EA Walk Forward Optimization - To maximize the performance | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Quant - I will run a forward walk for you! Analyze your risk! 0 replies

Is backtesting necessary before Walk Forward Analysis? 7 replies

Walk Forward Analysis - the process of testing 6 replies

Walk forward analysis for EAs 6 replies

Neuroshell Walk Forward Analysis, ATR Questions 0 replies

EA Walk Forward Optimization - To maximize the performance

  • #1
  • First Post: Edited 10:48pm Mar 9, 2014 7:40pm | Edited 10:48pm
Hi Guys!

I have profitable EA myself, but i wonder if I could maximize the EA performance. In fact, market movement has cycle, example Lehman Brother crisis has big average daily range, which is good for A setting, but recent year suitable for B setting.

Meaning to say, i need to optimize the EA once a times, so to apply the optimal setting in future trading. So now the question.
1. How much of history data is good enough for optimizing?
2. How long does the optimal setting valid to be used in future trading?
3. Any system or good method to look for the past optimal value and apply into future trading, which past and future exhibit optimal result?

Here is what i can find, "Automated Optimization....." from this link ,but importantly i must know the above cases first.
Anyone has this kind of work/ideas/method?
  • #2
  • Mar 9, 2014 8:02pm Mar 9, 2014 8:02pm
So in other words, you want the EA to be set based on past performance ?
MT4/MT5 EAs/Indicators/Alerts coder
  • #3
  • Mar 9, 2014 10:50pm Mar 9, 2014 10:50pm
Quoting FerruFx
So in other words, you want the EA to be set based on past performance ?
Yes, but it is not the issue of how to set or how to do by using EA. I want to know the 3 answers for the questions.
Thanks.
  • #4
  • Mar 9, 2014 11:16pm Mar 9, 2014 11:16pm
Quoting ~bull.bear~
{quote} Yes, but it is not the issue of how to set or how to do by using EA. I want to know the 3 answers for the questions. Thanks.
What you are trying to do is useless in my opinion. Optimizing an EA with past datas won't give you any clue of what might happen next ...
MT4/MT5 EAs/Indicators/Alerts coder
  • #5
  • Mar 10, 2014 3:04am Mar 10, 2014 3:04am
Hi bull.bear,

Ia have been coding, testing and optimizing an EA for a very long time now, 1y+ and it's still a work in progress. After going forward and backward trying different methods, this is the approach I finally adopted, and in response to your points:

1. Minimum of 5 years and with at least 2-3k trades generated. I also keep 6 months to 1 year forward test data.
2. You don't know in advance, or in other words, when you know, it's too late.
3. You must have back-test data on which you optimize your variables and forward (unseen) data on which you verify that your optimizations work well.

But what I consider the most important, you should also consider to focus on risk instead of performance.

Maximizing the performance will also maximize your risk and you should allow for some amount of unexpected scenarios you could have in the future that you didn't saw in the past.

Knowing that you have to risk something to get something, the key is to find the correct equilibrium that gives a decent profit for an acceptable risk.
Simplicity is the Ultimate Sophistication.
  • #6
  • Mar 10, 2014 6:16pm Mar 10, 2014 6:16pm
Quoting FerruFx
{quote} What you are trying to do is useless in my opinion. Optimizing an EA with past datas won't give you any clue of what might happen next ...
Logically say it is dangerous, and useless. But from my point of view, the volatility cannot be vanished/changed for recent period of time, so do the price movement range as well. As you can compare the movement from the past to present, range and volatility become smaller now, especially to those year 2008. TP and SL might be got triggered easily for that times but for today won't, not much. So what i believe it should be somewhere useful, at least those setting doesn't run/fall outside the optimal range even if the setting is not so accurate.
  • #7
  • Mar 10, 2014 6:23pm Mar 10, 2014 6:23pm
Quoting broketrader
Hi bull.bear, Ia have been coding, testing and optimizing an EA for a very long time now, 1y+ and it's still a work in progress. After going forward and backward trying different methods, this is the approach I finally adopted, and in response to your points: 1. Minimum of 5 years and with at least 2-3k trades generated. I also keep 6 months to 1 year forward test data. 2. You don't know in advance, or in other words, when you know, it's too late. 3. You must have back-test data on which you optimize your variables and forward (unseen) data on which...
Thank you.
I have searching online, finally found one forum they mentioned about take 25% of optimizing period to run the optimal value. It shouldn't be more on that percentage, otherwise it is not so accurate based on their studied.

So you must have a lot of statistic data to prove what is wrong and right. Yeah, i know risk is increased when looking for better profit, all must be equilibrium. It is best if those setting fall under the optimal range. However, we don't know. But we just assume it works within somewhere the range.

Since you have done more, so what is your opinion ? Do this work? Have you compare with and without walk forward optimizing result ?
Thanks.
  • #8
  • Mar 10, 2014 9:26pm Mar 10, 2014 9:26pm
u r right and wrong

of coz it is useful, or what do you think wall st guys do? anyone else has 'future' data for testing/optimization?

in this sense, you are wrong.

however, i also agree with you that he's doing useless things very likely.

the correct way, imho, is to categorize several typically different market behavior (volatility pattern), for each type price moving, a trader design a set of parameters(or even strategy) accordingly. the successful systems (automatic high frequency, i visited two funds on Street) can both self-adjust according to predefined condition change, and even so, a couple of traders need to supervise it all the time. and it is so resourcefully demanding that i doubt OP can afford to apply. and hence, the outcome is very likely to fail. the way for OP to make it requires strong coding skill, statistical knowledge, as well as cleaned data. in fact, any one of three things may easily prevent 95% FF users from making it this way.

