Big Mt4 Bug | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Bug with MT4 Historic Data on Simulator 11 replies

Strange MT4 output - mql bug ? 2 replies

MT4 Bug? 4 replies

A bug in MT4, any work arounds? 3 replies

Unusual Bug with DDE and Excel Scripts in MT4 8 replies

Attachments: Big Mt4 Bug
Exit Attachments

Big Mt4 Bug

  • #1
  • First Post: Edited 2:19pm Feb 10, 2007 8:08am | Edited 2:19pm
I'm dealing with MT from over one year, and now i feel cheated.

I did an EA that generates over 98% of positive transactions with 90% modeling quality. CHeck this out:

http://forex.webup.pl/mt/temp1.jpg

felt lucky millionaire till i check whats happen in M1 chart. Results were sad :|


In my script I used Bands values but only from M5 period so the code is like this:
double HBands = iBands( NULL, PERIOD_M5, 20, 2, 0, PRICE_TYPICAL, MODE_UPPER, 1);


I tried to understand where is the difference between M1 and M5 periods in tester procedure, so i did a small investigation. I modify my script in way to save calculated bands data to a file after every tick.
The results are that:
in M1 test period:

(in C column are ticks, column D and E are calculated bands values)

http://forex.webup.pl/mt/temp1a.jpg


and in M5 test period:

http://forex.webup.pl/mt/temp1b.jpg


As You see ticks are exacly thesame in M1 and M5 tests periods, but there is a difference between bands values for that ticks. There are't even similar.

After that I did a charts from those data that shows me what's going on, and as soon as i saw that i thought my heart get attack.

Let judge it by yourself:

M1 test period:
http://forex.webup.pl/mt/temp2.jpg


and M5 test period:

http://forex.webup.pl/mt/temp3.jpg

Sooooooooooooooooooo.

As you see MT4 has a bug that is a strumbling block for makeing money on forex. Very sad...........


If someone wants i can write point by point how to test scripts in my way.
Attached Images
  • #2
  • Feb 10, 2007 10:36am Feb 10, 2007 10:36am
Does your expert trade on interpolated data, or just off the bar opens?
The breaking of a wave cannot explain the whole sea.
  • #3
  • Edited 12:49pm Feb 10, 2007 10:56am | Edited 12:49pm
Quoting PabloPL
If someone wants i can write point by point how to test scripts in my way.
Dear friend,

I dont really know what are you delivering, but I'm sure the strategy tester has a major issue of transmitting history data to the trading signal's trigger. Can you please kindly share your method of debugging? And what is the criteria to differenciate a good/bad backtest with the buggy backtester from MT4? I recently made an EA....perhaps, I dont start the EA, I took an EA from odbc and have it reconstructed. The performance is extreamely good until NO LOSS TRADE for the entire year 2006. I did like to check how good my system is and whether its suitable for live trading. Because last week was a good week for my EA, and I did make some bucks on live trading with it. But I'm still skeptical of letting it run by itself. Pretty worry about it. Please guide.
Attached File(s)
File Type: zip Backtest V1.21.zip   47 KB | 387 downloads
File Type: mq4 stocEAv1.21.mq4   11 KB | 455 downloads
File Type: mq4 stocEAv1.2.mq4   10 KB | 436 downloads
File Type: zip StochEAv1.2.zip   59 KB | 410 downloads
  • #4
  • Feb 10, 2007 11:35am Feb 10, 2007 11:35am
  •  tesla
  • Joined Oct 2006 | Status: Friendly Neighborhood Programmer | 533 Posts
Okay, a couple of things here:

1. The smooth equity curve you're seeing is due to trades being opened and closed intrabar. Within a minute period, MT has to guess where the ticks were and it does a fabulously bad job of modeling performance if you open and close a trade within the same minute. Did you notice that you've got more than 18000 trades within 8 months? That's more than 100 trades per day. The details of this issue are pretty clearly identified in the backtesting information you can find stickied in this forum.

