System Testing - Sample Sizes

for n>= 30, the sample's average will approach the mean of the population being sampled as n increases towards infinity.

Personally speaking, I will define my expected value-per-bet before each wager, so am not too pushed about such sampling -> E[x] will always > specified margin before playing.

in excess of 1000.

I think it is most important if you are
including an logical ingredient seemingly
overlooked by others, but 200 picks, strike
rate, ROI as stated, seems good enough.
2 Other points, I personally don't like paper
trading or back fitting, beware of those.

Look at the P&L for each race and calculate the mean and standard deviation to do a z-test: http://en.wikipedia.org/wiki/Z-test

This can give you some confidence that the profit isn't just due to chance.

these stats guys will always say you never have enough data.

Are you offering or only taking bets? If you're offering, increasing stakes will likely reduce ROI due to getting more matched when are wrong than when you are right.

I think 200 results is good enough if not back fitting.

Just "curious" as to the sport you are betting on?

If HORSE RACING, then you can get runs of "good luck" and runs of "bad luck" that can distort results. So the more you check results, with thought about them, then the more it will be clear that a real "edge" does exist.

200 seems far too low to be "confident" re future events....if it is HORSE RACING. I've checked almost 2,000 races, so have a "deep" understanding of what strategies "work well" ......and what strategies "work", but there are longer losing runs . 

I'm certainly not going to check out any more results, as the time it takes to do so thoroughly is
"considerable"!

It might be that you are only applying your strategy to certain race types......if that is the case, then the 200 fig might be a reasonable amount ( my check was on almost every race in Britain, at different points in season and on NH/AW and Turf.)

Hope you get some more replies from mathematicians....but your initial post is not comprehensive enough re what you are doing, so perhaps that is why you are not getting many replies on thread.

GL

Under new username  (posted as martinch).

Ideally we would have more info on odds and comm level to answer this question properly and my answer below  isn't perfect, but from what you say we can do rough approximation:
1 - number of trials = 212
2 - Assume all odds are 3.4, so p = 1/3.4*=0.294
3 - we would expect to win on 212 * 0.294 = 62.4 events given 1 and 2
4 - you actually won on 72.6 bets (I backed this out from 35 points profit at 3.4)

If you use excel you can use a formula
=1-BINOMDIST(72.6,212,0.294,TRUE)
1 - this gives the probability of winning in 72.6 OR MORE events GIVEN YOU HAVE NO EDGE = 6.5%
So it's quite unlikely that you are purely getting lucky and therefore it looks fairly convincing that you have an edge.

I would estimate your edge at 8% (we could have a huge debate on how to get 16% to a guess of your ROI on your next 212 bets).  When you should ramp up... that's really a person specific question - I would just suggest doing things gradually (and stakes should always be a % of bankroll).


And the more your average odds vary around 3.4 the more issues with the stuff above.

...and meanwhile back in the real world....you can calculate until you're blue in the face but it don't mean a dam thing as far as predicting future returns.I have methods which will show good returns for 1500 bets and show a loss for the next 1500 bets.....the only way to find if a method works is by putting down the cash..... and if you're in profit it works!

isn't that an expensive way to find out that something doesn't work?

Not if you're betting in 1p units

you'll get banned for talk like that

ive tested my plan.
37.5 strike rate
272 % ROI
on 240 races]

if I give you £2.40 can you test it for me properly?

if you cant read the form book.
never going to make it pay.
there are plans can help you.
putin 27/1 winner reading form
found that one.

When should you increase it? Why not now?

Start with a hypothetical bank of £200.

Build conditions to bet 3% / 4% or 2.5%.  The 'conditions' are built around 'sub-groups' of your systems.  You start at 3%, notch up based on analysis.  Notch down if you have had weaker results in a certain type of event (i.e. if football, if your strike rate is low when the system brings up a certain team, pull back to 2.5%

Increase by the odds won / decrease on loss by the stake.

If you traded to reduce your risk, this doesn't get factored into your 'hypothetical bank' since it won't be an indicator on your performance.

roadrunner46     15 Mar 12 15:07 
if you cant read the form book.
never going to make it pay.
there are plans can help you.
putin 27/1 winner reading form
found that one.

I like this one. Reason I created a backing strategy in the first place was to break even with it and reduce my PC liability, it seems I have been too successful and created a profitable backing strategy which generates only 18% commission on profit - was hoping for 40%!

Still I bow to your superior knowledge roadrunner, must get my head in the form book as there is obviously no other way to make money, silly me.

JLivermore - thanks for the maths, I dusted off my old A level books - going back many years and came up with the same conclusion. I have calculated a margin of error of -/+ 6.9%, which means worse case scenario it is just better than break even. However, this is all theoretical of course.

so you make money laying horses.
how comes 215 bets you only made
ROI 16%. yet your strike rate was 38%
your bets werent level stakes?

yes your sums add up.forgive me.
my ROI 272% how many times
is that more then your ROI of 16%?
im not good at maths theory.

There are two main things to consider:

You need to verify your theoretical edge and...
You need to use a decent staking system, such as kelly (or a modern modification), to place optimal bets.

To do the first, you need thousands of bets to be fully confident.
For the second, you need to know the "true" price of what you are betting on and the price you're getting, from that you can work out what percentage of your bank to bet to optimally grow it over time.

In general, up your stakes slowly, don't get carried away if you end up with a sweet winning streak. Remember that if you're getting an edge of more than 5% you're probably wrong (it's not impossible, but very rare).

