system analysis

[onefurlongout]

I’m currently working on a selection process which is showing a 28% strike rate over 288 races. The average winning price is a little over 3/1

I just wondered if any members thought that this was an acceptable strike rate?

My aim is to analyse 500 races but wondered if I should perhaps go for more race to get a more accurate figure?

Thoughts welcomed from anyone that has completed similar analysis or that could just offer advice on how to implement the process (level stakes, betting plan??)

Thanks

[superkauto10]

have you actually put money on the system or is it a shadow system were you just been writing things down?

i normally try to cover any system over 2 months due to you can have different results at each course and some courses tend to have the same outcome when the ground the same but on different ground can be a different result al together.

what time period have you done the 288 races over?

is it just flat ? all weather? national hunt? or all the listed?

do you have the results set out for each distance or is the percentage an overall picture?

[onefurlongout]

just shadow so far and from the 1st May all flat racing

I have the data stored and recoreded on venue, distance, going, odds.

Still some work to do me thinks

[superkauto10]

i would wait longer than 20 days to see if a system works iv looked at a few before and over a few weeks it looks bomb proof then you go ahead into it and the cracks come out its best to wait even when you wanna go ahead with using it until your 100% that you can hit a profit that your happy with without it hitting any problems.

good luck with what ever you do!

[Artemis]

onefurlongout,

Sounds quite promising.

Some time ago, Dave Jay and I agreed that about 700 races was sufficient to appraise the effectiveness of a system using level stakes. If you are still showing a slight edge after such a test, I think you can be confident that the system will carry on in profit unless there is some underlying structural change in the way odds are returned or other major changes.

Once you introduce staking plans, you cannot make any safe predictions, especially if you are increasing stakes after a loser.

I know our friend, Seagull, has been monitoring such a system which has done very well indeed over the last 18 months, and covers thousands of races. This is a ‘Martingale’ type system which any statistician will tell you can only lead to ruin in the long term.

[onefurlongout]

Thanks for the advise Artemis.

I have currently completed assessment of 329 races and I’m hopeful of assessing more over the next few days.

Within my analysis spreadsheet I have the following headings

Date
Meeting
Race type (eg handicap)
class
going
Dist
Top rated won (Y/N)
Price of top rated
number of top rated selections (there is sometimes more than 1)
Second rated won (Y/N)
Price of second rated
number of second rated selections

I’ve also complete the above for the 3rd,4th and last picks

Does anyone else think I need any further headings?

As for the analysis I look at:

the winners from each section against all runs
the number of single picks wins against single pick wins
Average prices
Type of race against win/runs
Distances against wins/runs
Going against wins/rund

I’m sure there are more to look at and I would appreciate any advice from any members that have conducted this type of analysis.

Thanks

[Gerald]

I saw this post a couple of days ago, but thought I’d wait and let someone more mathematically astute reply first, before I make a fool of myself.

I don’t have any maths book anymore, so this is just off the top of my head.

Haven’t got my head around the actual system, and not really interested.

On the issue of how many races to analyse before becoming confident in it:

The Binomial Distribution springs to mind.

n = number of races, = 288 or whatever

p = assumed chance of picking a winner, = 0.28

This is the bit where I’m not sure that my maths is right. I think the Standard Deviation for the Binomial Distribution is the square root of npq, where q = 1-p.

So square root of npq = square root of 288*0.28*0.72
= square root of 58
= about 7 and a half.

As n approaches infinity, the Binomial Distribution tends to approximate to the Normal Distribution.

This is the bit where I really need a set of statistical tables. I think that the 95% Confidence Level is within 2 Standard Deviations (1.96???), and that the 99% Confidence Level will be closer to 3 Standard Deviations (over 2.5 SDs?).

Number of wins from 288 races, expected to be 288*0.28 = 80.6, say 81.
So if my figures and memory are correct, you can be 95% sure that you will get between 66 and 96 winners.

[edit: Putting it a different way, if your sample result is 81 winners from 288 races, then you can be 95% confident that the actual long-term figure is between 66 and 96, or that the 28% success rate in the sample is from a population with an actual strike-rate of between 23% and 33%.]

Now for the experts such as the Dice Man to come along and say what’s what.

[onefurlongout]

Once again guy thanks for the posts.

Just to clarify. I apply the same assessment criteria and apply points to every runner which then provides me with an overall result. When all the individual points are totalled the one with the highest number of points is the top rated etc etc

At the moment I am assessing all race types to establish any patterns. From the analysis I’m hopeful of develping a more refined approach which may improve the win ratio. eg all 5f maidens.

So far from all the races covered the win ratio is 28%.

