Originally posted by Zestfulmocha
Alright, time to share some of my initial EA back testing results... still very early days for me, I have developed 4 draft EAs now, only two of them have shown any sort of 'hope', one was really just learning to build so might toss it, and the fourth has some sort of errors I haven't fixed yet, so no idea how that'll turn out!
I have put most of my focus into just one EA so far - being an EA discussed in one of the TradeView tutorials, but I have taken it in my own direction. So far I have really only ran it through the EA tester and analyser, as I've been having trouble with MT4 back testing (I user pepper stone, for some reason having data issues?) and on top of that my work laptop which I was 100% operating form has has some admin changes across the company so I have lost some admin rights, so, I now have a setup at home and am currently downloading what I hope to be a comprehensive and high quality data set (better than broker or MetaQuotes data) in order to build an isolated back testing environment on that computer - completely cut off from any trading accounts (ad therefore brokers) and completely wiped of metaquotes data. My goal here is to have that platform running backrests, but when they are ready for demo/live I just run them on a VPS or separate broker / MT4 environment.
So until that back testing platform is ready, I'm just sticking to TradeView back testing - my strategy;
1. Run the EA through all instruments all time frames
2. Pick out any trends (which instruments/time frames does it do well on? Which does it fail? why?
3. Shortlist the time frames and/or instruments I want to focus on for this EA and take it to analyser to see where I am make improvements
4. Develop thesis on where improvements can be made and run those through EA tester again, at the same time run those shortlisted results through MT4 optimisation and see what results I can when stepping out my indicators (then use the combination of this feedback to rinse and repeat until I hone in on a strategy i'm comfortable to demo.
The approach here is instead of starting with a specific idea I want to remove my assumptions form the process, cast a wide net and just see what happens, working my way down to a solid strategy. This perhaps leaves me vulnerable to curve fitting and cherry picking results, but this is something I will need to learn to balance as we all do
So, for my 'main' EA, I have run it through steps 1-3 and have identified it works best on M30 and H1 time frames, but I have seen best results on AUDUSD M30, as follows (only run on 3 years data, currently re-running with some slight variations over 7 years).
Whilst the draw down is higher than what I would like, I figure position size can always be halved so no real issue for me. It suffers long periods of flat-lining, with most of the of the profits only falling out of 30% of the trading time. I haven't yet fully understood why this is the case (but it is somewhat consistent across all other results too), so i'm currently working on trying to isolate that and see if I can optimise around that to make it more consistent, or perhaps make it 'turn off' during those periods. If I am successful at doing this, my goal is to run this strategy across all of the instruments/time frames that I am confident with, and ensure each is running (and turning off) at suitable intervals - so in a portfolio I may be running 4+ versions of this, with anywhere form 0 - 4 being 'online' at any given time,
A work in progress, and still a lot to work on before I can say with any confident that these results will be replicated on the longer time frames and indeed on a demo account, but so far my small 'dent' into the world of back testing has so far shown some early promise.
Regarding the EA Analyser, some interesting results I saw were that P&L by weekday showed Mon-Thurs being good or 'okay', yet positions opened on Friday were hemorrhaging money. Again i'm not sure if this was due to too many positions being held over the weekend and that just wasn't working with the strategy, or if it was somehow FA related (I don't know enough about Fx and FA to guess), or if it due to simple change in trade psychology/price action going into close on Friday - more analyses needed to be sure, but interesting nonetheless. I turned Fridays 'off' in the above results, and prior to that annual P&L was around ~90% and DD was ~28%, so quite significant changes for simply changing the rule to 'not open positions on Friday'. I have not tested to see if the phenomena also occurs on other instruments or time frames, so cannot say whether this is a real trend that is useful to pull through, of it ifs just curve fitting.
Anyway - baby steps for me, still a long way to go with this back testing stuff but so far I am pleased that my initial efforts have not failed dismally, and that's good enough for me! If anyone read all of this, I welcome any advice or constructive criticisms.
One suggestion (in case you haven't done it) re the issue with testing on MT4 - make sure you've downloaded the TVI Library file to the terminal library folder.
