PDA

View Full Version : How good is your system?



mathare
18th January 2004, 17:38
I know quite a few people on this forum run successful systems. I suspect that many others, like myself, have tried to come up with equally good systems. But how do you know whether or not a system is working?

Obviously you can look at strike rates and profit figures. But can you tell when a good system has just had a bad start or conversely when a poor system has had a lucky run at the beginning? I have a pretty solid mathematical background (2 A-levels in maths and a masters degree in theoretical physics) but I understand that most punters don't. So I'm going to try and explain some maths that may help you decide how well your system is working.

Following a few simple steps, we can work out if a system is actually showing progress or whether we could achieve similar results through luck alone.

1) Record every system bet in a spreadsheet. This is essential and you should record at least the profit/loss for each bet and also the total profit/loss on the system to date. NB You can do this using something other than a spreadsheet but you had better be bloody good at maths for the later steps
2) Divide the total profit/loss by the number of system bets to get an average win/loss per bet
3) Calculate the standard deviation of the individual system bet results
4) Work out the standard error by dividing the standard deviation by the square root of the number of system bets
5) Divide the average win/loss by the standard error to get the t-value.

That's the steps to follow, but what does it all mean and what can it tell you?

I'm assuming everyone can understand the averaging in step 2. This just gives us an indication of how successful (or otherwise) an average system bet is. The tricky bit comes in step 3, calculating the standard deviation. This value gives an indication of how much each the profit from each individual bet varies from the average.

The good news is you don't really need to understand what this means too much, you'll just have to trust me on what it means and how it is used. The even better news is that if you are using a spreadsheet then it can calculate the standard deviation for you. If not, you are in for some long, complicated maths that I have no intention of going in to here. If you are using Excel and you're individual bet profits are in column E then entering =stdev(E:E) in a cell on the same page will populate that cell with the standard deviation of you individual system bet results. I'm sure there is also an equivalent in Lotus 1-2-3, Open Office, Star Office or whatever your chosen spreadsheet application is.

That's the tricky bit done and you should be able to complete all 5 steps now. Again, a spreadsheet can be used to help with the maths. They really are very powerful for recording bets and making sense of the data.

At the end of it you have the t-value. What does this mean? Basically, the higher that value the more confident you can be that the profits you seen have been as a result of skill or a viable system rather than just chance. If the value is negative then all it is telling you is that you have made a loss, and you probably knew that anyway. It's only really worth thinking about positive values. If the t-value is 0.7 you can be 75% certain the profits have come from a good system and you haven't just been lucky. With a t-value of 1 you are 80% certain your system is good; a t-value of 2 gives you 95% certainty and a t-value of 3 means 99% certainty.

I have been applying this to some of my systems that I have been developing. I'm not going to publish any results/qualifiers/t-values yet as the systems are still immature, more data is required before I would be happy to publish any figures. The more system bets you have recorded the more data you have and the more accurate and reliable the t-value will be. That said, the data can be obtained for any number of bets and will start to settle down once you have recorded 40 or 50 bets. If it looks poor at that stage it may be worth starting the system again, changing the staking plan or abandoning it all together in favour of something else.

Also the t-value is greatly affected by the staking plan you use with your system because the average win per bet is involved in the calculation. For my systems that use a staking plan I also record profits to level stakes and calculate t-values for both my staking plan and level stakes. A change in staking plan may make a losing system into a winning one but ensure that the reasons behind the plan are sound and that you haven't just changed the plan to suit past data. If I change staking plans on a system I restart my records and work out the t-value after 50 bets using the new plan to see how it has performed. The value of paper-trading systems in a spreadsheet is you can quickly apply a number of staking plans across the same qualifiers and record the profits separately for each.

I hope some of you find this useful. Of course, no amount of maths will turn a rubbish system into one that will make you a million overnight. What I hope this does is encourage anyone paper-trading systems to apply a bit of maths to them to see if the systems look as though they may be profitable long-term.

