Number of Factors

[Vortech]

Is it really true that the more rules you have in the system the less effective it is in the longer term?

I understand the concept of back filtering but still not quite sure that a system with more rules has more probabilty of failing.

If you generate a system with 3-5 rules with 40,000 selections - 25%SR - 10,000 winners at avg of $5.00 thats a POT of 20% ($10,000 - profit). Just an example.

If you restricted that to 15 rules and come up with only 2000 selections for 40% strike rate for 800 winners at avg of $5.00 you have a profit of $2000.

Betting level stakes would seem to continue with the 3-5 rules at Profit is more, but with confidence in a higher strike rate you can use a larger stake of your bank like 5%, instead of 1-2%.

Back to my original point, by filtering more rules why does everyone say its setup to fail? Is confidence increased by not on the amount of horse races bet on or is it by the number of meetings tested and eliminated when they don't fit the criteria?

Just interested to know others thoughts?

[UselessBettor]

I'll give it a stab and try to give a reason.

Backfitting is a problem when the rules you add are not aded for a logical reason. The more rules you have the more likely that you are backfitting some of those rules.

The other reason is that usully the more rules you add to the system, the more you restrict the result set. Less results mean less confidence.

I can find many systems that only have 2 rules that make 5% plus but I have to say that variance is huge in these systems. You go from -20% to +20% swings and the average punter couldn't handle them(me included). So I think adding a couple of extra rules to minimise the variance is acceptable.

[darkydog2002]

The more rules you have the more it leads back to the selection becoming the fav and at best scenario favs lose 67 % of races.
Cheers
darky

[Chrome Prince]

There's a golden rule I have.
Unless you can properly explain the reason for the rule, leave it out.
That's why so many systems fail.

[Vortech]

If your original system has a 10% POt and you add another 5 rules, does that mean the your system will drop below the 10% POT. You have have to say no in theory.

I using a 2 step model. Using 3 rules as a base to get a 10-15% POT and adding extra filters.

I know in the end of the day, the horses that should have won only to be let down by the jockey or sweating/playing up at the barriers.

[AngryPixie]

As few rules/filters as possible is best. I seldom use more than two or three with laying.

The problem with using alot is that they begin to overlap. The rules should be independent of each other.

[TheSchmile]

If you add the extra rules to tighten up the selection process or to specifically isolate the value runners then I don't see a problem. If it's obvious that your selections are disadvantaged in certain states/conditions and your strike rate and POT increases as a result then go for it.

The see systems as a set of rules constructed to either:

(a) Isolate undervalued runners in the market
(b) Isolate absolute good things that have many lengths on the opposition and will be able to sit 3 wide and still bolt in. As most know, these systems usually end up with very low action.

Whatever you do, keep monitoring the results so that you can justify when things go good or bad and tighten things up if it's appropriate.

The Schmile

[Bhagwan]

General the more rules one has , the less likely history will repeat itself.

When running tests on systems...

One method is to run them in batches of 150

Some batches will show a profit & other batches may very well break even.

That way one gets a stronger idea if the plan has any legs or not.

Is also a strong idea to test the price range the winners came from also.
If profit came from say 1-2 outsiders , then the warning bells should be sounding because the system is less likely to repeat that result.

The term Back fitting is often used as a derogatory word to describe a system that was shoe horned into making a profit using a hundred rules over layed on past data .
In other words the data was massaged into profit by adding a hundred rules.
.
I believe seeing that future data is not available then that's all we have to work with.
The question we are then left with, is will these rules repeat themselves into the future.
Some do & some dont,
That's why it is suggested to run the rules over previously unseen data to see if it holds up..

Once we have a lot of races say over 4.5 years , its an idea to batch them into lots on 150 races & see how consistent it may be.
We have all seen systems that absolutely nail it over say a 3 month period , then fall in a hole - Very frustrating.



Here is an example.

Lay Plan
Lay bet all 2nd Favs priced 2.00-4.90 that have a win SR of 20-23% in races with 9+ runners.

What stands out here as possible back fitting, is the 20-23% win SR.

This may tell us the the percentages have been cherry picked ,to force that data into profit.
This may or may not repeat itself.
One would have to then run this against unseen data to see if it holds up.
It may indeed be a loop-hole based on punters over betting these commodities because of this popular percentage range punters & systems ,like to use in their rules.

Because there are only a few rules , there is a strong chance that the data may repeat itself.

