ea optimization

Hoping for some opinions on this.
I have been using a commercial ea for 6 months and I'm happy with the results so far---but I'm wondering if they can be better.
The ea tests well over a 10 year period and presumably the developers used optimization to come up with the best settings for the whole of that period.
I know the historical results dont guarantee future success blah blah but I'm wondering if an optimization of the main indy settings say every month would make the ea more responsive to the current market conditions---or would I just be curve fitting and chasing my tail?
When I carry out optimization with the strat tester I can massively improve profits--sometimes with an increase in DD-- sometimes with a huge difference in number of trades. Is it simply a matter of finding the most inrease in profit with least increase in DD---or do the other figures hint at true performance.
Very interested in hearing from people who have played around with this stuff.
Cheers

Yes, I too have been considering that question. If I optimize an EA, theoretically that should mean if I could go back 1 year (or the time period I tested) and using those parameters, achieve a similar result. If we are basing our future trades with that EA on past results (optimized), then the big question is: How often to re-optimize and what time period? I'm thinking somewhere between 3 months and 12 months of data, and re-opt once a month. Right now I'm planning to use 12 months based on one opinion that I shouldn't use too short a time period, but I too would appreciate some knowledgeable insight from someone who has experience with this issue.

Just my opinions here to help you guys in your thinking. Here are a few options for you to try and consider.
1. Isolate ranging/trending markets and optimize in those areas only. Then manually determine wich market condition you are in and use that set of optimization parameters.
2. Simply optimized on a quarterly basis. So optimize for the last 3months, and use those parameters for the next 3 months.
It's alwasy nice to have other inputs and ideas as soundboards in your own head so I hope you the best in your trading.

I appreciate that opinions will be offered but I always ask myself, "Does this opinion help me?" Some people will ask questions hoping for suggestions or inspiration to trigger their thinking in the hope of reaching the answer themselves. For me, by the time I ask a question, I have considered every possible aspect that I can find and am hoping to get an answer from someone who has already considered or solved this issue. Samaronne, you offered 2 suggestions. Have you ever tried to optimize an EA for profitable use? Are your suggestions based on personal success? You suggest isolating ranging/trending markets: how do you propose to do that? Why are you suggesting a 3 month period? Have you had success with that in your own EAs?

I am not trying to offend, but after all my experiences, I do not accept any answers or advice without understanding "WHY" it is being suggested.

thank you for your input.

IamFX, I hope you don't mind me jumping in here but this question is important to me right now, too. I am trying to optimize an EA as well.

Yeah no worries Eric---the more the better.
I was discussing this topic on another forum and got this response from a bloke that has had experience. I got his permission to post it here.

Heya John! Doing it every month would definitely be curve fitting. Here is what i recommend....
Choose a time period of no less than 3 months up to a year, starting from the current date going back. Find the most profitable setting that has the lowest drawdown (personal preference comes in here....i don't like anything more than 20%). Then run a backtest with those settings. Look at the equity graph. What you should look for is as smooth of an upward equity curve as possible. Once you have a fairly smooth upward curve, then find an area within the time period (maybe even a couple different areas) that the EA does poorly. Run your test with your new settings in visual mode. Go through the trades to see if you can discern why the EA failed and if it looks to be potentially fixable. Always keep in mind, sometimes the trade setups cannot be fixed to be profitable during certain time periods. If you determine that the trades can be potentially profitable, then run a optimization for this time period and see what kind of results you end up with. Most often the difference is very minimal and will have a negative effect on your positive trades over the longer time period. Major point of importance: The number of trades is somewhat important! If you have too few trades, then you are just missing a lot of good trades to keep from having a few losses. If you have too many trades, then you are leaning toward the greedy side. Remember...Past performance is not indicative of future results. What my optimization method is intended for is not maximum profit, but permanent long term profitability. i want to make money without every having any loses. I have found this method of optimizing gets me closer to that than any other way.

Have you considered the effect of a black swan?

That's the main reason optimized systems fail. They do not account for paradigm-shifting events.

I just Googled that to make sure my understanding was correct---here's what I got.
What Does Black Swan Mean?
An event or occurrence that deviates beyond what is normally expected of a situation and that would be extremely difficult to predict. This term was popularized by Nassim Nicholas Taleb, a finance professor and former Wall Street trader.

Investopedia explains Black Swan
Black swan events are typically random and unexpected. For example, the previously successful hedge fund Long Term Capital Management (LTCM) was driven into the ground as a result of the ripple effect caused by the Russian government's debt default. The Russian government's default represents a black swan event because none of LTCM's computer models could have predicted this event and its subsequent effects.

The ea that I use has around 50% winners. So I get used to a fair succession of losers. A black swan event may just create a longer run of losers. It is also a close of bar type ea so conceivable a --for eg--- 300pip drop on along position wouldnt be much fun.
It all comes back to the old equation--high return=high risk and there's no way around it that I can see.
You could use a capital preservation method whereby ,for eg ,you started with a 10k deposit. Say the ea performed as it has done for a long period of backtesting and you doubled your money. You then removed your initial 10k and traded with "free" money. Any blackswan event that happened after this could perhaps be regarded in a different light.
Ronald--can I ask how you deal with the possibility of a blackswan?

I intervene and manually ride it.

That is something to consider, but that would point out any drastic change in current conditions, which is why I'm not sure if I should use the most recent 3 months' results or 12 months (which is what I'm testing with now) or more. You might say I'm already guilty of curve fitting by using different parameters for different currency pairs, but many forex traders have commented on the individual "personalities" of different currency pairs, so perhaps this is correct.

At least, congrats for finding a commercial EA that makes money. I got tired of spending money without finding one so decided to write my own.

And this is the main reason I would want to re-optimize frequently to see if conditions have changed. Kind of like using a weighted MA that gives more current conditions a bigger factor. Just a thought.

I Googled the question and found this link which explains how and why to optimize an EA plus an inexpensive program to automate what they call "walk forward" testing, that is optimize a period, then test those settings on the next shorter period of data just like you would live, and so on.

http://www.easyexpertforex.com/backt...imization.html with links on the page to their explanation of "walk forward" and their software.

Hello guys

During optimization I run through all the test results looking for the smoothest curve, not net profit, as its the smoothness of the curve that will let you keep the EA running. Increasing the lot size will take care of intended profits - BUT it takes for ever to manually search one by one down the list of the optimization results tab looking for an ideal balance curve - any ideas?

Thanks