So much seems to be wrong in your post that it defies belief, for starters....

point(1)
average odds of 3.5 multiplied by strike rate (SR%) of  38% would yield an average ROI of 133%, ROI meaning Return on investment.Your figure of 16% ROI neither refers to the implied ROI figure of 133% nor to ROI generally as its more near to POI figure format/profile or profit on investment.

point 2
A sample size of 200 is extremely small and no clear indicator to any robust profitable system and 2000 sample size would be a much better hallmark to gain confidence in any system!

point 3
Your average odds of 3.5 seem quite small and must refer to favorites of which if you were to look at many 200 sample size segments you could quite easily fall upon 116% roi or even 200% roi along with terrible losing runes equating to say 50% ROI, furthermore favorites are extremely difficult to make a profit with because all the value has been drained out with pricing mechanisms like trading/stake amounts and bookmakers over-cutting industry prices via track side bookmaking monopolization.Most pro punters will be betting on average odds of between 5.0 and up to 11.0 and certainly not relying on unprofitable short odds.

point4
You have not mentioned if you have back fitted your profitable sample runners as this is the easiest mistake to make when looking for profit as what appears to be profitable in the past may and does not generally reflect as profitable into the future and has more to do with natural variation than revealing hidden wealth and will actually mostly plummet into loss in the near future to naturally offset the peak that you recently observed

Can somebody please explain BACKFITTING ?


If my laying strategy consists of the following rules --- Have i backfitted in any way and if not can somebody give me an example ?  thanks


1.  Fav
2.  Won last time out
3.  Won by under 1/4 length

I shall try to explain a back fitted system and where it can lead to problems if I may;

When looking at past data it is possible to see many peaks and troughs in the data where strike rate and ROI are good and thus profitable but you cannot say that this data then is always profitable as is the data trend based on natural variation or on a robust bias with a theoretical edge that provides constant profit?

In your example you are using a system based on a theoretical edge that being you are sub-dividing all favorites on horse racing into ones that have won last time therefore more likely to have either a penalty, raise in race class or drop in starting price due to attracting stakes.
Then you are sub-dividing into horses that have won well and those that were not clear of the competition and less likely to repeat a win.

The system seems to have good theoretical sense as a possible profitable laying system and not one that was simply back fitted that looked good retrospectively with no theoretically biased profit.

p.s. your system for laying will make even more money if you look for a change of jockey since the last win and should give you as much as 5% gain in ROI%, this is because the factor that led the original jockey to make any correct deciosions in ridng the horse to its win is no longer present.

cheers bf - thanks for your reply

Could you please give me an e.g of a backfitted system ?

thanks

From FLATSTATS :-


Why Systems Fail

Systems fail mostly because the core rules are all wrong to start with. A punter creates a system with rules which are not sound and then procedes to back fit the system to make the results look good.

Back fitting is like shooting an arrow at a tree and then painting a target around where the arrow lands. In other words you are making the system fit the results. What usually happens is a punter thinks of a basic idea, looks at the data, and whittles away at the data until the results look good. This is not a sound way of creating systems and will lead to the system failing sooner rather than later.

Examples of backfitted systems include rules such as:

    Only back the horse when priced between 2/1 and 8/1 - Why 2/1 and 8/1? Why not 3/1 and 9/1?
    Don't back 3yo or 9yo horses - Why 3yo and 9yo? Why not 4yo and 8yo?
    Best with field sizes of between 11 and 13 runners - Why 11 and 13? Why not 12 and 14?

If a rule can not be justified then it is probably best not to use it. Only if there is a specific reason why the rules should be used should it be included. An example of this could be: 'Only back colts and geldings'. If this rule was applied to an all weather system then it is justified as there is a distinct gender bias on the all weather.

Systems can fail if the punter does not hold his nerve and ride out any (inevitable) losing streak. This all boils down to money management. If the punter does not stake an optimum amount on each horse, and prefers to place amounts based on hunches, or luck then he is going to find his bank becoming volatile and ultimately end up depleted.

A whole new topic can be started on staking plans but suffice to say: if a punter does not understand that he needs to control his stakes when the heat is on then he may as well forget the system from the start. Overbetting after a win, underbetting after a loss, or missing out a few bets to save money is not a sound way of betting and will also ultimately deplete the bank.

Another reason why systems fail is that some of them are boring to operate. The strike rate and profitability maybe good but the number of contenders it produces is low. Each day punters would have to check the racecards manually to find a system bet. This could go on for ages and many punters will just give up on the system.

One way to get around is this is to us the Flatstats System Alerts feature. This takes all the hard work out of finding contenders. Create your system, set the 'Alert Me' option to on and each day the racecards are scanned and any system bets found are listed on your own horse / system alerts page.

It will be easier to explain with an example what exactly a back fitted system would be and it is more to do with how one finds a system rather than it having any particular style or format which could be anything really.

Lets use a roulette wheel as an example, Fred wanders past the roulette wheel at his local casino of which it is his first visit with a friend, he notices that a player always puts chips on four squares on the table and seems to be winning more chips than everyone else, in fact after half an hour he takes what seems to be a huge amount of chips off the table and cashes them in much to the high praise of friends and other players who delight in his sudden good fortune.

Fred decides that he will will go to the casino tomorrow night and put chips on the very same numbers as he reckons this is a good system from what he has seen and the fact the numbers came up again after the winning player had left.

What do you think happen to Fred when he returned with a sum of money in the hope of winning big using this new system that he has noticed worked well the night before ?

The whole subject of backfitting a system leads on to a very important point that will occur in most gamblers systems as for me you have 2 types of selections.

(1) profile selected

This method is probably the most widely used but can be the most problematic as it avails itself to the backfitting negative method of finding profit.Profile selected assumes that a certain set of condtions are present in the form of the runner or team and/or inthe race or match type as well as weather and seasonal conditionas which play there part too.

