it here first NSS ;o)
So all the punters that took the 28/1 in good faith are told "Don't be silly fellas! As if we are going to get this account holder to pay out. It is so obviously a mistake"
Well you can bet your bottom dollar that if the horse had fallen at the last everyone who backed it would have had lost their stake and would have been told that "That's gambling folks!"
I see Betfair have issued a statement but have so far refused invataions from both BBC Radio and Talksport to come on air tonight to explain things further.
I am hoping for some clarity from AddickAddict who says he works at Betfair.
That's a really, really good question Chunes and, for me, not at all boring.....though the proper answer may be exceptionally so, so I'll give you a short one first.
The short answer is that I will ignore a result if it is clearly a freak. For example, my database goes back to the 1988 GN but it excludes input from one race (2001) when a massive melee led effectively to a 2 horse race. I didn't use a model then and it would be crazy to include either of those horses' characteristics as a guide to future GNs. But, much though I would like to think that Ballabriggs was a freak winner and Oscar Time really should have won, there is no way that it should be regarded as a false result. Ballabriggs simply exposed a deficiency in the model (relevant I think to quicker ground). In fact, I look for deficiencies even when the model has picked the winner (as explained below). One day the wheels will come off but so far it has worked OK.
At the risk of boring the socks off you, I'll give you a fuller explanation. A few background musings/statements of the obvious. Firstly, even a successful stats model (whatever its use) is not a crystal ball. With horse racing, you can pick a winner by judgement or luck but the aim of a stats-model is to improve your percentages and of course, over time, your returns. Any model will at some point fail (though that doesn't necessarily make it a bad model); arguably it always fails to some degree, even when it appears to succeed. Secondly, a model aims to stop you following your gut (or heart) when it comes to selections but, though it should be applied systematically (based upon objective data, hopefully accurately gathered), there is inevitably a good deal of subjectivity in constructing it; at the very least in selecting the criteria used to screen the runners. Thirdly, many stats-based systems tend to use a rigid "red-line" methodology with criteria drawn from previous winners only. They seek to eliminate from consideration all runners that fail to comply with ALL of the chosen criteria, thereby leaving the list of eligible candidates. The problem with a red-line system is that there is little or no allowance for the changing characteristics of a race that has changed significantly over the years and is continuing to do so. Just as important, the problem with only considering previous winners is that you ignore the fact that, but for a bit of luck here or there, a close runner up (with very different stats) could easily have won the race.
So, what's all that got to do with the price of fish?
I'll try to demonstrate with specific examples why, IMHO, a GN model should include closely-placed horses in its database, why it should use a largely "points-based" rather than red-line methodology and why it needs to be reviewed and, as necessary, tweaked every year, regardless of success.
Stat-followers used to say that since no French-bred horse had won the GN you should put a line through all French-bred runners. Then in 2009 Mon Mome popped up at 100/1 (cue nostalgic cheering from Plaaayer). But Mely Moss (French-bred) narrowly missed out to Papillon in 2000 and Clan Royal (French-bred) was a 3-length 2nd in '05 (jockey dropped his whip on the run-in) and 7-length 3rd in '07. Both of these could easily have won. There are many other examples of why the characteristics of closely-placed horses should be an essential component of the database.
Take the old 11 stone weight barrier. Until Hedgehunter won the 2005 GN carrying 11stone 1lb, no winner had carried 11st+ since 1988 and most GN stat-followers used to put a red-line through all runners that did. When he won, the threshold for elimination was simply raised to 11.02. I started using a GN stats model in 2006. Then it was a red-line system which had an 11.02 weight elimination threshold. Happily, it threw up the winner and 4th placed from 6 selections in '06 but I was uncomfortable that Hedgehunter had came a 6-length 2nd carrying 11.12. Moreover, the handicapper was beginning to compress the weights to give higher rated horses a better chance. So I decided to adapt the model to a points-based system, penalising horses for not complying with the selection criteria rather than eliminating them. It didn't have much effect on the model's selections in 2007 (which only identified 1 horse with a "winning score", McKelvey, which came a 3/4-length 2nd) and, since the first 5 home all carried less than 11 stone, normal service on weight was seemingly restored. But the change began to pay off in 2008 when the model scored with the winner, 2nd and 4th from 5 selections, with the 4th placed, Slim Pickings, carrying 11.03.
It picked the winner again (from 6 selections) with Mon Mome in 2009, carrying 11 stone but I was again uncomfortable, not so much that the other 5 selections had bombed (you simply have to accept that not all horses take to the race or run to their best in the GN for whatever reason) but because the 3 placed horses carried 11.06, 11.04 and 11.02. Clearly the handicapper's weight compression was beginning to take effect and so I decided to tweak once again the criteria and the weight penalty-scale and in 2010 the model had its second very good year with 1st, 4th, 5th, 6th and the simple fact is that if I hadn't made these adjustments, the model would never had identified Dont Push It, the first winner to carry as much as 11.05 for 28 years. While there is a risk (a certainty I guess) that my tweakings will one day cause the model to crash and burn, I prefer to try to account for unexpected results as a way of trying to accomodate the model to changing characteristics of the race.
So, for anyone still awake, what about the tweakings caused by Ballabriggs? There have been a number of tweaks but the main two are: a) to treat a Class 2 win at the Cheltenham Festival as a Class 1 (a tweak also verified when back-testing Silver Birch's win) and b) to adjust some of the criteria and penalties according to ground conditions, adopting 4 ground scenarios (from soft/heavy through to good/quicker) to give 4 respective scores to each runner. Of course, it won't be possible to know the precise going before the race but back-testing the model suggests that there are gradual changes to most horses' scores through the going range and I'm intending to make sure that the possible range of ground is covered in the final selections. It also suggests that on good to soft or easier last year, Ballabriggs would have had to settle for a place and Silver By Nature would have been Oscar Time's nearest danger. Shoulda, woulda, coulda!
Well you did ask Chunes :-)
Peanuts is our very own Stephen Hawking, which is meant as a compliment.
Saying that, the intricies of the model bamboozle me (and I'm doing a maths degree) but the detail contained within it and the knowledge you have behind it are amazing. Here's hoping to another successful year!
It would look a very weird system to a real mathmetician who knows how to do proper regression anaylsis. I'm strictly an amateur statistician but it's kind of worked so far. Got long to go at University?
Maybe more likely that Synchronised takes his chance (with AP)?
A man's got to know his limitations