Predictive Statistical Modelling

[Pacoboy]

Am looking for any London based enthusiasts who are keen on embarking on working on developing forecasting models for horse racing odds etc. People that can help steer my graft/ideas with better grasp of statistics/pure maths or programming than me – the two or three heads better than one mantra.

I have developed a programme which can handle large amounts of form data/manipulate different factors etc. I have a few ideas and have read many things on the kinds of statistical models that people develop (eg based on Chapman and Bolton paper in the 80s). Mutinomial Logit Models/Maximum Likliehood Estimations etc etc… all a little above my head at the moment (oh how I wish I took the statistics options when offered to me at uni back in the day!). I realise this kind of approach is not everyone’s ‘bag’ but if anyone is interested in offering help, I’d be most grateful. Following on…..

Anyone done any work on researching recency weighted means of form factors? Seems logical but needs to accommodate the random influences on results – I’ve seen Zipfs Law/Fibonacci series used but that intuitively to me places too much weight on more recent events when considering many horses have excusable blips LTO in an otherwise progressive profile…suggestions for an algorithm for removing the blips from the profile…anyone done any research on placing importance on factors depending on time elapsed?

Pace analysis in our non-sectional world – spread of lengths beaten/ average lengths beaten as a measure of estimating pace effect in a race?

Algorithms for predicting improvement or ‘coming to form’ etc??

Anyone interested in seriously tackling or collaborating in any of the above???

Cheers

[Gingertipster]

Pacoboy,
As someone who does his own odds comiling to find my idea of the value selection/s, I don’t see any way of making a programme to do it for you.

Each horse’s chance is not made up of equal amounts of each aspect of form.

e.g. With a real dog of a horse, temperament might make up almost all the percentage chance of winning. Where as one with an average temperament might have very little of its percentage make up from the "temperament" aspect.

If soft, one horse who loves soft might have a lot of its chance made up of "going" and less about "recent form" (if it has run twice recently on good-firm). Another might have soft ground form but ran poorly on soft last time on soft, so would have a bigger part of its percentage about "recent form". etc.

Also, each race is different. A race where there is only one prominent runner may favour that prominent runner. Where as a race with seven prominent runners may favour those held up etc.

Draw might be very important to some, but for a front runner will probably be able to get out and get the rail anyway, it’s far less of a problem.

So anyone putting information in to a computer will have to put in to it his own opinion of how much each aspect is going to affect the result. And so it won’t take "opinion" out of the equation.

By the time you programme all the relevent information in to a computer, you’d just as well do it yourself with your own brain. If you’ve got a love for statistics and maths Paco, you’re half way there anyway.

To start,
I’d advise you to buy a Timeform Racecard and a Racing Post. AFTER racing has finished, (doesn’t matter so much about the result, concentrate on SP’s). I take it you know odds as percentages? (3/1 = 25% and 15/2 11.7% etc). Study the Timeform and Racing Post and see if you can understand why the SP’s were as they were priced. Although SP’s are not always a correct estimation of chance, it is as good a guide a novice compiler has.

Once you’ve got the idea from a few race cards, buy Computer Timeform or Race Passes (I make do with the print version, Perspective, Racehorses and Chasers And Hurdlers)… And off you go. Don’t give up, it’ll take time to get it right.

Any advice you want, give me a shout!

Hope I haven’t depressed you too much Paco, just think you’ll be wasting your time trying to create a programme.

Value Is Everything

[Pacoboy]

Hey, many thanks for the words – some interesting points – I like the observations of each horse having different weightings of preferences – not necessarily constant across the population. I will chew on that for sure. Allow me to put up a counter argument too. Spend a few mins reading up on the Hong Kong Syndicates (Bill Benter etc) who employ just such models based on the ideas printed in (Chapman and Bolton) paper ‘Searching for positive returns at the race track’ –or something to that effect. Interesting stuff and touched a nerve for me.

With regards the human v computer issue. I would argue that the sorts of human processes you describe can be coded in a program and in a crude way they are already in mine – i.e a parameter such as pace being affected by other parameters such as field size/draw/going etc. To use your example, a computer could predict how many front runners are likely in today’s race and then analyse over thousands of past races how such a ‘shape’ affected results (ie favouring hold up horses etc and to what extent) this will be affected by the course/going/field size etc etc all of which can be accounted for in the process. Whether they know it or not, I’m sure successful punters will follow ‘routines’ as above and many of those could be coded as such. I approached the coding of some parameters by recording exactly what I would do manually to determine the info I require…from form/ratings etc then get the computer to copy. There are pro’s and cons for both – more flexibility from the human without doubt…but more consistency of approach (useful if its a successful approach!) from the machine unaffected by emotion…the computer able to digest much more data as well of course.

The hardcore statistical methods written about in some of the references above do appear to be a robust, mathematically sound basis for eliminating some (not all) of the uncertainty in the data which you rightfully point out is a big barrier to analysis. Don’t quite grasp the pure mathematical/statistical theories enough to begin thinking about seeing how it could be applied. Just wondered if there was anyone out there who did.

Do you keep records of your odds and the edge they give?

Cheers

[Gingertipster]

The problem with working out pace by numbers of runners is sometimes there are a lot of prominent runners in a 12 runner race and sometimes not.
There are so many different types of prominent runner which can also effect a race.

Pacemakers. Front runners who need to lead. One that needs a view of the front (leading or to one side of the leader but does not like being behind horses). Front runner who is also effective tracking the pace. One capable of anything from making the pace to being held up. One who just likes tracking the pace.
And at what pace does each horse like to race? Some prominent runners who are stayers at the trip will need to make a good pace. Some pace setters don’t have a change of pace at any trip so are best with even fractions. Some have a change of pace and can kick off slow fractions. Some yet to race at the distance might try different tactics (held up) to get the trip.

If there are just two prominent runners who both need to lead they could take each other on up front. Where as four prominent runners in a race where none need to lead may not end up fast run at all.

Value Is Everything

[Gingertipster]

Do you keep records of your odds and the edge they give?

Yes, I have a small pocket sized book which I take to the races. When studying a race I make a few notes on each runner the day before, put all runners in order best chance to worst. Like to give each runner a percentage chance working to an over all 100%. Then convert each percentage to odds and back the one/s I make a better chance than bookmakers. I won’t back anything which is only slightly better than my tissue, so as to give myself a margin for error. Most bets I’ll do before leaving. Sometimes I’ll change the odds I’m willing to take if seeing something positive or negative at the races.

I guess if I am making an over all profit, then that is the edge it gives me.

Value Is Everything

[Author]

Posts