2. I haven't had the problem you're stating with accessing data from other timeframes, but I would expect you to see that if you were trying to get data using a bar offset without accounting for the time diff. Post your code and I'll take a look.

3. The whole reason for limiting pictures to 600x600 pixels is because when you post huge screenshots like you did, it breaks the layout. If you abuse the huge-honkin'-screenshot feature, they're going to take it away.
  • #5
  • Feb 10, 2007 11:40am Feb 10, 2007 11:40am
Please reduce the size of your screen shots or I will be required to delete your posts... thanks Scott
  • #6
  • Feb 10, 2007 1:52pm Feb 10, 2007 1:52pm
Quoting Craig
Does your expert trade on interpolated data, or just off the bar opens?

Lets looka at the attachments - im using interpolation of every tick, but it isnt importatnt what what method You are using - even in "bar completed" Bands values should be very similiar to reality.

On my charts you can see, that bands values are calculated from feature currency movements so it schould't by like that. Its just impossible.
  • #7
  • Edited 2:52pm Feb 10, 2007 2:10pm | Edited 2:52pm
Quoting davidke20
Can you please kindly share your method of debugging? And what is the criteria to differenciate a good/bad backtest with the buggy backtester from MT4?
Simple method is like that:
1) add in your script init function opening file for data
2) in function start first line schould by saving data to the file
3) run it on couple different tester periods
4) load cvs data from files to a MS Excell and do a charts on the data
5) compare it on Excell chart.

Charts should be similiar and look like the way it schould be.
Thats why at first You schould have good klowledge about tested indicators. This way of debugging cant tell You about minimal differences, but You can see if overall process of calculating indicators are good or not.


----------------------------------------------------------------------
Im including sample code:
......
int f;

int init() {
files = FileOpen("test.csv",FILE_CSV|FILE_WRITE);
return (0);
}

int start() {
FileWrite(f,NormalizeDouble(Ask,4)," ",
iBands( NULL, PERIOD_M5, 20, 2, 0, PRICE_CLOSE, MODE_UPPER, 1));

...... (your code)
}

---------------------------------------
After that You will see if the indicators are calculated well.


Bellow i attach a simple code that store calculated data to future analyze in MS Excell. You can use it.
  • #8
  • Edited 2:54pm Feb 10, 2007 2:33pm | Edited 2:54pm
Quoting tesla
The smooth equity curve you're seeing is due to ......... Did you notice that you've got more than 18000 trades within 8 months? That's more than 100 trades per day. The details......
Thats normal i know that. But the case isn't about smooth of indocators data even of equity.

Quoting tesla
I haven't had the problem you're stating with accessing data from other timeframes
I dont have it too. Im using data from other timeframes, but why for tick in the present the indicator value for that tick are calculated from ticks in the future ?? I understand that it could be a bug - all soft has it, but in this soft is very damaging. People thinks that they will earn some money, and if the check it in real trade the are loosing all. THAT THE WORST BUG I HAVE EVER THINK IT COULD BE !!!

Quoting tesla
3. The whole reason for......
Okey - im newest user so : Im sorry
  • #9
  • Feb 10, 2007 2:49pm Feb 10, 2007 2:49pm
Of course i send to You a simple MQ code.
Attached File(s)
File Type: mq4 SampleCodeMTBug.mq4   1 KB | 433 downloads
  • #10
  • Feb 10, 2007 2:49pm Feb 10, 2007 2:49pm
  •  tesla
  • Joined Oct 2006 | Status: Friendly Neighborhood Programmer | 533 Posts
Quoting PabloPL
I dont have it too, but why for tick in the present the indicators for that tick are calculated from ticks in the future ?? I understand that it could be a bug - all soft has it, but in this soft is very damaging. People thinks that they will earn some money, and if the check it in real trade the are loosing all. THAT THE WORST BUG I HAVE EVER THINK IT COULD BE !!!
Look at it this way.
5 Minute chart, 12 ticks back, that's an hour behind.
1 Minute chart, 12 ticks back, that's 12 minutes behind.

