back fitting, staking and chi tests

[stephen12000]

Hi there, I’ve been think about these things lately:

I don’t understand the lack of logic behind back fitting.

My thinking is, say I start monitoring a system today for a year and, just for argument the basic system takes no account of whether the selection is suited to the going, however as a potential filter the going is noted and the horse’s previous performance on the day’s going is noted so that when they coincide ( selection won on this going previously for exmaple) it is noted as such in a database.

After the year is up i look at the results of the system and find it maybe made a small profit, but if i just pick out those real time selections where the going matched a greater profit is made, then this would seem sensible to continue with the system with that filter applied.

However, if I do the same thing going backwards in time, it is somehow not considered worthy and called back fitting. Unless I have completely misunderstood the term, isn’t back fitting excatly the same thing as monitoring a filter in real time? Isn’t it usefull to back fit? and if not why not?

Staking. I just wondered if anyone has tried a 20% rolling stake on a very small bank and turned it quickly into a large profit and reduced the stake percentage accordingly. By small bank I mean as little as £10 – no fear, sod all to lose, even if you blow it after awhile you’ll only be pennies down.

Chi Test.  I have read a lot about this lately, and I can see it is used as some kind of psychological boost that the system you are using is working due to logic rather than luck.

However I saw awhile back where the 5/4 had a quite high chi test figure, meaning it was unlikely that the results were down to luck, yet the figure has dropped drastically and one post I saw on another forum said it was now only about 65% that the results were down to the system being sound. In otherwords, it was only about 50/50 – the results could be due to luck almost as much as skill.

Doesn’t this show what a load of b*****ks

the chi test is related to horse racing? How can it be down to the ‘soundness’ of the system one time – ie, it’s not luck but logic, then later, it is as chi testers would say ‘just as likely the results are due to luck, not the ideas built into the system’.

What do you all think of those 3 things, have I completely misunderstood backfitting and ch itests?

[snowman]

Hi Stephen & welcome to TRF.

Personally I don’t like rolling banks, if you have a hundred quid & want to stake 5% then stake £5 until your bank has gone or you have managed to increase it to say £150, then by all means re-assess and increase your stakes if you wish to the new 5% figure, earlier if you want to.(£6 when 120).But I never reduce after a loss it just takes you longer to recover. This is a personal opinion you understand I would never dream of deriding anyone elses chosen method of betting.

As far as back fitting is concerned you raise an interesting point. Backfitting IS frowned upon I think when someone looks over the last 2 or 3 weeks and declares that the third clear fav, in all races under 2 mile, in Handicaps, shows a 35 point profit per week and therefore this is the system to be on. – Particularly if they are trying to sell it! Yet if someone has done the work and gone back far enough it is considered to be a well researched, viable, proven selection method.

I think possibly the problem lies with people who come up with a system that has no basis in logic but can prove it works by back fitting (ie the horse with the longest name in all maidens – HANG ON! Didn’t Westmoreland Road win the 5.20 at Newbury!! I think i’m onto something here!!!)

Don’t have a view on chi tests Stephen don’t know enough about them.

Again welcome – interesting post.

[stephen12000]

Thanks for the welcome and the reply.

I guess it doesn’t make much sense to reduce stakes in the rolling bank tradition, I have started a little (£24) bank to experiment with 20% stakes just to see how far i get, if i don’t reduce stakes at that percentage then i think it would go pretty quickly, but with a normal larger bank and more sensible percentage there is not much reason as say you in reducing stakes.

The backfitting question just bothered me because we are always being told to beware, as you say you could backfit a third letter R type system and flog it. Clearly there is positive and negative back fitting.

I was hoping someone who is up on chi/archie tests would express an opinion. My math/logic skills are not up to much and really i just want to know if the chi test is trully usefull to horse racing punters or just a bit of a red herring.

<br>Thanks again snowman:) <br>

[ray7]

Hi Stephen

On the subject of the chi test. I don`t know how to work it out without going through old posts but from what I have read the more data the more accurate the result.<br>For example the 5/4 system (and most others) is going through a bad patch at the moment and if you applied the chi test just on the past few weeks it would show a bad result but if you applied it on the results for Feb it would look fantastic.<br>From that I think that for horse racing you need to use at least a full years results.

On the question of backfitting I think the problem with that is that most people dont distinguish between backfitting and backtesting. By that I mean they call it all backfitting. The way I see it is backfitting is looking at past results and working out a system that would have picked the winners. This would probley only work with the races checked and loose you money if you tried to use it on future races. Backtesting on the other hand is where you come up with a system and set out the rules then use the past few years data to see if it would have produced a profit. Still no reason to belive it would show a profit in the future but a better bet than the first example.

These are just my personal thoughts on the subject but I hope they help you.

Good punting<br>Ray

[snowman]

Hi Ray,

I think your differentiation between Backfitting & Backtesting has hit the nail on the head, don’t you agree Stephen?

BackTESTING is a useful tool that means you don’t have to wait 18 months to get 18 months of results on your system, whereas BackFITTING is taking 40 winners and finding some way of linking them all together even if there is no basis in logic.

Nice one Ray.  

[dave jay]

Hi everyone … I don’t think Chi-Square Test can be successfully applied to racing data on it’s own … for those who don’t know the formula for this it’s ;

(E – O) squared / E

E = expected frequency<br>O = observed frequency

In racing deduct the number of winners you actually have from the number of winners you thought you would have and multiply this figure by itself … then divide this number by the expected number of winners ..

I suppose all this will tell you is that when your going through a bad patch you’ll have the figures to prove it .. by way of a Chi factor …<br>:laugh:

Stephen, I think it might be better if you looked at Binomial Distribution instead, IMO, this is far more useful when betting and assessing your probable chances of winning ….

Backfitting … backtesting … I’d agree with the latter. Got to be done ..

[ray7]

At the risk of sounding thick.<br>Whats Binomial Distribution

Ray

[Bricoman]

Hi Everyone,

I have observed that systems tend to do well until they get in the public domain. <br>The 5/4 system (currently going through a bad patch) started off with a chi score of 99% and since it has become popular the prices have (on the whole) plummeted. It dosn’t suprise me that a 2/1 shot 3 months ago would now be an odds on shot.

I would also assume that when punters realise that they are not getting the returns that they expected, they will stop backing the selections, the prices will get better, and the remaining "die-hards" will start making money again.

It’s just my theory – but it’s an obvious one!

Regards

Bricoman

[dave jay]

I tend to agree with you there Bricoman … I think the recovery period for any data coming into the Public Domain is three to five years. Sometimes the edge never returns, like using good but less than well known Tipsters. Once they are uncovered the edge is gone within a season or so and it won’t be coming back.

[dave jay]

For Binomial Distribution see

http://www.ruf.rice.edu/~lane/stat_sim/binom_demo.html

Ray.

Click on the title Binomial Distribution on that page to get the full calculation.

(Edited by dave jay at 11:09 pm on April 16, 2003)

[Author]

Posts