This method also can lead to much fewer selections as the more fussy you become about perfect data the less selections over the same time you end up with and some backers may only back once or twice a week using this method which cant be good in the long run to make good monthly units of profit.

(2) rating selected

This is the best and powerfullest method for finding winners as it does not reduce the amount of selections possible but puts all the runners or teams in a set order reletive to there merits which can be anything you wish to add in what ever amounts suit best using a points system.

The beauty of this method is that you cover all the runners or teams and have a broader or total spectrum of chances realtive to points that can be checked overtime to index realtive chances.

I wonder if anybody else has made this discovery as I have tried both types of system selection and know that raing is by far the powerfullest method possible and highly lucrative!

typo*rating

I have used RANKING order re RATED races ...then if odds are "value" to lay or back, depending on position in ranking ordr, bets should be placed.

EG ...If horse is ranked 3rd top rated and is 2-1 SP......there is no "value" backing or laying it.

If odds are say 3-1 SP....it is an automatic back.


If odds are say evens.....it is definitely not a back....and could be considered as a lay....but as in top 3 rated I probably would not lay it.

=============================================================

I am taking the RANK order as a strict guide to finding value re bets....

Of course I have had to BACK CHECK results on approximately 2,000 races against system ratings and odds to "see" the uses for the system. Some might believe this is BACK FITTING.......but is not IMO.....it is just arriving at a strategy that works due to the value in odds and the strength of system re rating races.

very intresting points DFC and quite true, I am compiling at the moment and not at the point of back checking(not back -fitting) the ranking and points aspect againt market price and position asI have not reached adequte sample size to find reliable trends.You have quite a volume I see and can afford this approach and its strange that I am curremtly noticing the same effects that you mention, If a favourite is far down the pecking order in terms of pointer/points then it fares much less well than when it is clear top-rated. I am convinced that in reverse one could find very good lays as some runners mark up very poorly and yet can even be odds on some times!

I agree that back-checking and back-fitting are 2 different methods,
sample volumes and thoerical edge being the deciding chalk marks !

p.s. When I refered to "not finding relible trends" my reference was in aspect of ranking order against price and not top rated which are proving quite profitable.

Recently for instance at the Chelenham festival I achieved about 130 points profit BSP with my top three rated for each race, not bad for a four meeting range of races!

Chelts Festival
my top 3 rated

avg BSP12.46
72.00
16.00
127.29

avg BSP 12.46
Sels    72.00
winners 16.00
BSP pft 127.29

8.53    Trifol    769.2728034    -1
12    Cinder    633.3841858    12
8.91    Steps     594.0085982    -1
12.17    Menora    978.2538647    -1
1.82    Sprint    977.123353    1.82
4.97    Al Fer    680.0416801    -1
90    Mon Mo    1812.208459    -1
7.8    Quanti    1391.394003    -1
7.25    Hold O    927.2809957    -1
1.79    Hurric    2389.644199    -1
12    Zarkan    1330.242344    -1
27    Overtu    1041.156383    -1
22.26    Gone T    880.117675    -1
19.07    Garde     864.0259387    -1
5.03    Scotsi    838.8437792    -1
1.61    Queveg    2431.737105    1.61
14    Shop D    718.2122933    -1
90    Dare T    701.4774868    -1
9.7    Triolo    598.6016709    -1
24.47    Educat    537.3360625    -1
9    Hunt B    459.7239613    9
6.6    Teafor    797.6351086    6.6
8    Allee     731.6793619    -1
    Soll42    618.8784105   
3.31    Simons    840.1837352    3.31
12    Cotton    702.3888444    -1
6.87    Monksl    673.0626609    -1
2.42    Grands    1737.413922    -1
31.42    Call T    640.4572072    -1
5.7    Bobs W    615.9766028    5.7
1.96    Sizing    2525.099322    -1
8.8    Big Ze    1807.29125    -1
5.9    Finian    1435.056067    5.9
34    Final     1108.520744    -1
8.74    Get Me    1056.558612    -1
0    0    893.0219852    -1
8.91    Kazlia    778.0535313    -1
36.43    Kapga     487.3386989    -1
55.32    Une Ar    474.0886466    55.32
17.73    Clonba    835.6933389    -1
16.5    Pique     787.9381496    -1
22.54    Champa    631.3158147    22.54
3.95    Peddle    1968.602944    -1
8.83    Crista    970.535644    -1
4.7    Sir De    854.6717242    4.7
19.28    Restle    764.5284285    -1
82.14    Reinde    581.7928427    -1
21.97    Pineau    510.3777877    -1
13.85    Albert    3004.542835    -1
7.43    Rubi L    2019.578899    -1
4.39    Rivers    1941.51851    4.39
1.87    Big Bu    1906.427506    1.87
5.5    Oscar     1789.737935    -1
11.5    Thousa    1640.162972    -1
11.84    Divers    1364.903682    -1
23.4    Giorgi    972.9666438    -1
9.04    Notus     940.5859334    -1
14.53    Becaus    2531.704456    -1
9.53    Sunnyh    1368.883744    9.53
81.06    Piraya    1229.202506    -1
9.74    Balder    1036.504591    -1
46    Countr    675.68133    46
6.37    Grumet    622.4553352    -1
    Clerk´    1154.313733   
24.39    Moon D    1124.326071    -1
10.5    Raya S    944.6095868    -1
2.5    Boston    906.8627315    -1
9    Brindi    857.5619839    9
18.1    Rocky     476.2006066    -1
357.89    The Mi    4375.997412    -1
4.5    Kauto     3414.765158    -1
2.86    Long R    3344.150816    -1
5.27    Cloudy    3141.163353    -1
13.5    Barber    996.641694    -1
20.72    Roulez    813.0969442    -1
17.18    Moloto    875.3076416    -1
    I´msin    875.1976888   
50    Rigidi    690.6798971    -1
26    French    960.6869577    -1
6.26    Kid Ca    781.8388778    -1
56.69    Idarah    690.1396637    -1

