This system: good, bad, or so-so?

[jackad]

I've developed a system that shows these results in 288 races: 21% win frequency: ROI 12%.

I will greatly appreciate your reactions to these questions:

1) Given the sample size, what is the liklihood of these results holding true in the future?

2) Do you consider these results to be bad (you do much better with your approach), good, or just so-so?

3) Using this system, is it reasonable for me to bet 2% of capital on each selection?

Thanks for your responses.
Jack

[tanda]

Briefly and assuming completely objective testing,

1) Not as likely as many would think (the sample size is not all that large);

2) Decent results;

3) 2% of bankroll is too much.

[Larry Hamilton]

you havent asked enough questions. Lets assume that eveyrthing you just said is true and satisfactory...You have a secret money maker that makes 2% profit. Exactly, what does a 2% profit mean?

Well, if you are comfortable putting $1,000 a week thru the window, you can expect to make $20 bucks. Ok, lets kick it up and see what happens. Put $10,000 thr the window, you make 200 bucks. (minus expenses)... Put $100,000 thru the window in one week and you can finally make a living--2k a week.....by the way, you can get nearly 3% in a savings account...and never mind the potential LOOOOOOOOOOOOOOOOng losing streaks that can accompany a 21% win rate, which is actually the flip side of a 79% losing rate.

Whats the question you forgot to ask? Do you think my wife will let me put 100,000 a week thru the windows?

[GameTheory]

Agree with tanda on all 3.

I would also like to point out how important it can be HOW you got those results on that particular sample.

If you got those results completely independently of creating your method, and they were "unseen" races, chances are much better of repeating those results on another sample. If you tweaked things here and there, and ran tests on the same sample repeatedly until you got some decent results, it is quite likely that your results will never be duplicated and you've simply "fit" that particular sample.

Either way, you need to test it on a bigger sample. 2% bet is too much, although if you always bet 2% of your CURRENT bankroll that it much different than starting out with unit bet size of 2% of your starting bankroll and betting flat. Not sure which you were talking about...

[ranchwest]

I basically agree with the previous comments.

You don't say how selective you were in coming up with these 288 plays. If feasible, I'd suggest coming up with 288 more plays and see if those have a similar result without any modifications to the system.

Since you don't seem very certain of what you've got, I'd say that what you currently have is something to play $2 on for entertainment. When you have a money maker, you'll know.

[andicap]

Originally posted by Larry Hamilton
you havent asked enough questions. Lets assume that eveyrthing you just said is true and satisfactory...You have a secret money maker that makes 2% profit. Exactly, what does a 2% profit mean?



Uh, Larry. He said 12% profit, not 2%. He's betting 2% of bankroll.

[jackad]

Thank you all for your responses. They're helpful. I can conclude that my sample size is too small to be predictive of future results, snd that betting 2% of my starting bankroll and flat betting is too much.

May I ask specifically:
!) What do you think of a 12% ROI using any system (tanda described it as decent)? What return do you get or look for when betting?
2) Assuming an approximate win frequency of about 20%, what percentage of starting bankroll would you recommend for flat-bet size? What about if my win frequency was about 30% (I'm thinking about another system)?

Thanks.
Jack

[ranchwest]

When I get a return on previous races, I like to assume that I may only get 1/2 that return on future races. Then, I take it easy with it until it is apparent that it is predicting as well as I had looked into the past. So, if I want an ROI of 1.12, then I model a 1.24 return and see if I get at least 1.12. Personally, I attempt to be more selective, with a higher return. I mostly go for high percentage plays.

[anotherdave]

Originally posted by jackad
May I ask specifically:
!) What do you think of a 12% ROI using any system (tanda described it as decent)? What return do you get or look for when betting?


My actual results are not far from yours. I am currently running an ROI Of about 12.5% and have a winning percentage of about 27%.(Although 21% compared to 27% can make a difference in losing streaks) I bet 1% flat and find that comfortable.

AD

[andicap]

You could make a living on 12% ROI, especially if you use off-shore bookies who give you a rebate.

[Larry Hamilton]

ooops, never mind

[Lefty]

Hey Larry, now he only has to bet one sixth of that hundred grand. Piece of cake.
Seriously it's amazing how much money has to be put through the windows. Good news is a lot of it can be SAME money, the old "churning" effect.

[superfecta]

Originally posted by jackad
Thank you all for your responses. They're helpful. I can conclude that my sample size is too small to be predictive of future results, snd that betting 2% of my starting bankroll and flat betting is too much.

May I ask specifically:
!) What do you think of a 12% ROI using any system (tanda described it as decent)? What return do you get or look for when betting?
2) Assuming an approximate win frequency of about 20%, what percentage of starting bankroll would you recommend for flat-bet size? What about if my win frequency was about 30% (I'm thinking about another system)?