If you're on a 5 minute chart referencing Period_M1 data without taking into account the time offset you can get into trouble. Not always, but it's very possible.

I've done alot of work using multiple timeframes and I've never seen this problem. Is it possible I missed it? Sure is. Is it possible that you've screwed up somehow? Sure is. Post your code (not just a snippit) and I'll take a looksee.

And please, easy on the caps and bold. Getting fired up doesn't help troubleshooting at all.


Edit: Sorry... posted while you were posting. I'll take a look at the code.
  • #11
  • Feb 10, 2007 3:07pm Feb 10, 2007 3:07pm
  •  tesla
  • Joined Oct 2006 | Status: Friendly Neighborhood Programmer | 533 Posts
Nope. No errors here. I modified your code just a touch to output both the M1 and M5 timestamps, I suggest you visually check the data against your chart. I wish you'd done so before putting together your original post. Just choose a couple bars M1 and M5 bars and compare values to your output.

Inserted Code
int f;

int init() {
      f = FileOpen("data.csv",FILE_CSV|FILE_WRITE,",");
      if (f == -1) {
         Print("error(",GetLastError(),"): data.csv");
         return(-1);
      }
      return (0);
}


int start() {    
        FileWrite(f
            ,TimeToStr(iTime( Symbol(), PERIOD_M1, 0))
            ,TimeToStr(iTime( Symbol(), PERIOD_M5, 0))
            ,NormalizeDouble(Ask,4)
            ," "
            ,iBands( NULL, PERIOD_M5, 20, 2, 0, PRICE_CLOSE, MODE_UPPER, 1)
            ,iBands( NULL, PERIOD_M5, 20, 2, 0, PRICE_CLOSE, MODE_LOWER, 1)
            ,iBands( NULL, PERIOD_M1, 20, 2, 0, PRICE_CLOSE, MODE_UPPER, 1)
            ,iBands( NULL, PERIOD_M1, 20, 2, 0, PRICE_CLOSE, MODE_LOWER, 1)
        );
         return(0);
}
And... no, the M1 and M5 values should NOT be nearly the same. Think about what bollinger bands indicate.

I'm heading to the mall with my daughter, if you need further clarification it will have to wait until tonight.
  • #12
  • Feb 10, 2007 3:09pm Feb 10, 2007 3:09pm
Quoting tesla
Look at it this way.
5 Minute chart, 12 ticks back, that's an hour behind.
1 Minute chart, 12 ticks back, that's 12 minutes behind.
Yes but why:
5 Minute chart, M1 Bands data with 1 ticks back, that's one hour ahead ?

Quoting tesla
And please, easy on the caps and bold. Getting fired up doesn't help troubleshooting at all.
Okey - im very nerwous about that. I spent lot of time till i notice that i'm "in a forest" as a Polish people say.

Of course i'm very glad that You will check it. Thanks in advance.
  • #13
  • Feb 10, 2007 3:28pm Feb 10, 2007 3:28pm
Quoting davidke20
Dear friend,

I dont really know what are you delivering, but I'm sure the strategy tester has a major issue of transmitting history data to the trading signal's trigger. Can you please kindly share your method of debugging? And what is the criteria to differenciate a good/bad backtest with the buggy backtester from MT4? I recently made an EA....perhaps, I dont start the EA, I took an EA from odbc and have it reconstructed. The performance is extreamely good until NO LOSS TRADE for the entire year 2006. I did like to check how good my system is and whether its suitable for live trading. Because last week was a good week for my EA, and I did make some bucks on live trading with it. But I'm still skeptical of letting it run by itself. Pretty worry about it. Please guide.
About your results with Stoch1.2 for instance I would be very careful because 30 trades is not statistically significant (setting aside any MT4 bug). In my opinion 100 is a strict minimum. And if you use already only the year 2005 you will see that the results are different.
  • #14
  • Feb 10, 2007 4:29pm Feb 10, 2007 4:29pm
Quoting tesla
Nope. No errors here.

