sample size

[Rastis]

Hi Folks,

I have been testing a football trading theory and have compiled data on about 400 trades spanning 5 months.

I was hoping for an opinion or two re whether this is a large enough sample to start having confidence that it isn't a fluke.

Cheers

[rubysglory]

Hi Rastis,

Disregarding the actual act of putting down your hard earned for a wager, and the Psychological effect that has, from a statistical viewpoint, probability theory is a generalised inference of a given outcome from an independent repetition of a random experiment. Given the independant and random nature of gambling on sports, words of wisdom may be had from the song "History Never Repeats".

Generally speaking however, with such a sample base, I would say that you should be able to assess with some degree of confidence wether or not your theory is a complete waste of time. IMO opinion however, what is of most importance in any strategy, is the ability to withstand an extreme run of outs. What ever your sample shows as your worst case loss sequence, double it ! If results are positive this in itself should provide you with some degree of postiveness to at least start playing for real - with small stakes.

rg

[donfede]

It also depends on the number of possible outcomes.

If your data is about correct score, 400 data is not too may because there are 17 possible outcomes.

If it's about over/under market 400 might be ok, because there are two possible outcomes, like a coin flip.

Have you tried your statistics on historic data as well, i.e year 2009 or 2010?

[Fidib665]

There is no easy way to answer this - it depends on the market you are playing in as well as the average quotes. As Donfede pointed out, in the CS market it might not be a sufficient sample size given that you have 17 possible outcomes. But as important: If you are always betting on, say bets with a quote of 50 - and if you have exactly one such bet per game - then you would expect in your 400 game sample 8 bets to win. If by chance, 10 quotes win (which could be easily possible due to randomness), this would show you a highly profitable strategy with ~20% profit per game. Clearly sample size is too small there to make any correct evaluations there.
If you obtain the same 20% on average for 400 games with quotes of 1.5 each, I would say, the chances are much better that you really have discovered something.

I have ~14.000 games to test my strategies on - and still the statistics often tricks me in believing that something is profitable which out of sample is not ...
Cheers, F

[Rastis]

Thanks for the input Folks.

As I suspected; if there has to be a definitive answer right now - even when considering that the trades are very carefully selected (the 400 trades came from 6000 - 7000 possibilities), the answer would be that the sample is too small.

Thus, I suspect all I can do is carry on compiling and see where it leads...

All the Best

Rastis

[freddy]

Why not just use small stakes and build up ?

Paper trading ok for a while but you will learn a lot more allot quicker with small stakes.

[Fidib665]

Depends very much on what you are doing. Might be necessary for trading but for punting it would lead nowhere...

[Iron]

I want to test whether the profitability of a betting system is due to random chance using the Chi Squared test.

What's the minimum sample size I'll need for the results to be statistically significant?

Thanks

Jeff

[rubysglory]

Jeff, that would depend I suppose on if any deviation in the resultant data was itself known to be caused by chance or perhaps some other factors.

rg

[Iron]

I don't see what's so funny.

Let's use an analogy. I think a coin is loaded, and is more likely to give me a head than a tail. If I throw it 4 times, and get 3 heads and 1 tail, you'd tell me that proves nothing, as it could easily be due to random chance. If I threw it 4000 times, and got 3000 heads, then my guess is that you'd see it as compelling evidence.

Similarly, how many bets do I need to have in my sample to be able to undertake statistical tests which can tell me whether it was extremely likely that the positive ROI I obtained was due to random luck?

Jeff

Jeff, that would depend I suppose on if any deviation in the resultant data was itself known to be caused by chance or perhaps some other factors.

rg

[CaerMyrddin]

Independently of your sample size, the test will always come up with some kind of metric that will allow you to understand where you are