That Lay Plan above, is currently showing...
+36% POT
80% SR

[TheSchmile]

If profit came from say 1-2 outsiders , then the warning bells should be sounding because the system is less likely to repeat that result.


Nice post Bhagwan,

Completely agree.

The Schmile

[darkydog2002]

Couldn,t agree more.

[Vortech]

Thanks guys.
Appreciate your time to post your comments from your history of racing knowledge.

I've got heavy money on one today SR1-1 Rosie's Reward

[norisk]

Like the 7 myself Vortech, but you're on class runner, no doubt, good luck:)

[KaiserSoze]

So, how about this... Let's say that based on 3 months of data you come up with a system based on 3 factors; A, B and C. They could be anything such as last start winner within 7 days over less than 1400m, or any other oft talked about criteria. This is just an example, but lets say that A, B and C give you 300 selections in the 3 months with a win SR of 20% and a POT of 6%.

Then you notice that 50 of those selections had a maiden as there last start and only 5 of those backed up to win. So you add the filter that last start can't have been a maiden. You now have a win SR of 22% and a POT of 8%.

Then you notice that there were 50 of those selections that were ranked 7th or worse for place SR and only 5 of those won again. So you add a filter to consider only those ranked in the top 6 for place stike rate. You now have a Win SR of 25% and a POT of 10%, say.

Then you notice that there were 30 of those remaining selections with a win SR of less than 15%, of which only 4 won again. Add a filter that only considers horses with a win SR greater than 15%. You now have a win SR of 27% and a POT of 11%.

Then you notice that there are a further 30 selections that are outside the top 4 for neural ranking, of which only 5 won. Add a filter that only considers the top 4 neural ranked horses. You now have a win SR of 29% and a POT of 13%.

Then you notice that of the remaining 140 selections there are a further 25 that have a UniTAB rating of less than 95, of which only 5 won. Add a filter that only considers only horses rated more than 94. You now have a win SR of 31% and a POT of 14%.

Then, in a moment of inspiration, you notice that there are 20 of the remaining selections with a pre-post price of $6 or more. of which only 3 won. Filter them out and you have a win SR of 35% and a POT of 15%.

Then, in a moment akin to something that could only have been experienced before in human history by the likes of Newton, Tesla, or Eistein, you notice that there are 14 of the remaining selections that ran from barrier 11 or higher, NONE OF WHICH WON. Add a filter to consider only horses running from barrier 10 and lower.

You are left with 81 selections over 3 months and a win SR of 41% and a POT of 19%. BUT, you have 10 filters, most of which have been fitted to the existing historical data. Backfitting?

BUT, you have also only selected criteria that hones in on the best chances given your original starting point.

Yes, it's true that you are also honing in on the favourites, but with the confidence of a 41% SR...

Yes, it's true that 81 selections over 3 months isn't a lot of action, but it's hardly non-repeatable...

Can any of the more experienced punters offer comments on the above method?

[Vortech]

This was my original arguement but in better detail.

Thankyou

[TWOBETS]

Morning Kaiser Soze,

This is exactly the way many (most?) folks tackle backfit system manufacture, and because I know diddly squat about horse racing it's the way I went about it to the letter. It's a well known fact that you can't beat the odds this way and yet somehow I do???? (Betting Unitab!!!)

I'm down to an average of less than four bets a day (only 1 yesterday) and yet I'm doing nicely.

The only widget missing in your arsenal may be finding a filter that might not be as obvious as those types you mentioned or you run the risk of forever diminishing bets AND forever diminishing odds.

Having said that , I must admit that my average payout for the last three years is a meager $1.16(Unitab) or $1.09(Betfew). Place Divvies.

[Chrome Prince]

I think you are sifting through the data and relying on too many factors to make a profit.
With all due respect, it actually is classic backfitting.
You've gone from 300 selections to 81 selections with 8 additional rules.
Even if you did apply the rules, you'd need at least 10 times the sample size to have any real confidence about it repeating.
Just my opinion.

[Barny]

I totally agree with the others here that a good system should have a few rules, 4 or 5 max and these rules must be logical. If you do the same as everyone else is doing regarding filters then surely you must lose though.