Totals

Bets    Wins    Win Strike Rate    Win %Return at Estimated Betfair Odds
All Selections    922    101    11.0%    70.6%


Results broken down by Year

Bets    Wins    Win Strike Rate    Win %Return at Estimated Betfair Odds
2004    103    15    14.6%    84.3%
2005    125    15    12.0%    77.6%
2006    134    14    10.4%    72.3%
2007    143    15    10.5%    65.1%
2008    155    14    9.0%    57.2%
2009    137    15    10.9%    76.5%
2010    125    13    10.4%    67.0%






The above is a laying system i derived off the massey site  - AUG/SEP  - Flat

How likely is this to perform well in the future ?

cheers fellas .

Let me calculate your AMPP figure(average monthly points profit) BBS

Your AMMP is 3.227 which means at £100 stakes you would make £322.70 per month profit

Pretty low frequency system and unless starting with good capital the infaltion period for your tank could take a year to reach maturity! profit is pretty robust against sample size but was the profiling based on therectical edges or back fitted?

Using the RANKING order and value in odds......I have STRATEGIES for :-

1. TOP 3 in market ( you can go to top 4/5 where JT/CFS etc)....

2. TOP 3 rated.

3. Laying those in lower part of ranking oder, where odds are value ( I will ignore certain rank positions re laying below the line ( line drawn at 50% + 1 of field size).

4. BACKING horses above the line that are within certain odds range ( you can get long runs of losers with this...but still profitable...theoreticaly...)



You can get days where a huge amount of points can be made.....and now and then days where you will lose a big chunk of points ( you can get a few bad days in a row....depending on which strategy you are using....)

Main thing is STAKE to take account of the "bad run" possibility....as they will happen due to there being SUBJECTIVE input in ratings....and I will have "off days" where brain not working well. Such days are few in number ...but if I have 2/3 days in a row where the subjective input is "wrong". the HITS can be very costly.

Generally though using RANKING ORDER and odds to determine bets produces PROFIT on every strategy mentioned above.

Sounds as if you are doing really well on your ratings BFF....wd


I'm still updating system re trainer stats.....BEFORE I can use system properly. It is such a boring "job" that I admit I have been more interested in keeping myself up to date re how race results are currently going.

Due to my personal circumstances, I just don't have time to use system properly...... however, I am finding that with small bets and ad-libbing from race to race can produce good profits, without resorting to HIGH STAKING.

I know that I should be using the system and strategies devised.....but it really takes a lot of energy and time to use system properly.....and currently i might continue just betting the way i am ( only started re HORSES just before CHELTENHAM using small stakes....and have made a profit from HORSE RACE bets amounting to £ 1,333.02 ( 12 days) .

Just doing what I am doing....I could up stakes by 5, or more,(could easily be financed)as there is sufficient liquidity for the type of bets placed.

It is frustrating not being in a position to use system properly.....but it is far easier, and less stressful, doing what I am currently doing. Finding it relatively "easy" to make £100 a day with small stakes......so say upping stakes by 5 on some races where I feel my edges are stronger could be done ....so will be doing so next week.

GL BFF with system and application of it ...though LUCK is irrelevant generally as you will know.

Cheers the nooooooooooooooooo

I cant agree more with the comment about rating and compiling data as being boring but as long as one keeps a long term goal to reaching a point of completion and maintain database daily with results and perform mod changes every so often this can be achieved ultimatlely!

It is also true that no matter how proftiable you are in the long run, losing runs and losing days will occur but this problem is more to do with correct staking and correct stake sizes are the only way forward  , you have to take the rough with the smooth!

Great content in this thread and a mine of information to help all ventures accross the minefield of betting

thanks for the replies lads

bf - its theoretical yes .

Are there any other sites you'd rec for researching info ?

bf - Where did you get your 322 a month profit from ?

All Selections    922    101    11.0%    70.6% over 7 year period as per your post brian
makes out for 3.227 points profit per month as (SR 29.5% X 922)/ ( 7 year of 12 months) delivers this figure so at £100 stakes makes £322.70 APPM.

£322.70 = £100L/s @ 3.227 APPM

bf - its theoretical yes .

Are there any other sites you'd rec for researching info ?

There are many sites that are useful but most charge for the data, the only free one is
the one we use , Adrians.

I am building my own database of races as I cant find all the races as a download
apart from a couple of ones that require an expensive payment .
I cant understand why the BHA dont sell the past races to punters for research purposes and also sell the current race cards like the do in America and Australia whos cards are brilliant.

bf_fananatic

Re. your Cheltenham ratings and mine which are shown below - several similarities so maybe we're using the same base source.

Bets: 81
Average BSP: 17.83
Wins: 13
Returns: 85.24 (BSP, 5% comm.)

