Weight-For-Age system

[rmwesley]

I've been playing around trying to build systems based on commonsense logic, as opposed to a systematic backfitting approach which can't then be explained in the real world. I've come across the following scenario and was wondering if anyone had any ideas what was going on.

As a starting point, I have a dataset going back to the 1st Jan 2010 which covers all races in the UK and Eire. To build the system I have done the following, filtering to a smaller subset of the data at each step:

1) Selected UK races only
2) Selected Flat races only
3) Excluded all maiden, novice, selling, claiming, classified, auction, apprentice, conditional, amateur or hunter races
4) Selected handicap races only
5) Selected races with only one 3yo (4 or more runners), and/or only two 3yo (8 or more runners), and/or only three 3yo (12 or more runners)
6) For the races with two or three 3yo I have picked only the 3yo which started at the lowest price (based on Industry Starting Price)
7) I then exclude any of the 3yo horses which are left, which started at an ISP of greater than 5/1 (to maximize strike rate & not rely on random big priced winners)

I should now have a list of 3yo horses, hopefully with a weight-for-age advantage running against older horses. I bet £1 to win on each horse using the ISP to determine profits.

The resulting bank roll (i.e. net profit) looks like this:



The system looks to be fairly consistently unprofitable up to the start of 2015, at which point it turns around to start looking fairly consistently profitable until the tail end of 2021, at which point it then reverses the trend yet again to become fairly consistently unprofitable up to the current date.

It is entirely possible I've messed up my code and done something stupid, but I have double- and triple-checked it.

This definitely doesn't look random to me. Anyone got any ideas what might be going on (e.g. WFA rule changes) to make these trends form, then flip not once but twice? Thanks for any insights, Richard.

 

[Outlander]

I don’t know anything about systems but I know something about the accuracy of the betfair market. Sophisticated people way above my intelligence level are churning this data to within an inch of its life to find an edge including yourself.
My thoughts are market correction would be the answer/reason.
The strategy shows a loss to betfair sp, edge is identified people lay , prices get bigger , eventually they get too big and profits start, then the boffins decide this is a back strategy so the market eventually adjusts and the prices get smaller and then the profits nose dive again.
This is the problem with systems by the time you have identified an edge the market has already adjusted.
I‘m a contrarian, if had a betfair account that didn’t have a massive premium charge and I was into systems and analysing data as soon as I identified a positive trend I would assume I was too late and bet against the trend and vice versa, but that’s just me and I haven’t really got a clue what I‘m taking about on the subject but it does interest me and a good post.
That aside I think they did do a bit of fiddling with the WFA scale something definitely rings a bell there also.
Some people were mentioning that 3yo had an unfair advantage on the scale maybe they did some remedial work to correct.

just found this and the surprising thing is that they made an adjustment against 3yo in 2017 and that coincides with an upside in profits for the strategy, maybe not surprising to the contrarian in me who would assume the boffins would make a market adjustment against the 3yo’s knowing the amendments made against them , and go against the logic anticipating they would be fighting for a perceived logical edge and overdo it.

BHA to implement alterations to Flat Weight-For-Age Scale | British Horseracing Authority

www.britishhorseracing.com

 

[TheBluesBrother]

just found this and the surprising thing is that they made an adjustment against 3yo in 2017 and that coincides with an upside in profits for the strategy,

The reason changes were made had something to do with 3yo's running in the Ark de Triomphe, where 3yo fillies received a WFA allowance of 11lbs, how can you give a horse like Treve that?

It was back in about 2013 that Dave Edwards mentioned to me that the Racing Post uses its own WFA table, you only have to compare the difference between the allowance the BHA allows a 2yo over 5f in March versus the Racing Post. there is a 21lb difference.

Mike.

 

[rmwesley]

Many thanks both for the replies.

I've investigated other scenarios to try and find conditions / filters which can improve a target variable, e.g. to improve strike rate, or profit per £1 bet, or Sharpe Ratio, or to reduce the length of losing streaks etc. These scenarios have included horses running again after a quick turnaround, horses changing trainer, horses moving up or down in class etc.

It is usually possible to improve this target variable by making some sensible decisions to filter a list of possible bets, e.g. select all horses who have been turned around to run again in a week or less, who won last time out, and whose odds have come in by at least 20%. However, these filters seem to be independent of time, i.e. they have exactly the same effect on the target variable today as they did every year since 2010. It is also very difficult (impossible) to find filters which create a profitable system without backfitting massively.

This WFA has been the only instance where a) there is such a clear cut time-dependent trend, and b) the system is (was) profitable with absolutely minimal backfitting

Just seemed very odd to me.