Thank for the advice Gerald I’ve just wiki’d Binomial Distribution and all I can say is wow!! I think your version is much more concise

[Gerald]

If the chance of picking a winner is 0.28, then the chance of picking a loser is 0.72

The chance of picking 10 losers in a row will be 0.72 to the power 10.

Now, 0.72 squared is approximately 1/2

So, 0.72 to the power 10 is approximately 1/2 to the power 5, which is 1/32

So you will have a 10 loser streak 1 in 32.

A 20 loser streak will be 1/32 squared, which is approximately 1/1000.

If you are having 100 selections a week, then about once every 10 weeks you will have a 20 runner losing streak.

[KevinTHFC]

Hi OFO,

The margin of error can be obtained form the following:-

MOE = 1/Sqrt(NoOfSelections)

so if you had a 1,000 selection, your MOE would be +/- 3.16%

On your 288 samples the MOE is +/-5.89%.

This means that the expect S/R for you system icould be as low as 26.35% or as high as 29.65%

Note that you would need 10,000 samples to getthe MOE down to +/- 1%

Kevin

[Pompete]

While obviously I accept the principle of paper-trading a system I do believe there are limits to extent of real knowledge to be gained from doing so to x amount of selections. And with respect analysing it like that weirdo bloke in Numb3rs. As, it only really matters when the moneys down.

Imv to continue to paper trade is a waste of time. What now needs to be tested is your ability to get the money on. Clearly (as with all of us) the amount risked should only be an amount you’re happy to lose and it doesn’t have to be huge. A tenner will give you 100 x 10p bets. But, the routine of doing so will teach you more about the merits of your system than another 500 paper-trades.

For example, if you’re putting the bets on before work are the markets available? Have you had time the previous evening to sort your selections? If placing the bets at work – what happens if you’re called into a meeting or you’re working into your break. These are just silly examples and I’m sure you’ve already thought about the practicalities but many a good system will come unstuck when you can’t get the money on (and you can guarantee it’s the good winner that ends the losing streak you miss out on).

If this is successful then the process of filtering your results to enhance profitability can begin.

Good luck with this Jeff.

http://www.fanrush.com/gallery/data/508/thumbs/whiteboard.gif

Forget that math bollocks and put a pony on it

[onefurlongout]

I take your point Pompete.

Had I put the money down I would have been well up by now

[douginho]

I like all these mathematical postings. Did a bit of statistics at uni and found it fascinating, albeit boring when not applied to Horse Racing!

Good luck onefurlongout as it sounds like you are putting in a lot of effort with this.

I have one question. If you are using all different types of flat races so far then in one year you will have many different types of races over many different distances. If you look back and find that you seem to out perform the median on 2m handicaps and on listed races over 1m would you then concentrate exclusively on these sorts of races in future?? Or is that too risky?

[onefurlongout]

Thanks Douginho.

The plan is to work out from a large amount of data which races are showing the most favourable results 5f races are showing the best early results so dfar but this could be refined even further.

[Gerald]

I put 28% and 3/1 into the Kelly formula, and it comes out with the figure of 4% of the bank on each selection to maximise profits.

Of course, the actual price, and probably probability vary from these figures from case to case, but it gives some kind of ballpark figure.

In addition, it isn’t really feasible to keep on working out how much to put on when you are dealing with multiple selections in a day.

I’d suggest splitting the bank into 50 or even 100 bets. If you do the latter, you will carry on regardless through a bad patch, without faltering.

Recalculate the bank and the bet sizes once a week. Of course, it is all a matter of taste and judgment – I suppose you could do it once a day.

The thing to realise is that if you have something that is working there is no need to be greedy or panic, just keep on chipping away.

[Artemis]

If you use Gerald and Kevin’s calculations and onefurlongout’s method of selection, you have as good a system as you can get.

4% of the bank staked on a method with a very slight profitability and a high number of bets would get you barred very quickly from any casino. If you think about it, you are virtually acting as a bookmaker – a steady relatively small profit with very little risk.

The maths and statistics will stand up – I’m certain of that – but will the selection method? I look forward to hearing the results from 700 selections.

The profitabilty is very small, so stakes will have to be relatively high and the bank will have to be a very sizeable sum. Aiming for a profit of £500 per week, would require bets of about £20,000 a week and an available bank of half a million. Of course, the bank only needs to be ‘available’ and should not be needed, but there is a very small (quantifiable) risk that the system could lose more than half of it.

[Gerald]

Hadn’t really thought about it like that.

I was thinking of perhaps a £1000 bank, and £10 bets, not spending any of it, and steadily working your way up.

Maybe I’m too much of a coward.

[Author]

Posts