Date	Time	Horse	Rank	Result	BSP	Returns
13/03/12	13:30	Steps To Freedom (IRE)	1	14th	8.91	-1.00
13/03/12	13:30	Cinders And Ashes	2	1st	12.00	11.45
13/03/12	13:30	Trifolium (FR)	3	3rd	8.53	-1.00
13/03/12	14:05	Al Ferof (FR)	1	4th	4.97	-1.00
13/03/12	14:05	Sprinter Sacre (FR)	2	1st	1.82	1.78
13/03/12	14:05	Cue Card	3	2nd	7.97	-1.00
13/03/12	14:40	Quantitativeeasing (IRE)	1	7th	7.80	-1.00
13/03/12	14:40	Our Mick	2	3rd	14.02	-1.00
13/03/12	14:40	Hold On Julio (IRE)	3	10th	7.25	-1.00
13/03/12	15:20	Hurricane Fly (IRE)	1	3rd	1.79	-1.00
13/03/12	15:20	Zarkandar (IRE)	2	5th	12.00	-1.00
13/03/12	15:20	Binocular (FR)	3	4th	7.59	-1.00
13/03/12	16:00	Scotsirish (IRE)	1	PU	5.03	-1.00
13/03/12	16:00	Dancing Tornado (IRE)	2	UR	14.29	-1.00
13/03/12	16:00	Garde Champetre (FR)	3	PU	19.07	-1.00
13/03/12	16:40	Quevega (FR)	1	1st	1.61	1.58
13/03/12	16:40	Violin Davis (FR)	2	15th	50.00	-1.00
13/03/12	16:40	Alasi	3	4th	51.74	-1.00
13/03/12	17:15	Triolo Dalene (FR)	1	11th	9.70	-1.00
13/03/12	17:15	Hunt Ball (IRE)	2	1st	9.00	8.60
13/03/12	17:15	Bless The Wings (IRE)	3	10th	11.81	-1.00
14/03/12	13:30	Four Commanders (IRE)	1	3rd	14.33	-1.00
14/03/12	13:30	Harry The Viking	2	2nd	9.40	-1.00
14/03/12	13:30	Alfie Spinner (IRE)	3	5th	10.50	-1.00
14/03/12	14:05	Sous Les Cieux (FR)	1	5th	7.00	-1.00
14/03/12	14:05	Monksland (IRE)	2	3rd	6.87	-1.00
14/03/12	14:05	Simonsig	3	1st	3.31	3.19
14/03/12	14:40	Grands Crus (FR)	1	4th	2.42	-1.00
14/03/12	14:40	Bobs Worth (IRE)	2	1st	5.70	5.47
14/03/12	14:40	First Lieutenant (IRE)	3	2nd	6.20	-1.00
14/03/12	15:20	Sizing Europe (IRE)	1	2nd	1.96	-1.00
14/03/12	15:20	Finians Rainbow (IRE)	2	1st	5.90	5.66
14/03/12	15:20	Big Zeb (IRE)	3	3rd	8.80	-1.00
14/03/12	16:00	Third Intention (IRE)	1	8th	27.00	-1.00
14/03/12	16:00	Balgarry (FR)	2	7th	7.45	-1.00
14/03/12	16:00	Get Me Out Of Here (IRE)	3	2nd	8.74	-1.00
14/03/12	16:40	Vendor (FR)	1	3rd	4.97	-1.00
14/03/12	16:40	Kapga De Cerisy (FR)	2	11th	36.43	-1.00
14/03/12	16:40	Une Artiste (FR)	3	1st	55.32	52.60
14/03/12	17:15	The New One (IRE)	1	6th	19.48	-1.00
14/03/12	17:15	Horatio Hornblower (IRE)	2	15th	14.16	-1.00
14/03/12	17:15	Village Vic (IRE)	3	13th	22.03	-1.00
15/03/12	13:30	Sir Des Champs (FR)	1	1st	4.70	4.52
15/03/12	13:30	For Non Stop (IRE)	2	3rd	10.50	-1.00
15/03/12	13:30	Cristal Bonus (FR)	3	PU	8.83	-1.00
15/03/12	14:05	Restless Harry	1	11th	9.28	-1.00
15/03/12	14:05	Rick (FR)	2	6th	44.00	-1.00
15/03/12	14:05	Reindeer Dippin	3	5th	82.14	-1.00
15/03/12	14:40	Rubi Light (FR)	1	5th	7.43	-1.00
15/03/12	14:40	Somersby (IRE)	2	7th	5.81	-1.00
15/03/12	14:40	Riverside Theatre	3	1st	4.39	4.22
15/03/12	15:20	Big Bucks (FR)	1	1st	1.87	1.83
15/03/12	15:20	Mourad (IRE)	2	6th	50.72	-1.00
15/03/12	15:20	Voler La Vedette (IRE)	3	2nd	35.79	-1.00
15/03/12	16:00	Notus De La Tour (FR)	1	F	9.04	-1.00
15/03/12	16:00	Hectors Choice (FR)	2	17th	16.50	-1.00
15/03/12	16:00	Giorgio Quercus (FR)	3	14th	23.40	-1.00
15/03/12	16:40	Start Me Up (IRE)	1	5th	13.50	-1.00
15/03/12	16:40	Adams Island (IRE)	2	19th	2.40	-1.00
15/03/12	16:40	Up The Beat	3	4th	10.19	-1.00
16/03/12	13:30	Grumeti	1	3rd	6.37	-1.00
16/03/12	13:30	Baby Mix (FR)	2	9th	8.40	-1.00
16/03/12	13:30	Countrywide Flame	3	1st	46.00	43.75
16/03/12	14:05	Raya Star (IRE)	1	13th	10.50	-1.00
16/03/12	14:05	Citizenship	2	14th	7.40	-1.00
16/03/12	14:05	Sailors Warn (IRE)	3	3rd	22.37	-1.00
16/03/12	14:40	Boston Bob (IRE)	1	2nd	2.50	-1.00
16/03/12	14:40	Rocky Creek (IRE)	2	8th	18.10	-1.00
16/03/12	14:40	Brindisi Breeze (IRE)	3	1st	9.00	8.60
16/03/12	15:20	Long Run (FR)	1	3rd	2.86	-1.00
16/03/12	15:20	Kauto Star (FR)	2	PU	4.50	-1.00
16/03/12	15:20	Midnight Chase	3	7th	17.46	-1.00
16/03/12	16:00	Enter Paradise (IRE)	1	UR	44.00	-1.00
16/03/12	16:00	Bradley	2	8th	38.00	-1.00
16/03/12	16:00	Keenans Future (IRE)	3	12th	216.54	-1.00
16/03/12	16:40	Molotof (FR)	1	13th	17.18	-1.00
16/03/12	16:40	Make A Track (IRE)	2	4th	19.97	-1.00
16/03/12	16:40	Imsingingtheblues (IRE)	3	18th	27.69	-1.00
16/03/12	17:15	French Opera	1	F	26.00	-1.00
16/03/12	17:15	Kid Cassidy (IRE)	2	12th	6.26	-1.00
16/03/12	17:15	Astracad (FR)	3	7th	7.80	-1.00

