Betfair | Stop Making Sense: The problem with betting on trends

Stop Making Sense: The problem with betting on trends

Simon Rowlands considers an uncertain universe and tips one tipster to hang from a lamp post before the weekend is out in his look at the Eclipse Stakes preview.

I promised to return to the subject of trends at some point, so here goes.

As stated previously, I do not much go in for trends. It is not that they don't exist, but that they exist less than those who have a vested interest in writing about them sometimes pretend and that they are often measured in a manner that encourages this misconception.

I wrote about this in The Sportsman in 2006 and cannot think of a better example of the issues involved than the one I used then. As Fanny Cradock once said in a rather different context, "here's one I prepared earlier".

I considered the forthcoming Cambridgeshire Handicap and pointed out that in the previous nine years only one three-year-old had been successful.

"Hopefully, few people would jump to the conclusion from this that three-year-olds have an 11% chance of winning the race: to do so would be simplistic in the extreme.

One step would be to consider the proportion of three-year-olds that have contested the race. There have been 300 runners in the Cambridgeshire over the last nine years, of which 71 (23.7%) have been three-year-olds, so perhaps we could expect roughly two three-year-old winners on average over nine years rather than just one.

A clear problem with this is that just one more three-year-old winner in the period under review would have turned a 'poor' performance into a creditable one.

Judging race trends (and plenty of other racing statistics) on winners - and winners alone - is far too crude a means of measurement. But that does not stop many from doing just that.

What about considering winners and placed horses (i.e. those that finished in the first four)? That is at least a step in the right direction.

There have been 36 opportunities to be in the first four over nine years. You might expect three-year-olds to have accounted for eight or nine placed horses in that time, or thereabouts.

As it is, they have accounted for 14.

Judged by one measure three-year-olds have performed poorly in the Cambridgeshire. Judged by another measure they have performed well. But neither method is an entirely satisfactory way of arriving at a conclusion.

Logically, if a 'trend' affects a certain category of horses it will affect not just the winners but also the losers, and it will affect the degree to which the winners win and the losers lose as well.

You properly need to consider ALL horses which ran in a race - all 300 over nine years in the example given - and consider not just whether they won, placed or lost, but the degree to which they won, placed or lost.

As mentioned previously, cumulative lengths beaten (once adjusted for field size and a few other factors) do this very well. Their use in this context confirms that three-year-olds seem to have had an advantage, though not a large one. More importantly, using cumulative lengths beaten makes full use of all the data and results in a conclusion that is statistically significant (less than 5% probability of its happening by chance).

If you used more conventional methods of judging the outcomes of recent editions of the Cambridgeshire - such as concentrating just on winners or on horses who figured prominently - you got the illusion of a bigger effect but at a substantial cost in reliability. There was up to a 40% likelihood of these apparent effects happening by chance.

The Cambridgeshire is by no means an isolated example as well.

There are more interesting categories to look at than age and weight, of course. But the essential point is that, no matter how 'clever' the categories you are considering, if the measurement of outcome is crude the research will be compromised.

It is human nature that some - not least those who peddle the messages in the first place - gain comfort from embracing simple and unquestioned certainties in an uncertain universe.

'Only one three-year-old has won the Cambridgeshire in the last nine years, therefore don't back three-year-olds in the race' certainly gets the message across.

Unfortunately, it is the wrong message."

The only thing that I would like to add at this point is that an alternative approach to average cumulative lengths beaten per rival is to consider % of rivals beaten, which is simpler and rather easier to calculate but which conforms to the same principles.

Oh, and that three-year-olds have won both editions of the Cambridgeshire since those words were written - 'bucking the trend' according to some (but in fact quite the opposite) - and even managed a 1-2-3 in 2007...

The Eclipse Stakes at Sandown on Saturday is the big race on the horizon, and any punters wishing to get involved ante-post can do so already on Betfair.

It's fair to say that it looks a slightly substandard race this year - no Notnowcato (last year's winner), much less a Nashwan, a Giant's Causeway or a Hawk Wing - but it is perhaps a more intriguing race from a betting point of view as a result.

I have to disagree with the views in Racing Post of Mark Winstanley, who has promised to hang from a lamp post in Esher High Street if Mount Nelson wins, as I think the horse (who can be backed at [4.3]) has a first-rate chance.

A very promising two-year-old, he had only one run the following year but shaped very well indeed on his first two starts of this campaign before finishing a close fifth in the Queen Anne Stakes at Royal Ascot last time.

That was a messy race and might have flattered Mount Nelson, but it's more likely that it did him no favours, as he shapes very much as if he will be better suited by a truly run mile and a quarter than a falsely run mile. He'll be carrying some of my cash, in any case.