Hey guys,
I have been working on a system which uses 4 parameters along with a TP and a SL. This is a H1 time frame with the expert always being in the market (back in on the start of a new hour after closing a position). I am currently working on my third version of this EA with each version offering improvements over the previous.
I noticed when optimizing that i could always achieve good back test results, i would then take these results and try them in the future (typical walk forward). I used begging of 2010 to beginning of 2013 to optimize and the time until now to walk forward test.
After running live i noticed a interesting thing which i probably should have known anyway. Basically if you take your optimization and run it immediately after you may or may not have good results. However if you run it a day late, eg start it on the 2nd of Jan 2013 instead of the 1st, you may have completely different results. This is because the optimization is being subjected to slightly different data on the walkforward, data that is a day shifted.
So i noticed that some of these optimizations work well regardless of how much i am stepping this optimization start date. I think of this as this system having more degrees of freedom (i know this is likely not the correct term but it makes sense to me). Some of these optimizations work almost on every step i try to take to offset the start date on the walkforward, others yield completely different results.
So what i am looking to do is take the list i have of 10 - 20 passes which i deemed as good on the walkforward and somehow quantify how well or how robust they are to having the start dates offset and still working. Then at the end i will have a large number of passes which i can sort through and pick the best one which produces the most consistent results on the walkforward date step analysis. From here i will select these parameters which pass the walkforward start date step process for a number of pairs and run them on the demo.
My logic is that if they worked on the optimization, and continued to work well in the multiple walkforward start date offsets, then they will likely work in my account since they are made of parameters which are likely a good fit for the current market and not just a fancy computer optimization.
Does my logic make sense? This is the only way i can think of how to overcome this problem of start dates and some of the optimizations being less robust and more affected by when the trade was taken. I figure if i pick ones that are more robust they will have a higher chance of producing profit during the live test.
Ways I have thought of improving these “degrees of freedom” is by increasing the number of trades and lowering the TP and SL.
Any help is much appreciated.
Thanks