Quoting FerruFx
{quote} What you are trying to do is useless in my opinion. Optimizing an EA with past datas won't give you any clue of what might happen next ...
  • #9
  • Mar 11, 2014 2:53am Mar 11, 2014 2:53am
To your last question, I can't say yet if it works or not, I explain...

For now, as my EA trades on my live account as you can see on my explorer, I can say that it works on the live account like in the tester, so, so far so good.

If it goes through 5+ years of live trading and that I can withdraw my initial investment and keep the account running with "free" money I then will be able to say that it worked.
Simplicity is the Ultimate Sophistication.
  • #10
  • Mar 11, 2014 5:53am Mar 11, 2014 5:53am
Quoting broketrader
If it goes through 5+ years of live trading and that I can withdraw my initial investment and keep the account running with "free" money I then will be able to say that it worked.
That's exactly how we should all look at this. But most newbies look at their backtest in order to say that it will work on Live trading.

BTW, I am not talking about you Bull.Bear ...
MT4/MT5 EAs/Indicators/Alerts coder
  • #11
  • Mar 11, 2014 2:57pm Mar 11, 2014 2:57pm
Quoting FerruFx
{quote} That's exactly how we should all look at this. But most newbies look at their backtest in order to say that it will work on Live trading. BTW, I am not talking about you Bull.Bear ...
Yeah. Honest speaking, i don't believe backtesting, because too many factors can cause the system not likely to be the same result or fall out of the expected equity curve. Slightly different result can cause massive different result in the end. However, backtest is always the last resource that we have, better than nothing.

The most accuracy bastesting result will be 99% tick model, with apply LIVE broker account information (Fixed spread, leverage, tickvalue, commission, swap, and etc). This is what i am doing now.
  • #12
  • Mar 11, 2014 3:15pm Mar 11, 2014 3:15pm
Guys! Thanks for your comments.

However, i am curious about how your guys personal EA or strategy, do you adjust the SL or TP once a times, meaning to say if market movement was changed (Big volatile and big range) ? What are you going to do then ? Or do you based on price action ?

IMO, based on price action can be the right choice, maybe. Actually i have trade management function in price action based, which i coded myself. But if i use this in my EA, profit doesn't look good as current swing high and swing low range is quite small especially for EURUSD. Maybe i should diversify to other pairs, or adjust my MM lot size in order to raise profit but to keep in equilibrium result (Risk and profit).
  • #13
  • Mar 11, 2014 5:18pm Mar 11, 2014 5:18pm
You should believe in back-tests first, because in my opinion, it's the first survival test any EA should go through. But of course you have alternatives like testing on demo account with real live data and having the patience to wait several months of test at least. Or going directly live at your own expenses.

I do not adjust any SL or TP as my system is entirely automatic and mechanic based, and everything was previously adjusted according to back-tests. I just watch like a movie when action is to be taken to verify that it does its job as expected, that helped me to make some improvements and remove some bugs.

I think that unlike me, you use a semi-automatic EA that needs to be adjusted in front of known coming volatility like during the release of NFP for example. If that is the case be aware that you could be in trouble in case of unexpected volatility.
Simplicity is the Ultimate Sophistication.
  • #14
  • Mar 12, 2014 7:16am Mar 12, 2014 7:16am
Quoting broketrader
You should believe in back-tests first, because in my opinion, it's the first survival test any EA should go through. But of course you have alternatives like testing on demo account with real live data and having the patience to wait several months of test at least. Or going directly live at your own expenses. I do not adjust any SL or TP as my system is entirely automatic and mechanic based, and everything was previously adjusted according to back-tests. I just watch like a movie when action is to be taken to verify that it does its job as expected,...
Well. My EA is full automated, it is about price breakout. I don't think my EA works under optimal result forever, because i have backtested (Optimized) in separate year 2012, and 2013. If i apply optimal parameters from year 2012 in 2013, the result is ok but not so good. The same conclusion if i use the optimal parameter from year 2013 in 2012.

In fact, future investment cannot be foreseen, but to assume that it works. Maybe my requirement for EA performance is too high and perfect ? What i mean, it can be improved if there is something that exists. That the reason why i open this thread.

I think if EA system is based on trend, and trade management (SL and TP) is also based on price action, so volatility and movement range cannot be an issue. Therefore, the SL and TP is considered to be logic and appropriate range, so it does work forever. Maybe my system is breakout, which is sensitive to volatility, can say doesn't work all the time, this is the fact.

Thanks.
  • #15
  • Last Post: Apr 18, 2021 4:00pm Apr 18, 2021 4:00pm
Has anybody been able to use the Walk Forward Optimization library found here. https://www.mql5.com/en/blogs/post/677264 If so could you show me some example code as I am a beginning programmer.
0 traders viewing now