Now I feel I must apologise for the length of this post. :snooze If anyone has read and understood it all then go and have a well-deserved cup of tea :laugh

Win2Win
18th January 2004, 18:11
WOW!! :) You've been busy, glad to see our education these days is of use to gambling :laugh Not sure that's Mr Blurts intentions though......

Generally the higher the SR, the lower the ROI%. The more profit you want, the higher the risk you need to take. I have system that easily throw up over 100pts a year, but have SR of 13%. That means if you can't handle losing runs of around 40, don't even go their.

Generally a system trial should last 100 qualifiers, but I always like to use a minimum of 250, 500 is best though.

However, if your system only selects 50 qualifiers a year, then the trial would last 10 years!!

Another good way is to graph all bets on level stakes, this shows you the ups and downs over time, so you easily pick out upward, and downward trends.

Not saying anymore though, as it's not the Member's Area.

mathare
18th January 2004, 18:17
If the government didn't want us to use our education for gambling why did they teach us about huge debts by replacing grants with loans?

bigcumba
18th January 2004, 18:49
Excellent post Mathare!

Littlewesty
18th January 2004, 19:36
I left school many many moons ago, and for most of my life after I have only used maths in a small way. Since taking horse racing seriously you would be amazed at the new things I have learned, when I was a youngster if amyone mentioned a spreadsheet, people would want to know if it was for a double or single bed :)

With the aid of my computer, one or two evening courses I have come to use excel more and more everyday, I design my own spreadsheets and use colours and lots of 'formulas' to make life easy. I must say after Mathare's post I have a long way to go, but I shall print it out and see if I can work my way through it.

Bit more homework :)

Thanks guys

mathare
18th January 2004, 19:39
All the maths in my post can be done using formulas Littlewesty, and unusually for most Microsoft products the help in Excel can be pretty useful.

But if you do find yourself struggling I will be happy to help :)

GLENCROFT
19th January 2004, 11:38
Mathare ,

I followed your calcuations on one of my systems

Results as follows

147 bets
average win per bet £2.92 (level £10 stake)
SD 13.33
SE 1.10

T-Value 2.92/1.1 = 2.65

But doesn't the stake affect it ?

If I had done it to a £1 stake the t-value would have been 0.265

or have I missed something ?

Glencroft.

plater
19th January 2004, 12:20
Yet another great read "Thanks Mathare", this Forum is an excellent window on what we should all be doing and contributing towards, I do not contribute myself much lately but I'm working on it :) , and will reveal all (now there's a horrid thought) at a later date.

Plater

Workshy
19th January 2004, 12:48
Jeez Mathare! I left school 4 and a half years ago and I thought I left statistics there! :D

A useful post nonetheless and some food for thought...

Vic
19th January 2004, 13:22
Fantastic stuff Mathare - will try this to see what I get. My recent results are so bad that I don't need the maths, but applying the maths to the earlier results on current systems will be interesting. I forgot all I learnt in Maths about 35 years ago, but am looking forward to re-educating myself - I guess this really is applied maths rather than pure maths, took both at school but as an accountant all you really need to know is subtraction!
Thanks for a really helpful post.

Cheers Vic

mathare
19th January 2004, 20:21
T-Value 2.92/1.1 = 2.65

But doesn't the stake affect it ?

If I had done it to a £1 stake the t-value would have been 0.265

or have I missed something ?


You have indeed missed something, I am sorry to say.

The results you give are for a £10 stake and you are averaging £2.92 profit per bet. To a £1 stake this would only be £0.29 per bet. But your standard deviation to a £1 stake would no longer be 13.33, it would be 1.333 (i.e. 1/10th, in line with the decrease in stakes).

Completing the calculation gives a standard error of 0.110 and thus a t-value of 0.29/0.110 = 2.65, exactly the same as you started with for a £10 stake.

The actual value of the stake is irrelevant in these calculations, so you can put 1p or £1000 on each selection and still get the same t-value.

John
19th January 2004, 20:42
Fantastic stuff Mathare, I've just come off the face of a maths exam (which included standard deviation) so don't really wish to talk about it... but!

If you carried out the Variance i.e...