I must agree becouse i didint tell You one thing. I check this one again and i found that if script gets data from MT data history module all indicators were calculated well.

But i'm using my own tick file-real tick file tick by tick. I have stored it in simple cvs file (some tickt for example):
2006-04-01 01:00:00.847000000,1.2117
2006-04-01 01:00:36.503000000,1.2117
2006-04-01 01:46:36.430000000,1.2117
2006-04-01 01:47:36.367000000,1.2117
2006-04-02 23:01:10,1.2112
2006-04-02 23:01:20,1.2112
2006-04-02 23:01:30,1.2112
2006-04-02 23:01:40,1.2112
.... and so on
Im conversin it to .fxt using a script named: simplecsv2fxt witch i have from MQ code base (http://codebase.mql4.com/)
After that i overwrite those fst files in tester/history directory, and i'm testing my scripts once again.
After that i'm getting such bug.

What's wrong ?? If You want i can put my tick files onto my server.
You can check by Your own this process.



Quoting tesla
And... no, the M1 and M5 values should NOT be nearly the same. Think about what bollinger bands indicate.
Yes - thast is clear for me that if im testing my script on (for example) M1 chart, M1 and M5 bands values are are different. Thats normal.

But i'm telling about that i'm testing thesame script in M1 and M5 charts and in those charts M1 bands values are different:
M1 bands values in M5 tester period are diffrent from M1 band values in M1 tester period procedure... and the difference is about one hour ahead.

So i compare only M1 band values but between M1 and M5 tester periods.

My charts from Excell are included above in first post - i didnt paint it by myself.

Please help - what is wrong ?
  • #15
  • Feb 10, 2007 4:35pm Feb 10, 2007 4:35pm
Quoting jlpi
About your results with Stoch1.2 for instance I would be very careful because 30 trades is not statistically significant
Those are not trades, thats bands values tick by tick. (in my pictures in excell data i show only 30 of them (30 of 30000))
  • #16
  • Feb 10, 2007 10:51pm Feb 10, 2007 10:51pm
  •  tesla
  • Joined Oct 2006 | Status: Friendly Neighborhood Programmer | 533 Posts
Quoting PabloPL
I must agree becouse i didint tell You one thing. I check this one again and i found that if script gets data from MT data history module all indicators were calculated well.

But i'm using my own tick file-real tick file tick by tick. I have stored it in simple cvs file...
Im conversin it to .fxt using a script named: simplecsv2fxt witch i have from MQ code base (http://codebase.mql4.com/)
After that i overwrite those fst files in tester/history directory, and i'm testing my scripts once again.
After that i'm getting such bug.

What's wrong ??
I've never used the script, so I don't know how it behaves. Want me to guess what's going wrong? It's a long shot, but see if the script imports to a non-standard chart. Running your EA against that non-standard EURUSD chart will give you values from that non-standard history for 1M data, and will have to reference the standard history for 5M data. Running your EA on 5M charts means you're once again on standard chart data for both 5M and 1M data. So what you ask? Well, if the tick data you imported is off by an hour, you'd get exactly the kind of wonky values that you're seeing.

That's the best guess I've got.
  • #17
  • Feb 11, 2007 9:14am Feb 11, 2007 9:14am
Quoting PabloPL
Those are not trades, thats bands values tick by tick. (in my pictures in excell data i show only 30 of them (30 of 30000))
Pablo,

this comment was for Davidke20 for the results he published, not the values of the bands
  • #18
  • Feb 11, 2007 11:43am Feb 11, 2007 11:43am
I wrote an indicator to plot the moving averages from 1 min - 4 hour time frames on a single chart. It doesn't look that good on a chart but hopefully you can understand the code. I am attaching it and perhaps it will help you will your problem. My understanding of this is not perfect but I will try and offer my advice.

