All forward testing on Metatrader is flawed - page 5

 
SIMBA:
Hi Newdigital,El Cid

I don`t agree with you that backtesting is nothing,I would say that it is not everything,for sure ..but that it is a necessary precondition for trading any system/method/EA.

Why?..Because you can forward test your trend following EA like crazy for months,and be extremely profitable,just because you are trading in a very favourable environment..And account loss is waiting there,just at the turn of the corner..the moment the trending environment turns into a range trading one..your Account is dead.

The only way to approximate how your EA is going to behave long term,with trending and ranging environments is to backtest for a very long period,with the following preconditions..

1-The backtested period should contain both trending and ranging subperiods

2-Once you optimize the settings of your EA/method/strategy..you should backtest again,without modifying these settings,on an OUT OF SAMPLE different period,that should contain both trending and ranging markets too,and this out of sample results will give you a fairer idea about your SYSTEM POTENTIAL,both in returns and Drawdown,than what you got at the optimization phase..it is usually a reality shock

3-You should backtest(and trade ) at a fairly liquid broker database,after checking that deviations between demo and live datafeeds are minimal..FXDD,as OceanFx rightly pointed out,is the perfect example..though not the only one

4-The optimal settings found in the IN SAMPLE testing should stand a "common sense" test..ie if your MA crossover optimal is 5/25..the suboptimal settings should be "near and similar"..ie..6/25,4/24,etc..if there are considerable differences in backtesting results from very near settings,the test is flawed..just throw it to the trash bin

Then,and only then ,you can start forward testing or trading your system with a minimum expected edge..and no guarantee at all that you will make it ..but this is how the markets work

Of course,if somebody backtests IN SAMPLE only,with a non representative sample,overoptimizing the settings,at an illiquid broker,with extreme differences between live and demo datafeeds..I agree with you,then backtesting is nothing...

Regards

Simba

If we can look inside the code of EA so we can backtest. Because in this case we will understand why backtesting will be different from forward testing (and may be it will be the same one), why we will get different forward testing results with different brokers and so on.

But if we can not look inside the code so backtesting is nothing in this case because commercial sellers will not tell us how it was coded really.

Besides if EA was created from manual trading system so we know the settings already without backtesting.

And if EA was created just from "i attached indicator to the chart and I see that it is very good can you create EA for me" so the only way to see something in the beginning is backtesting of course. Just to understand it is necessary to forward test or not.

 
newdigital:
If we can look inside the code of EA so we can backtest. Because in this case we will understand why backtesting will be different from forward testing (and may be it will be the same one), why we will get different forward testing results with different brokers and so on.

But if we can not look inside the code so backtesting is nothing in this case because commercial sellers will not tell us how it was coded really.

Besides if EA was created from manual trading system so we know the settings already without backtesting.

And if EA was created just from "i attached indicator to the chart and I see that it is very good can you create EA for me" so the only way to see something in the beginning is backtesting of course. Just to understand it is necessary to forward test or not.

Newdigital,

1-My previous post was based on knowing the settings and the logic of the EA/method/system...Of course,if you can not look into the code nor see the trading logic,not only backtesting is nothing but..Why would you want to forward test it ?And,more important..Why would you plan to trade it?Every trading system has losing streaks,you need to have confidence in the system to stick with it during those periods..and to do so,you need to know its trading logic

2-If the EA was created from a manual trading system you know the settings without backtesting...but you don`t know the limits of the system,nor what can you expect in different trading environments..after all,every manual trading system has been created by trial and error..and backtesting it in an OUT OF SAMPLE environment will expand the trial , reduce the error and improve the system

3-I am a manual and discretionary trader,I don`t use EAs for trading,but I use them for testing my ideas..and thanks to backtesting I can filter the bad ones and improve on the ones that show potential..Imagine that I tell you that trading the breakout of the 12:00 GMT H1 bar ,only in the direction of an H4 SATL,is extremely profitable..probably I reached this conclusion due to the last month of trading events or so,I saw it happening successfully,so I noticed it,and I am prepared to trade on it..How would you know this is not due to some short term fad?How would you know if this system could stand the test of time?Only way to START KNOWING is to backtest it extensively..If the backtest shows the system won in the past under very different trading conditions,it is just saying you that the idea has potential..if the backtest shows this to be a losing system,it just saved you a lot of money..and this is worth something,isn`t it?

So,the right backtest is extremely useful if you use it as a disproving tool,try to disprove your idea..and if you can`t..the idea has potential

Additionally,as a side note,regarding commercial EAs,just backtest them out of sample,for a different period to the one shown in the ads..and you will see how most of them,if not all,fail miserably.

Regards

Simba

 

Simba,

Your cases are usefull and you are right.

But to understand my point so I am talking mostly about commercial EAs. Because to understand why El cid opened this thread you need to read this post https://www.forex-tsd.com/forum/commercial-talks/8803-evaluation-of-oilfxpro-ea-portfolio#comment_228227

He wanted to sell his EA without any forward testing results. The same with dynamit system guy. That is why El cid opened this thread.

