[roasthawg]

let's say you have a system you're thinking about trying and you want to find the expected win% in order to bet kelley. say the system has gone 70-30 in the last 100 games it said to bet on. how do you go about determining expected win%, do you find the confidence interval at 99% or what? it's been a minute since i had a stats class and i can't figure out how to do this for the life of me...thanks.

[Justin7]

To approximate for games near 50%:

Take square root of sample size, and divide by 2. For 100, this would be 5. Every 5 above 50% would be on deviation.

If your system went 60-40, that would be 2 standard deviations, or about 98% confidence.

On another note, if you're "data fishing", you should expect to find one system like this for every 50 you check, even if there is no logic to it. The more crap you search, the more crap you find.

In addition to having good statistical results, you need to have a logical basis why this system is better than flipping coins. And if your idea is simple, it's a fluke (other people would already know it, and the market would correct for it).

[roasthawg]

Originally Posted by Justin7

To approximate for games near 50%:

Take square root of sample size, and divide by 2. For 100, this would be 5. Every 5 above 50% would be on deviation.

If your system went 60-40, that would be 2 standard deviations, or about 98% confidence.

On another note, if you're "data fishing", you should expect to find one system like this for every 50 you check, even if there is no logic to it. The more crap you search, the more crap you find.

In addition to having good statistical results, you need to have a logical basis why this system is better than flipping coins. And if your idea is simple, it's a fluke (other people would already know it, and the market would correct for it).

so if you're system went 60-40 are you saying you would use 55% for your expected win%? thanks for your help...

[Justin7]

If you went 60-40, you are 98% confident that your system is at leat 50/50. At 55%, you're only one standard deviation above, which is about 68%. But see my many warnings on this.

[roasthawg]

Originally Posted by Justin7

If you went 60-40, you are 98% confident that your system is at leat 50/50. At 55%, you're only one standard deviation above, which is about 68%. But see my many warnings on this.

i definitely understand the risks of any system betting strategy, but gambling itself is a risk right? also, i agree 100% that a system must be backed up by logic. thanks for your help on the win% estimate...it appears that as far as kelley goes it might be better to just estimate the percentage at around 53 to 55%.

[durito]

None of that suggests anything about what rate the system will predict at going forward. It simply explains what has happened. Just because an angle has been statistically significant looking backwards, that doesn't mean it will be going forward.

[Data]

Based on the information you have you can only calculate a range where "true" win-rate should fall with any given confidence level. The problem is that as soon as you move your confidence level away from zero the range gets very wide making it unusable for calculating Kelly stakes.

[roasthawg]

Originally Posted by durito

None of that suggests anything about what rate the system will predict at going forward. It simply explains what has happened. Just because an angle has been statistically significant looking backwards, that doesn't mean it will be going forward.

fa sho unless it's a significant system of course.

[roasthawg]

Originally Posted by Data

Based on the information you have you can only calculate a range where "true" win-rate should fall with any given confidence level. The problem is that as soon as you move your confidence level away from zero the range gets very wide making it unusable for calculating Kelly stakes.

exactly, i've been doing pretty good by just estimating a win rate but i wanted to see if i could get a lil more scientific...at this point i think a conservative estimate is probably the best way to go right?