An example of an all too often used filter in a system is "days last start". You get the impression that almost everyone uses this filter, let's say it's Must have run within 21 days. All those horses that have run within 21 days (which incidentally will be a lot of the field) will be carrying far more money than they should compared to their "real odds / chances", and hence they'll be under the odds. Another example is Must have run within the first four at it's last 4 runs ..... A system has to have limits doesn't it, BUT, just to be different, and get value, why not allow the horse to have one of it's finishes to be 5th or 6th ? That will set you apart a little and give you a chance of some value.

[TWOBETS]

Even if you did apply the rules, you'd need at least 10 times the sample size to have any real confidence about it repeating.

Agree with you there CP. Or maybe 50 times the sample size.

I've tried to explain this before but everyone thought I was on drugs. Still, I'll have another go....... When backfitting systems it is impossible (almost then) to isolate a particular input factor. You may think you are adding only "first up last start" runners for example. But in fact within that group you have a million other factors that may or may not be having an input on the result.

Thus when you add or subtract any single input it actually affects the other input factors.....There, clear as a bell

[Chrome Prince]

You may think you are adding only "first up last start" runners for example. But in fact within that group you have a million other factors that may or may not be having an input on the result.

Thus when you add or subtract any single input it actually affects the other input factors.....There, clear as a bell

Bingo.

[Vortech]

I'm actually experimenting a little at the moment with horses 2nd up.

1st run back from a spell - 1st or 2nd on settling, 1st or 2nd on turn and finished outside of top 3 but only 3 lengths max off the winner. 2nd run back often sees huge improvement. Some advise from the old punters in the west.

Some great results of late, but wanted to apply more filters. Should I keep it to this simple system and not apply any additional filters?

[Chrome Prince]

I've been experimenting on first up from a spell horses, there seems to be some tremendous value on horses that would have otherwise probably have been favourite.

[darkydog2002]

I,ve got hundreds of these backfitted systems from a well known source (not PPM)
None would be worth the paper their printed on.

I,ve actually posted a few on here.

Rules like "Last run in 14 days /2nd last run in 35 days/3rd last run in 46 days"

I,d say there would be more than 1 person on this forum who fell for them too.

Cheers
darky

[Barny]

TWOBETS, that's brilliant, absolutely brilliant.

I actually "get" what you're saying, and I will honestly and unashamedly admit, I HAD NEVER THOUGHT ABOUT IT THAT WAY BEFORE.

Brilliant, just brilliant ..... thanks for sharing that.

[darkydog2002]

Syd /Melb/Bris/Adel
Field size after scratchings = 16 Max
Age of Horse = 3 - 5
Place % = 40 - 100
Barrier 1 - 13 after scratchings
Weight today 52 - 58 KG
Days to LS = 7 - 21
Days to 2nd LS = Max 49
Runs this preparation = 3 - 10
LS finished 2nd
2nd LS = won
3rd LS = 2nd
4th LS= 1 - 10
LS SP to Max 25/1
Distance change = Not coming back more than 300 M

Not going up more than 2 KG
Av Prizemoney Rank = 1 - 6

Enjoy,

Its a oldie and sold for $200 +

[Vortech]

Was it tested over a period from like 2008-2009 and then fell into a hole afterwards?

You often have to wonder why someone sells a system with great returns, instead of keeping to themselves and backing it with there own coin

[darkydog2002]

Not sure but the usual blurb that goes with most of these is "$300 turned into $168,000 betting 3 % of bank."
Trouble is the minute you purchase, it loses and continues to lose.
Cheers
darky

ps.The next one might be called "The Resurrection"

[darkydog2002]

Anyone game enough to test it ( To see how it performed over the last 2 years)
Cheers
darky

[Vortech]

I'll give it a run tonight

[darkydog2002]

Thanks Vortech.
Cheers
darky

[Vortech]

Overal 125 Bets - 26 winners return 127.90 Unitab

2011 - 20bet, 4 winners, return 29.20
2010 - 42bet, 9 winners, return 33.3
2009 - 29bet, 6 winners, return 23.7
2008 - 34bet, 7 winners, return 41.7

Is this a lesson being taught to the novice punter using backfiltering?

[gunny72]

My best system so far had one rule, viz, from the top three ranked place percent horses back the two with the longest prices.

I excluded 2yo and 3yo races and only bet on metro meetings so I suppose that is two more rules.

This gave me a modest pot and a lot of fun for a decade but sadly the value is no longer there as lots of punters use place percent now.

[darkydog2002]

Thanks Vortech,
If it isnt it should be.
Cheers
darky