On a one minute chart you should see the 5 min MA having 5 one minute bar values exactly the same because it is only calculated once every 5 minutes. On a one minute chart you will see the 1 hr MA with the same value for 60 bars because it is only calculated once per hour.

Hope that helps.
Attached File(s)
File Type: mq4 MATimeFrames.mq4   5 KB | 366 downloads
  • #19
  • Edited 4:08am Feb 12, 2007 3:09am | Edited 4:08am
Quoting tesla
I've never used the script, so I don't know how it behaves.
Converts normal cvs with ticks to fxt format. I convert it to M1, M5 M15 and replace them in tester/history before testing. In MT4 help file is written that replaceing files or using non-standard history is allowed. Of course "Recalculation" option is disabled.

This is very important in my opinion to have tested script real ticks, not only interpolated or approximated.
Have You ever tested Your own scripts on such that real ticks i MT platform?

Quoting tesla
Running your EA against that non-standard EURUSD chart will give you values from that non-standard history for 1M data, and will have to reference the standard history for 5M data.
Is that normal even if I have converted non-standard chart to all needed time periods ?? For me is a bug. Authors of MT4 that write this help file schould warn user of such like process tha MT4 can mix history values from standard and non-standard history if its true.
If You think about earning money on forex - in my opinion - its extremely important to test sctipts on real ticks. Am I wrong ?

Quoting tesla
Well, if the tick data you imported is off by an hour, you'd get exactly the kind of wonky values that you're seeing.
After i readed some post on this topic I found that some people dosnt undestand my point of view and my problem so I want show You once again the problem but in other perspective.
I tested my script on standard MT4 history files calculated from MT4 history data center. I found that Bands values calculated from those files are OK in M1 and M5 tesrer periods.
See below:

http://forex.webup.pl/mt/STDm1.jpg


http://forex.webup.pl/mt/STDm5.jpg


After that i tested my script on my own tick file converted to fxt file format.
I convert tick file to M1, and M5 periods nad replaced them in tester/history as it's written in help file from MT.
Results:

http://forex.webup.pl/mt/NONSTDm1.jpg

http://forex.webup.pl/mt/NONSTDm5.jpg



As You see there is an error calculating Bands Values.
In my chart in point A You can see, that bands values sended to a script are very wide, but they schouldnt be like that (wide in height).
If You are in point A where You dont know the future so why bands values are so wide ?? The schould be as on other showed charts.

Lets imagine - we are playing on forex, currencies are changing and once we are in point A.
We dont know anything about the future and what we see on Bands ??

http://forex.webup.pl/mt/NONSTDm1BUG.jpg

Its a BUG.

Other hand - if You compare charts calculated from oryginal MT history data You can see that bands data are very similar - and its normal,
but calculated on ticks file there is an offset one hour.

At the end please belive me - I understand well what is resolution of calculated indicators data related to tester period
and interpolation of ticks calculated from diferent tesret periods - You think I'm not.
Im a computer science degree person and I undestand as those problems as backtesting well.
I know that my written english language might be sometimes funny - but that schould't be a barrier
becouse the bottom line for me and i thik You is to get better soft in the future.
I dont want to be annoying, but I know Im right, and if You want to help just study my cahrts cerfully.


PS: Are You involved in developing MT4 platform - are You a programmers team member ?
  • #20
  • Feb 12, 2007 4:03am Feb 12, 2007 4:03am
Quoting dumbmoney
On a one minute chart you should see the 5 min MA having 5 one minute bar values exactly the same because it is only calculated once every 5 minutes. On a one minute chart you will see the 1 hr MA with the same value for 60 bars because it is only calculated once per hour.
Thanks for help but resolution of calculated indicators related to tesrer period is not the problem in this topic. Let look at my charts - especially on the last one.
You can see, that indicator values are strange - why bands data are so wide in point A if the future schould by unknown ?
I will tell you why - becouse MT4 knows the future in that point, and that is the BUG. Is it possible that MT knows the future ? If Yes - everyone on this forum is a millionaire
0 traders viewing now