N.B.  Removing the two longest priced winners (which amounts to just 15.4% of the total number of winners) alters the picture completely and would have resulted in a level stake loss. 

Did you guys not lay any horses at Cheltenham ?

Due to number of runners in most races, I would back many and lay 1 or 2..... Now I know this was me only "ad-libbing", using my brain, and not bets originating from my "system" use ( though the brain does know what the numerous factors are that are generally relevant to each race), however, LAYING in such large field races , as well as BACKING is one way to make a prfit from such a meeting.

I noticed the 2 high odds ones as well BFF...and how they would be ignored by most statisticians ......however, such high odds winners are "bread and butter" if your system of ratings produces such bets TIME AFTER TIME .

I expect to get high odds winners......they are "normal" for me to find or my system to find. They should not be excluded from stats....but just noted, as they do effect figures.

The fact is TTR......BFF got the winners in their selections......so "removing the two logest priced winners" is distorting the facts....

STATISTICIANS probably would not find such winners

.BFF got the winners in their selections -----should be ---.BFF got the winners in the selections

great content in this thread?

Where?

Glasgow Brian - any system which uses market position of the horse is going to be a nightmare when it comes to putting your money down in reality.

You are looking for the fav in your system. But what horse is fav?  OK a lot of the times it is obvious and doesn't change, but what if there are two or three at the top of the market, flip-flopping about. What if one horse if fav on BF but not on the show? Do you lay it or not?  Also when do you lay?  10am your "fav" could be out with the washing as it drifts come race time.  Just before the race - then you have to stick at your PC right until the off.

It's all very well looking back in time and saying the fav did this, that or the other - but that is when you know what the fav was.  Even worse if you have a system that says look at the top 3 horses in the market as somebody else said later on this thread.

TheVis - Of course you are correct that the horses at top end of market will change position as FAV/2nd FAV/3rd Fav ....or be JF or CF etc .

However, it is the odds that horse is at that are important near the off.....not whether a hoprse is FAV or not. Using the RANK position in ratings to indicate whether odds are a value back or lay is what determines whether a bet should be made or not.

I have not tried to do live runs for STRATEGIES yet ( system not updated re trainer stats..at present)......however, in theory, it will be the odds of probably the top 6 in market that will be checked firstly from LAY possibility. If the odds are value to LAY, then horses will be layed....the RANK position remaining STATIC .

A BOT set up would be "best" way to apply strategies.....but initially manually is what will be used.

I agree that the practicle aspects of applying strategies pose difficulties re manual approach, but as rewards are high, then this reflects the difficulties that will be encountered.

racing about to start.....so GL with bets today.

The forecasted BSP at time form is pretty accurate but a pain in the rear to convert in its current format and although you cant always be sure what is going to be favorite you can get a good idea of the price band that all the runners will occupy, non runners affect the market much more but you can use auto-betting apps to have a reserve set up using a ranked set-up
The bet engine allows this type of bet automation and I can recommend it totally having tested it out.

so , bf

What you are saying is if you have logical criteria , maybe 900 selections may well be enough ?

The only systems apart from trading that truly work or those that have underlined
types of form or data that increase the strike rate above or below the implied chance via better than standard odds compilation and other backers changes to the market when live.

But what the real question boils down to is that are the methods that are used to locate
mis-priced selections in reality genuine valued calculations or products of the continually shifting sands of natural variation that move peaks and troughs of returns across the spectrum
of available data that can mislead.

To find this answer and to answer your question Brian , 1 to infinity may be either profitable or a loss as natural variation smooths out.

BF
Please could you dumb down your posts a bit please ?.
I don't know about others, but my head is spinning with the apparent complexity of them.
In fact I have no idea if what you are saying most of the time is absolutely brilliant or absolute codswallop.
No offence meant btw.

LOL FAFH try this one for size....All theories and calculations used to work on the subject matter of probability of an event are only rational and thus useful at certain points in time and beyond certain points in time they become void.

E.G. all markets in betting of sporting events that involve chance are meaningless once the race is finished at that point in time, as the winner has not even 100% chance of winning though it has won as it cant be a predicted chance anymore but a conclusion of the probability calculation that itself is now redundant in its useful form.

The logical conclusion of the above observation is that

1 X probability = 0 x result and also 0 x probability = 1 X result

But how can this be as 1 cannot equal zero unless the 2 separate multiplying factors are them
selves opposite thus 1 X 0 can equal 0 x 1, 1 and zero are in fact opposite as one exists and zero does not.