No way sorry.

By the way it is very good thread and very good discussion about backtesting and forward testing.

El cid, please ... don't be emotional and don't take anything personally.

 

Because you wanted or not - but you opened good thread with very good topic. It will help many members to understand the differencies between backtesting and forward testing and between forward testing and real trading.

 

I think that ND is right about asking garanties about what is sold on the forum he is managing.

For instance, I think that it is justified to ask for forward tests and most of the things listed.

The only thing that shouldn't be required is the source code as it is really the property of the developper.

Then I would say that basically ND is right and if people don't like it they can try to sell the EA somewhere else.

For developers, if there is an official pool of developers it would be good that ND requests references from clients they worked with. It's simple and could be a first filter.

But I think that the worst uncheck commercial abuse is just to say "I am a professional trader for 50 years, I have a very good system (by the way it was invented 2 hours ago, forgetting my 50 years of trading), join me I will teach you but you have to pay".

So nothing material is sold (then no control) and people are "extracted" from the forum to take their money in a peaceful environment only controled by the guru (exactly like for some sects)

 
newdigital:
Simba,

Your cases are usefull and you are right.

But to understand my point so I am talking mostly about commercial EAs. Because to understand why El cid opened this thread you need to read this post https://www.forex-tsd.com/forum/commercial-talks/8803-evaluation-of-oilfxpro-ea-portfolio#comment_228227

He wanted to sell his EA without any forward testing results. The same with dynamit system guy. That is why El cid opened this thread.

No way sorry.

By the way it is very good thread and very good discussion about backtesting and forward testing.

El cid, please ... don't be emotional and don't take anything personally.

Newdigital,

Thanks for the additional information,I hadn`t known about that issue when I posted my comments..anyway,as you said,the thread is good and informative with different opinions that can contribute to improve everybody`s understanding about back and forward testing

Regards

Simba

 

Nothing wrong with forward testing.

Just go to this page https://www.mql5.com/en/forum/general and download the instruction about how to use VPS from ckowyong. It is free. Then take VPS for $19 per month, or for $10 per months, or for $12 per month. Do forward test for 3 months or few weeks (depends on how often your EAs are trading) and sell as many EAs as you want. And no one will tell you anything.

If I want to sell something I will do the same.

 

Hes talking about 1 bar in one instance.On demo who cares.Its worse on live.Brokers constantly changing them.The history is constantly being changed so that indicators or ea's do different things than expected.This is why no serious trader uses broker data.They pay for there own private data.This is why no EA has ever given any decent long term profits.That is why people like beluck earn 2 grand for programming e-signal feed into MT4.What is the point of backtesting on perfect data when live data is constantly messed with to the point that no one can keep track of it.Its pointless unless your just a scammer trying to sell things and to make an EA look like an atm machine.Which is what happens to 90% of traders who eventually realize they cant beat the data.They end up selling things.If you want to beat the data the only way is to not use it at all and pay for real data that hasn't been messed with.I look at simple MA on E-signal compared to live MT4 broker and there is massive discrepancies.I even ask the broker about this show them pictures they say well you signed a disclaimer knowing that our data although we try our best is not perfect.Thats just there get out of jail free card for messing with your trading.Why you think companies likes FXDD have campaigns with incentives urging people to deposit more money.Because thats how hey make there money mess with your trades so you lose then they just keep deposit.That alot more profitable that pip spreads.

 
SIMBA:
The only way to approximate how your EA is going to behave long term,with trending and ranging environments is to backtest for a very long period,with the following preconditions..

1-The backtested period should contain both trending and ranging subperiods

2-Once you optimize the settings of your EA/method/strategy..you should backtest again,without modifying these settings,on an OUT OF SAMPLE different period,that should contain both trending and ranging markets too,and this out of sample results will give you a fairer idea about your SYSTEM POTENTIAL,both in returns and Drawdown,than what you got at the optimization phase..it is usually a reality shock

3-You should backtest(and trade ) at a fairly liquid broker database,after checking that deviations between demo and live datafeeds are minimal..FXDD,as OceanFx rightly pointed out,is the perfect example..though not the only one

4-The optimal settings found in the IN SAMPLE testing should stand a "common sense" test..ie if your MA crossover optimal is 5/25..the suboptimal settings should be "near and similar"..ie..6/25,4/24,etc..if there are considerable differences in backtesting results from very near settings,the test is flawed..just throw it to the trash bin

Then,and only then ,you can start forward testing or trading your system with a minimum expected edge..and no guarantee at all that you will make it ..but this is how the markets work

Of course,if somebody backtests IN SAMPLE only,with a non representative sample,overoptimizing the settings,at an illiquid broker,with extreme differences between live and demo datafeeds..I agree with you,then backtesting is nothing...

Regards

Simba

Simba, thanks for one of the very few helpful posts in this thread! Let's go study now

 

you can argue and argue, but some skulls are too thick. dont bother trying to convince these knobs.

as far as im concerned, mbtrading feed with ninjatrader or quotetracker may be more accurate.

people, metatrader is FREE for a reason! do you know why ?