Trading Probability


one of the things i like about this site is it's based a lot on statistical probabilities and edges off the initial balance.

ive been developing a nice spreadsheet of data on the ES to base my trading plan off of and was wondering what kind of sample size do you all think is ample enough to get a good edge?

i can condense my 10 years of ES data down to 10 days, you think thats significant to trade off of or what?

just curious if any of you are using historical probabilities to base a trading plan around.

thanks!
I used to consider 300 trades in backtesting as a minimum, however for the longest time I ignored the need to have 2 sets of data in backtesting, the "in-sample" one being used to develop & optimize the trading strategy, the "out-of-sample" one to verify the strategy on virgin data.

Of the dozen+ systems I developed, the only one that has (so far) survived is the only one for which I used that "in-sample" / "out-of-sample" testing methodology. I had about 16 months of data for 200 trades "in-sample", and 18 months of data for 230 trades "out-of-sample". So far 70+ live trades for that system, and a P/F of 1.5 on that sample, vs 1.8 on the aggregate 430 trades in backtesting.


If you have used those 10 years of ES data to develop & optimize your strategy, I would advise to use some additional data for the "out-of-sample" validation. Unless those 10 years represent 1000+ trades, in which case there is a little less risk of having curve-fitted your strategy.
here is a link that may provide some useful ideas.

http://www.linnsoft.com/homework/


In general I think it would be wise to not only look at the traditional IB in the ES but also use the first 90 minutes of trade. Lots of great fading trades happen off that range.
I agree that Dom's approach makes a lot of sense. In Dom's case he's testing and trading the same market so he would need to carve out some historical data to develop the strategy with and then the balance to test with.

Some traders will develop a strategy on one symbol with the intention of trading it on multiple symbols. In cases like that you might be able to develop a strategy using all the data you have on the ES and then test it against the YM, TF and NQ. However, there is certain degree of correlation in those markets so you will not end up with a perfect test.

Good idea about using the 90 minute range Bruce. I know that when I was testing the gap trade I varied away from using just 9:30am EST as the start and tried some other times. I found that 8:30am EST was almost as effective as well.
great input!

thanks for the opinion and content. looks like i have some reading to do!

so for an example today how i ran my numbers, and after the initial balance the edge was to break the high and look to fade around 48-51. but no i reran the numbers and here is my criteria: gapped out of range, unfilled gap after the IB, timeframed higher, and in this period, we stopped timeframing higher but did not break the IB, gives me a 7 sample size out of my 10 years of data. The edge is to break lower and go long around 1235-1231 for a rotational trade back in balance.

although the sample size is small it still gives me a good idea of how traders are positioned and how the day moves. so i use these edges to help develop my trading plan for the day but not specifically base it on off it. obviously use price action and volume to tell the story.

maybe i should look at less specific criteria. or here is another question for you guys....how much significance do you put on each individual timeframe? i mean do you think the first 2 timeframes (RTH) is more significant than what the third period does?
Bruce you’re an encyclopedia of knowledge thank you for everything if one day I’m going to be a successful day trader it will be mostly because of you

here is a link that may provide some useful ideas.

http://www.linnsoft.com/homework/


In general I think it would be wise to not only look at the traditional IB in the ES but also use the first 90 minutes of trade. Lots of great fading trades happen off that range.