Thanks.
Jack Question 1-a 12% return is not bad if you are steadily hitting a winner,and not riding out long losing streaks.You have to pay yourself for being at the track and the time you spend figuring races.So if you can do that,and bet the needed amount of money,more power to you.I personally look for an overall return of 25-1 for every dollar I bet,per race,not overall ROI.So in most instances,since I'm betting exotics,I need the higher return to justify each bet.I have to project what the bet will pay,and be honest with myself,and be willing to pass races that I might hit ,but have a greater chance of losing.Sometimes I get burned,but thats the risk I am willing to take.
Question 2-If you are asking for a specific amount,can't tell you.Once again,you have to factor in how much your time is worth,and decide what bet unit will get you paid if your win % holds up.Then multiply that unit by 20.So if you run a string of losers ,if you lose 20 consecutive win bets,something is wrong with your betting angle.

[Foolish Pleasure]

Hi Jack,

Here is a generalized formula which allows for a different bet size and a different bankroll goal.

IBR = initial bankroll
b = bet size ("to win") as a % of BR.
L = total number of losses
W = total number of wins

Now that "b" is generalized, the effect of a winning bet on the bankroll is a factor of (1+b) and the effect of a losing bet is (1-1.1*b). So:

BR = IBR * [(1-1.1*b)^L] * [(1+b)^W]

Let's now define a "goal" of G = BR/IBR. If we are trying to turn $100 into $100K, G=1000. Dividing both side by IBR:

BR/IBR = G = [(1-1.1*b)^L] * [(1+b)^W]

N = number of races
p = probility of winning each wager

So, W=N*p and L=N*(1-p). Plugging these values in for W and L and solving in the similar way as before yields the generalized solution below. I am skipping the steps because there are about ten of them and they are difficult to follow when typed like this. Once again, the base of the log is arbitrary. You will find if you plug in G=1000 and b=.05 and then multiply by some constant, it will produce the same formula as in the previous example.

***** BIG GENERALIZED RESULT **********

N = log(G) divided by the quanity "log(1-1.1*b) + p*log[(1+b)/(1-1.1*b)]"

as long as the denominator ("the quanity") is positive. (There were too many parentheses, I think adding the words makes the formula more clear.)

That formula allows one to calculate the number games (N) needed to meet any goal (G), using any bet size (b), with any winning probability (p).

**********************************************

If the denominator of the above formula is negative, the betting strategy will never reach the goal without really strong luck; and if the betting strategy does get lucky and meet the goal at some point, there is zero chance that the goal will be maintained in the long run. The denominator will be positive when:

log(1-1.1*b) + p*log[(1+b)/(1-b)] > 0

p > -log(1-1.1*b) / log[(1+b)/(1-1.1*b)]

This formula allows one to calculate the MINIMUM winning percentage to experience ANY STEADY GROWTH AT ALL for a given bet size.

Please at least attempt to actually do the work, I'll be back next weekend to field questions.

[alyingthief]

two of mitchell's works--thoroughbred strategies, and common sense betting (this is the better exposition)--contain explicit and useful formulae on determining sample sizes necessary for determining the differing confidence levels (ie, 90,95, and 99% confidence require successively larger samplings) of a proposition. (confidence is a general statistical term that points at the number of times an event will occur in a certain region of a gaussian, or "bell" curve--99% means that only one time in a hundred will an instance fall outside the general region the formulae have determined as probable). these equations generally solve most handicapper's questions about the pertinence of a handicappinng methodology--but will only do so within parameters..so that if you have a win rate of 21% and an ME of 1.12 (12% roi), you may find that indeed it is viable, but will prove unprofitable too many times to be useful--that is, for every thousand bets, sometimes you will have a negative return, break even sometimes, sometimes have a greater return.

this is probably the case with your sample. additionally, there are other questions to be answered before someone can arbitrarily make a good estimate: how many of your winners were very large? did you discard these results as anamalous when you determined your return? etc., etc.

since i don't have the books open to the page i can give you a good ball park estimate on how many races you will need to examine to insure a high probability of success at the limits you have listed: certainly more than 2500.

[tanda]

We also need to know the risk you are willing to accept of tapping out.

Assuming that you have an actual 21% win rate (and the above tests of confidence levels can lead you to an answer), you can calculate an appropriate flat bet amount based on your risk tolerance. For a percentage of bankroll, the calculations are more complex but possible.

[hurrikane]

Whoa, when someone does that math let me know the answer. It's too late in the day for me.

2 things.
1. when you run your sample and get your results you need to run again over a completly untrained set of data. Otherwise I would guess 9 out of 10 times..you backfitted to the data.

2. Are you willing to stick it out through the losing streak. At 21% you could possibly take a 50 hit losing streak...are you going to stick it out to 49 and then fold up your tent and go home(that is, if you haven't tapped out). If so you lose.