Σ (x * the mean) * frequency
__________________________
n

('n' above should be in the middle)


...then to calculate the Standard Deviation, just square root the Variance. And if you want to be even cleverer :rolleyes: divide your answer to this by the Mean to get the Co-efficient of Variation.

Don't you just love Maths. Ooo

Top post, I'll try out the forumulas in my spreadsheets.

Many thanks!

mathare
19th January 2004, 21:19
Nice one SJ. I love people who can rattle off the maths formulae for stuff like that. Especially when they then say they will go and get a spreadsheet to do it all for them :) :laugh :)

Win2Win
19th January 2004, 21:46
Nice maths guys, me love it. I fancy Carol Voderman as well, just imagine running of endless formulae in the heat of passion.....

bigcumba
19th January 2004, 21:54
Nice maths guys, me love it. I fancy Carol Voderman as well, just imagine running of endless formulae in the heat of passion.....

Especially when you put your variables in the wrong place.... :yikes:

John
20th January 2004, 00:59
.....At the same time as her saying "one from the top and any other five" :laugh

GLENCROFT
20th January 2004, 10:19
Thanks Mathare - I thought that it should not matter.

Hard to believe I 've got a degree in Pure Maths + Statistics !

In my defence it was over 20 years ago.

Interesting post.

podtog
21st January 2004, 01:13
Excellent post Mathare, the maths completely baffles me but the reasoning behind it looks good, I,m good for subtraction, percentages and the like then i,m stuffed, wish I could do it though. Now 6 beers minus 5 beers = 1 beer left.

mathare
21st January 2004, 22:05
Heehee :)

That's the beauty of spreadsheets these days, no need to understand the underlying maths, you just need to know what to type where

Bin57again
2nd March 2004, 18:01
Mathare
Thanks for that post. Most interesting. Learnt more relevant maths in your post than in the whole of secondary school. No doubt you found maths' vocational usefulness before I did.
Anyway, I have a system that registers 0.8 as t's value. In effect though, all I'm doing is using a standard staking plan - knowing I'll come up with a winner before long. I know you've touched on this elsewhere but could you reiterate. How is t's value affected by an increasing staking system? Presumably, it must be affected otherwise even roulette would register a success (based on past results and expectation of future results...not probability).
Best regards.

WillsDonkey
1st February 2006, 18:13
This site is one hell of a read, good though!
I've only recently joined the forum so I am gonna start banks for the A+/A bets shortly.

Anyway my point is im also papertrading a system of my own, the thing is when i calculated the t-value for it, it came out as 1.6 for level stakes. But when i re-calculated the t-value when using a 3% staking plan it came out as 1.2.

Does this mean the level staking plan is better in the long run, because at the moment the 3% staking plan has a greater profit or does it mean my system should be filed in the bin.

Thanks in advance

mathare
1st February 2006, 18:19
Anyway my point is im also papertrading a system of my own, the thing is when i calculated the t-value for it, it came out as 1.6 for level stakes. But when i re-calculated the t-value when using a 3% staking plan it came out as 1.2.

Does this mean the level staking plan is better in the long run, because at the moment the 3% staking plan has a greater profit or does it mean my system should be filed in the binIf the system is profitable long term then it shouldn't be binned, no. You don't say how many selections you have. The more selections the values are calculated over the more accurate they will be.

Level stakes may also be showing a better t-value as it may have a higher ROI than the 3% staking.

Without knowing more about the sample size and things like LLR, LWR, average price etc it is hard to say whether you should use level stakes or 3%. But you're doing the right thing in considering alternative staking plans for the system :)

WillsDonkey
1st February 2006, 18:58
Right then here goes. Selections over last 2 months.