So basically you cannot have probability and result in the same time as both only exist in separate times, one the future and one the past which are opposites.

Well if you have read up to this point and wondered what the hell is this guy on about well
its a lesson in pure observation and logic and also leading up to a powerful quotation which I want all you Betfairians to remember.

If you want time to spend on a result then spend time on probability

And for sanity's sake don't waste time understanding someone else's version of probability
like you have just done now and when you look at the average tips and prices that are floating around.

So after all this, what is the minimum sample size needed to know whether you are a winner of not please?

So after all this, what is the minimum sample size needed to know whether you are a winner of not please?

There isn't a simple answer to that - in theory you can fluke an unlimited number of results plus it depends on the way in which your samples are derived.

That said; if you're looking for a rule of thumb - at odds of 3-ish you'd probably want a couple of thousand. At significantly longer or shorter odds you'd need more. Whatever you think you need - it's always more.

There are various statistical tests you can do to assess the strength of a correlation but personally I've never found them that useful, there are however a couple of simple heuristic tests that I use all the time:

1: Look at the neighbors - IE those that almost met your criteria but didn't quite. Say you're scoring selections out of 10 and backing those that scored 8 or above, how did the 7s do? If your strategy is good you should see a sloping improvement across the score range from negative to break even to positive. The 10s should do better than the 9s and so on. If you're seeing peaks and troughs you're either not onto anything or you don't have enough data. How you construct the neighbors depends what you're analysing. Say you've got an indicator that's good for 5f races (and the distance is somehow relevant) it should still roughly hold up for 6f, a bit less for 7f and so on.

2: Divide your data. The stronger a trend is the smaller the data set needs to be in order for it to be evident. If you divide your data in half you should see the broadly the same trend in both sub-sets. The better your indicator the more times you can make the division whilst still observing the trend in each sub-set.

3: Check the opposites. If your indicator highlights good value to back then selections with the opposite characteristics should be value to lay. If they're not then you've got to ask yourself why.

4: Check your reasoning and your process. First; just ask yourself honestly whether what you've 'found' is probably just b0llocks. If you cant explain convincingly WHY a trend exists then you should treat it with great skepticism. Ditto if what you've found is incidental - IE you were expecting or looking for something else. If you came up with a theory, made a prediction then observed it then that's great. If you just went fishing or back-fitting then you not only need a bigger data set but a different data set - i.e. one from the future not the past.

Above all - decide your methodology BEFORE you do your analysis. Otherwise all you'll do is keep going through different techniques until you find one that makes it look like you're going to be rich. That won't work. Not ever.

LOL FAFH try this one for size....All theories and calculations used to work on the subject matter of probability of an event are only rational and thus useful at certain points in time and beyond certain points in time they become void.

E.G. all markets in betting of sporting events that involve chance are meaningless once the race is finished at that point in time, as the winner has not even 100% chance of winning though it has won as it cant be a predicted chance anymore but a conclusion of the probability calculation that itself is now redundant in its useful form.

With respect - this is just wrong. Probability is a property of information, not of an event. When I say that a there as a probability of 0.5 that a flipped coin will land on heads I'm not saying anything at all about the coin flip. I'm saying that the information I have about the coin flip allows me to predict heads at an accuracy of 0.5 (50% if you prefer). This remains true if the coin flip has already taken place. If you don't understand this then you don't understand probability at all.

Thanks for your replies.

Beginning of 2015 I decided to take things seriously with the horses, after 30 years of failing & subscribed to a ratings/tissue service and started keeping results. At one point I was £12k up, then lost the lot. I finished the year £3.2k up. Last year exactly the same thing happened I was £6k and again lost the lot. It forced me to look back at my records and it seemed to me the races that were killing me were big field competitive races, handicaps, (the type of race Pricewise tips in) backing more than one in a race and backing ew. Beginning of August last year I decided to change my strategy.

Now, I only consider races where the horse at the head of the betting on the tissue is 4-1 or less, so open races where they are 9-2+ the field on the tissue I use are an instant swerve for me now, and I only back horses that are 4-1 or less on the tissue to try and cut down the horrendous losing streaks I have had (and still have) I very rarely back more than one in a race and I never back ew.
Obviously although the tissue price on a horse I'm looking at is 4-1 or less I'm looking to equal or beat these odds when I place a bet. With a nearly a year behind me here are the results:
Month: in (returns £) out (amount staked £)brackets: profit/loss Return On Investment strike rate
Aug 2016 in: 10618  out: 9470 (+1148) ROI: 12% 11/68 16%
Sept 2016 in: 15257  out: 13975 (+1282) ROI: 9% 21/102 21% in: 25875  out: 23445 (+2430) ROI: 10% 32/170 19%
Oct 2016 in: 13364 out: 10490 (+2874) ROI: 27% 24/100 24% in: 39239  out: 33935 (+5304) ROI: 16% 56/270 21%
Nov 2016 in: 8825  out: 5850 (+2975) ROI: 51% 15/57 26% in: 48064  out: 39785 (+8279) ROI 21% 71/327 22%
Dec 2016 in: 14471 out: 15200 (-729) ROI: -5%  24/127 19% in: 62535  out: 54985 (+7550) ROI: 14% 95/454 21%
Jan 2017 in: 28520  out: 21976 (+6544) ROI: 30% 44/168 26% in: 91055  out: 76961 (+14094) ROI: 18% 139/622 22%
Feb 2017 in: 21716  out: 22721 (-1005) ROI: -4% 36/159 23% in: 112771  out: 99682 (+13089) ROI: 13% 175/781 22%
March 2017 in: 60901  out: 54198 (+6703) ROI: 12% 60/243 25% in: 173672  out: 153880 (+19792) ROI: 13% 235/1024 23%
April 2017 in: 47947  out: 55752 (-7806) ROI - 14% 34/191 16% in: 221619  out: 209640 (+11979) ROI: 6% 269/1215 22%
May 2017 in: 71209  out: 70264 (+945) ROI : 1% 52/237 22% in: 292828  out: 279904 (+12924) ROI: 5% 321/1452 22%
June 2017 in: 38766  out: 37978 (+788) ROI: 2% 25/131 19% in: 331594  out: 317882 (+13712) ROI: 4% 346/1583 22%

