EA Tester is useful?

To add comments, please log in or register
sharklife
15
sharklife  
I tested several EA with the tester and I got almost incredible results with 90% quality models.
Example. EA tested results in 1 year it turned 5000$ in around 90000$
EA tested on the real market it turned 5000$ in 3500$ in less than 2 weeks.
The comic thing is that the EA tester for that 2 weeks gives out positive results wit this EA and not a -30% performance and for every period tested it never gives out a such big drawdown(maxdrawdon for 1 year was 208$). Even the realtime demo testing wasnt bad.
So is it useful to test EAs with a tester so imprecise? I'm working with the live account now, but as I thought it's not safe at present to put any serious money into a MT4 driven account.
pierfrancesco
6
pierfrancesco  
Hi I'm italian, sorry for my bad english.
However, only 2 days i'm testing the tester and changing parameters i loss or i win.
Only one solution:
A historical quotes tick by tick
and a perfect tester (programmable in mql4 for exemple)
You're a programmer?
Iggy
474
Iggy  
For most possible accurate results in backtester:

1. use alpari data only (MT history gives weird results with sensitive EAs)
2. synchronize all timeframes (I wrote a little script to do that) and put them into history folder
3. hardcode your timeframe in tester (i.e. use iTime, iHigh, etc. use explicit PERIOD_?? instead of 0 or Period() function)
and run your EA on 1 minute frame only)
4. Run your EA on demo for a week or two then compare results with tester for the same period.

Then maybe you will get closer to real life. That is my bitter experience with backtesting.
Files:
pierfrancesco
6
pierfrancesco  
I'm not a programmer.
It possible to make a mql4 program that record all the ticks for a good historical data?
It possible to make a perfect tester using mql4 o others?

In other words, it's possible to make a tester better than MT strategy tester?

thanks
Iggy
474
Iggy  
It's no prolbem. But you need to keep it working for 2 years every trading day for every currency you want to keep tabs on to collect enough data for good testing.

Here is a simple EA that collects ticks. There is also something called FXT collector on this site that collects them and puts them right into a test file.
Files:
Wackena
589
Wackena  
irusoh1 wrote:
For most possible accurate results in backtester:

1. use alpari data only (MT history gives weird results with sensitive EAs)
2. synchronize all timeframes (I wrote a little script to do that) and put them into history folder
3. hardcode your timeframe in tester (i.e. use iTime, iHigh, etc. use explicit PERIOD_?? instead of 0 or Period() function)
and run your EA on 1 minute frame only)
4. Run your EA on demo for a week or two then compare results with tester for the same period.

Then maybe you will get closer to real life. That is my bitter experience with backtesting.
Is PeriodGen a script that is attached to M1 Chart (off line) to synchronize all timeframes for that currency pair? Does PeriodGen auto put timeframe .hst files in History folder? Does it replace the time consuming Period_Converter script?

To compare backtest to demo, do we use M1 for both?

Thank you very much for this idea.

Wackena
Iggy
474
Iggy  

To make things simple here is the sequence:

1. you create subfolder .../experts/files/databank/

2. copy from alpari 1m data into databank (It is called M1_XXXXXX.hst

3. attach script periodgen to the same chart XXXXXX, frame doesn't matter

4. It will create subfolder .../experts/files/databank/XXXXXX in which you will find all timeframes up to H4 generated based on alpari data. (it will not create M1 you already have it, just copied it from alpari, remember)
It runs very fast (maybe less than a minute).

5. save old time frames in your history subfolder and replace them with new ones including 1M (you need to rename it XXXXXX1.hst) which you copied from alpari.

NOTE: this script does not change time zone it uses alpari time zone. I basically created this script for backtesting on a defined data set so I don't care if ti's different.

As long as alpari will continue publisihing data you can repeat this once a month to get more data.
At least for me alpari data completely synchronized gives much more sober results in my testing.
The most sober tests result from following steps in prior post.

alapri has 2 1/2 years worth of data.
I'd say if your EA didn't make any money in 2 1/2 years on alapri time it probably won't make it anywhwhere.

Wackena
589
Wackena  
irusoh1 wrote:

To make things simple here is the sequence:

1. you create subfolder .../experts/files/databank/

2. copy from alpari 1m data into databank (It is called M1_XXXXXX.hst

3. attach script periodgen to the same chart XXXXXX, frame doesn't matter

4. It will create subfolder .../experts/files/databank/XXXXXX in which you will find all timeframes up to H4 generated based on alpari data. (it will not create M1 you already have it, just copied it from alpari, remember)
It runs very fast (maybe less than a minute).

5. save old time frames in your history subfolder and replace them with new ones including 1M (you need to rename it XXXXXX1.hst) which you copied from alpari.

NOTE: this script does not change time zone it uses alpari time zone. I basically created this script for backtesting on a defined data set so I don't care if ti's different.

As long as alpari will continue publisihing data you can repeat this once a month to get more data.
At least for me alpari data completely synchronized gives much more sober results in my testing.
The most sober tests result from following steps in prior post.

alapri has 2 1/2 years worth of data.
I'd say if your EA didn't make any money in 2 1/2 years on alapri time it probably won't make it anywhwhere.

Thank you very much. It worked like a charm. Also, it was very fast as you said.

Wackena
zap
182
zap  
I've been able to make around 95% same results using a sensitive, short time EA, when compared my demo account and my special 99,7% accurate backtester.

I started a demo account one and a half months ago, only to test my backtester.
For this period,
-I had exactly 80 trades both on demo and backtest.
-Largest drawdown was exactly the same both on demo and backtest.
-Actual trades were near regarding opening/closing times (with 0-2 minutes of difference, well backtester doesn't care about time needed for real trades/server outages)
-Actual Profits/Losses (takeprofit,stoploss) were almost the same: a little bit better on the demo than on the backtest (It could have been the other way too, though). This was due to different tick data for the different brokers (I got the historical tick data from a different broker, than I ran the demo with), and requotes/slower response of the realtime demo account, which isn't simulated on a backtest.

Although not getting the very-cent-same results, this accuracy was more than enough to see the actual behaviour of the EA. That is, I could see the very same patterns on the two balance graphs.

So I can say: yes, an accurate backtester is possible! I can be reached at[Deleted] if anyone would like to hear more about it.
Fajar Kurniawan
383
Fajar Kurniawan  

Where I can download strategy tester and how to test my EA in this strategy tester.

If there is a link to study , please inform


thanks

Marco vd Heijden
Moderator
12735
Marco vd Heijden  

This must be a new record 2006 <> 2018.

It's funny because the info is still very relevant:

sharklife:
I tested several EA with the tester and I got almost incredible results with 90% quality models.
Example. EA tested results in 1 year it turned 5000$ in around 90000$
EA tested on the real market it turned 5000$ in 3500$ in less than 2 weeks.
The comic thing is that the EA tester for that 2 weeks gives out positive results wit this EA and not a -30% performance and for every period tested it never gives out a such big drawdown(maxdrawdon for 1 year was 208$). Even the realtime demo testing wasnt bad.
So is it useful to test EAs with a tester so imprecise? I'm working with the live account now, but as I thought it's not safe at present to put any serious money into a MT4 driven account.

Even this day we are discussing these things...

@Fajar Kurniawan the tester is build in it the two desktop terminals.

Download one of those and it will be available for you.

12
To add comments, please log in or register