Selections = 205
Wins = 52
SR% = 25.37%
LLR = 11
LWR = 3
Average Price = 7.13 on betfair

Profit with Level staking plan = + 57 points

Hope this helps you to help me...cheers

tovarich
1st February 2006, 19:01
Surely, if a system is showing a profit - on paper - over a reasonable period of time (100/200/500 selections or a full racing seaon) then it must be considered a winning system? I know that figures can be made to mean anything but all the percentages and averages mean nothing if the system doesn't show a profit? No?:dunker

Win2Win
1st February 2006, 20:24
Surely, if a system is showing a profit - on paper - over a reasonable period of time (100/200/500 selections or a full racing seaon) then it must be considered a winning system? I know that figures can be made to mean anything but all the percentages and averages mean nothing if the system doesn't show a profit? No?:dunker
No, it means you can start a bank on it you are prepared to lose :)

mathare
1st February 2006, 20:29
Surely, if a system is showing a profit - on paper - over a reasonable period of time (100/200/500 selections or a full racing seaon) then it must be considered a winning system? I know that figures can be made to mean anything but all the percentages and averages mean nothing if the system doesn't show a profit? No?:dunkerAs I said earlier - if it is profitable long term then don't bin it. What constitutes long term differs from person to person and system to system though.

mathare
1st February 2006, 20:32
If I were you will I would paper trade for a bit longer. You're looking at around 100 selections a month. Another couple of months will see you into the start of the flat season. If it is still looking good then I'd start a small cash bank and follow the system to whichever staking looks best over the data you have at that point. I'd calculate your ROIs for all staking methods too though.

And remember a 3% staking with a LLR of 11 means you can drop a third of your bank in one hit so make sure you can ride out fluctuations like that.

tovarich
2nd February 2006, 10:47
No, it means you can start a bank on it you are prepared to lose :)

Can't agree, Keith. A losing system is a losing system and I don't see what can be done except bin it and start wth something new. ("SOMETHING NEW", I seem to have seen that before somewhere.:D )

If a system can't make a profit using level stakes (LSP - the standard we work to) then I don't think any amount of messing about will turn it into a winning one. :doh All you need is a system that picks more winners than losers and your in betting heaven.:D

mathare
2nd February 2006, 10:50
If a system can't make a profit using level stakes (LSP - the standard we work to) then I don't think any amount of messing about will turn it into a winning one. :doh All you need is a system that picks more winners than losers and your in betting heaven.:DI agree about LSP - a system that doesn't profit to LSP is very unlikely to profit on other staking plans. And if profits are seen on other staking plans I think it is a product of insufficient data to get a true long-term view of the system.

I disagree on the fact you need more winners than losers though. Winners at prices that more than cancel out the losses from the losers is all you need, not an SR over 50% :)

tovarich
2nd February 2006, 11:14
That's exactly what I meant to say, Mathare, enough winners at prices big enough to cover the losses, e.g. a strike rate of 10% would be fine (I think) if all the winners were 20/1. But then I'm no mathematician.:doh

mathare
2nd February 2006, 11:17
That's exactly what I meant to say, Mathare, enough winners at prices big enough to cover the losses, e.g. a strike rate of 10% would be fine (I think) if all the winners were 20/1. But then I'm no mathematician.:dohI know Tov - just joshing with ya :) And if the winners are 20/1 then you can drop the strike rate down further to around 5% if you wish and still make a profit.

Profit Seeker
4th February 2006, 12:50
Imo, there is no true winning system, OR every system is a winner. :doh
For example, we know people including myself that run systems that have profited every year for the past millennium (ok then 10+ years), but no matter how good that looks, they can still collapse at any minute. But then again, you can take a system that has a few losing years with some good profiting ones, with the bank and staking to account for the poor times, you still may come out on top, obviously huge patience and discipline is need for this and you would need plenty of other things to make your money in the meantime. If someone only monitored this system in it's losing year they would say it's no good, but not necessarily true.

Win2Win
4th February 2006, 13:29
Your starting to sound like a pro PS :) All you need do now is learn how to make some money! :D

Profit Seeker
4th February 2006, 13:46
That's the hard part of the pro criteria to meet. The patience and discipline side. I'm making the same as everyone else on the PLs so can't be going far wrong.

Win2Win
4th February 2006, 13:51
YOU'RE making money :yikes: I'll have to 'fix' the systems :laugh

Profit Seeker
4th February 2006, 14:24
Ye, only recouped losses though :yikes: 70% increase in Jan about same as most :)