The figures in italics are rolling totals. At the end of June I had staked £317882 which returned £331594 for a p/l of £13712 - a ROI of 4%. I've backed 1583 horses, 346 of them have won for a 22% strike rate. What I'm worried about is, my strike rate has remained pretty consistent around 22% but my ROI is steadily decreasing. Am I reverting to the mean, have I just been lucky, will I eventually become a loser or is 4 maybe 5% a realistic roi and I should continue. Any thoughts would be appreciated, and feel free to ask any questions. TIA

McCoy,

How many hours "study" would you say you do in a month?

I mean, do you work as well as bet? Or are you able to devote all of your time to this?

Anyway, on the face of it, I think your 4 or 5% is very realistic and 1583 bets is a decent sample
(for what my opinion's worth!) so carry on and well done !

Roughly 2 or 3 hours a day, including putting bets on. Normally up between 7 and 8, turn on the computer and get started. Used to be 4 or 5 hours, but by avoiding races where they are 9-2 plus the field I've cut down the number of races I need to study, which is good because I ended up burnt out. No, it's not full time, I have a business to run, but it's seasonal and I have a lot of time on my hands especially during the week and for 4 months in the winter the business closes down.

Probability is a property of information, not of an event.

Agreed which is why using the concept of 'probability' in horse racing and sporting events (even 2 outcome events) is fraught with danger.

"Chance" is a safer concept to consider---if you are that way inclined-----being a conspiracy theorist I doubt if concept of "chance" even in the most calculated sense is wise but that is another story.

Gamble responsibly

I must read this thread in detail as it looks like some people know stuff.
For a long time I've been working on pedigrees.
At the start of 2017 I wrote programs to analyse pedigrees.
My sample size is 159,000.
It was a bit of a moment to look at the results for the first time to see if years of work were pointless or not.
Results were favourable.
It isn't for betting, although I might use it as a guide later.

I agree with all the points aye robot writes about - it's solid advice. Just to complicate things though, I think it's also possible to have *too large* a sample size and it can give you overconfidence.

Imagine you are looking back at past races and discover a profitable betting strategy, with a reasonable basis for why it makes sense, and so on. You might have found an edge, fantastic! The problem is, edges tend to disappear over time, as other punters grow wiser. The flip side of this is that the further back in time you look, the larger the edge is likely to be. So if you test your strategy over more races, say the last two years of racing, you could find it produces better results than it does over just the last year. You might think that this is great news and confirms your strategy, and it does, but it can also make your edge look far bigger than it really is.

TheInvestor's comment is also true - beware when increasing your stakes, it will always reduce your ROI. You will always get more money matched on losers than winners.

Imagine you are looking back at past races and discover a profitable betting strategy, with a reasonable basis for why it makes sense, and so on. You might have found an edge, fantastic! The problem is, edges tend to disappear over time, as other punters grow wiser. The flip side of this is that the further back in time you look, the larger the edge is likely to be. So if you test your strategy over more races, say the last two years of racing, you could find it produces better results than it does over just the last year. You might think that this is great news and confirms your strategy, and it does, but it can also make your edge look far bigger than it really is.

This is a fair point, though I would say it's an easily avoidable trap if you're graphing your data - which you DEFINITELY should be. In this case you'd see the trend emerging visibly over time if you binned (grouped) all your results by date and graphed the mean result.

More widely I find graphical approaches to data analysis tremendously useful. Any phenomenon that's strong enough to exploit will almost certainly be readily apparent on a graph. It's my "go to" approach for edge discovery. I only really use more advanced statistical tools to work out how indicators relate too each-other, all the foundational work finding out what's relevant and comprehending how the markets work is done with graphs. I have tried other approaches but I've never got them to yield much. I even hired a statistician to look into some data for me, he was a university professor who specialised in machine learning so he was qualified, but even he didn't come up with anything I didn't already know about.

very interesting thread....

i have been tracking a system since the beginning of october 2016. it is based on a very specific set of criteria and is a back betting system. from the start of october to the beginning of May there have been 203 horses that met the criteria delivering 101 points profit (calculated after paying a 5% commission).
if a horse hits the criteria it should be backed at any price.
what i have noticed lately is one month delivered 70 points profit, another 40 points and the rest were either flat or at a loss to deliver the 100 points profit.

Do you believe it has just been lucky so far (based on the 2 high performing months) or that it potentially could be an edge?
does a system need to deliver on a more consistent basis over time to be considered legit?

id really appreciate some feedback.
ps i have data up until now but need to enter it into the template (its quite arduous) and can update at the end of the month with a full years data (there are some gaps)

Peter May
@PeterMayRacing
·
6h
My latest blog piece is already on the ATR website.  It covers an approach based 3yos in all age handicaps which I referred to in my last book back in 2013.  Good to see it still working.

well beaten last time out ,drop in class ,1st 2nd fav ,+24 off24 bets 27 days 100%roi, longest losing run 2