99% modelling quality but with a red bar - page 3

 
gangsta1:
The only thing to worry about is the modelling quality bar is red, but this is perhaps normal with 99% modelling quality. ANyone else experienced this?
Nope, never . . . have tested with GBPUSD, AUSUSD and EURUSD so far . .
 

so what is more reliable, 99% modelling quality with a red bar or 90% with a green bar?

99% modelling quality with red bar dukascopy data: 28189 bars in test and 66403328 ticks modelled

90% modelling quality with green bar broker data: 26695 bars in test and 34321186 ticks modelled

From the above it seems as though the Dukascopy data is much more accurate with almost double the ticks modelled...

 
zzuegg:

drop the 99%, it's only a number in the fxt header.

Duka dasta should be real tick data from their stream, but the main advantage for me is to be able to test on variable spread


What variable (real) spread is used? Dukascopy spread?
 

You can try moving the spreads around and see how the system performs by using this tool. I cannot help but think you're looking for some form of absolute confirmation from the back-tester. With your scalping system, you have to be very careful because those are very sensitive to spreads and data quality. I'm afraid you're not going to get any form of absolute confirmation on how the system will perform in the future based on back-testing results.

The purpose of a back-test is best used to determine if system-A is better than system-B in the past of course. As most of us wouldn't choose a system which performed poorly in the past to our real-account, you get the idea. A modeling quality of 100% wouldn't help you much because A) the data is not your brokers data. B) the spreads are static and C) lord alone know how the broker will respond to your scalper.

If you're showing Excellent results with it so far. Then I say its time to place it on a demo or penny account for about 3-months and go from there.

 
ubzen:

You can try moving the spreads around and see how the system performs by using this tool. I cannot help but think you're looking for some form of absolute confirmation from the back-tester. With your scalping system, you have to be very careful because those are very sensitive to spreads and data quality. I'm afraid you're not going to get any form of absolute confirmation on how the system will perform in the future based on back-testing results.

The purpose of a back-test is best used to determine if system-A is better than system-B in the past of course. As most of us wouldn't choose a system which performed poorly in the past to our real-account, you get the idea. A modeling quality of 100% wouldn't help you much because A) the data is not your brokers data. B) the spreads are static and C) lord alone know how the broker will respond to your scalper.

If you're showing Excellent results with it so far. Then I say its time to place it on a demo or penny account for about 3-months and go from there.



Thank you.

Even more frustrating now is that even with the Dukascopy data I am getting good results on the h1 but not on the m1, when the entry criteria is the same across all timeframes!!

Also confusion over the GMT offset. Dukascopy data is GMT+0. So I assume that I can backtest with the GMT offset set to 0, despite my broker currently being GMT+2 and following DST? I am guessing the tester data is completely independant from the broker, thus 17pm on my broker would still be 17pm GMT+0.

 
gangsta1:


From the above it seems as though the Dukascopy data is much more accurate with almost double the ticks modelled...

This is incorrect . . . with the Dukasdata data the number of tick modelled is 0. This is the point of using the Dukascopy data . . it is tick data so there is no modelling going on . . . Normally MT4 uses the M1 data (and the volume) to synthesise tick data, so it models the tick data, i.e. it makes it up.
 
But now there is another problem----My EA backtests with no losses using the Dukascopy data, 99%. However, with the normal history data and 90% modelling quality it suffers losses!! What results to trust?!
 
If you are talking 1% losing trades vs 2% losing trades that may not be significant, if you are talking 1% vs 10% then that probably will be significant. What time period are you using 2 years or greater ?
 
From 2007 until today. The history data from the broker shows 4 losses, compared to 0 losses with the Dukascopy data (out of 200 odd trades). I really need to know which is more reliable. I have read so much about tick data and Dukascopy inparticular that I am temptd to trust those results more. However, the red bar issue worries me slightly... Very confused as to which to trust. I have says and days of backtestign to do and dont want to waste my time!
 

Re: the red bar, are you 100% certain that you are copying over all the hst files, from M1 to MN1 ?

For example,

GBPUSD1.hst
GBPUSD5.hst
GBPUSD15.hst
GBPUSD30.hst
GBPUSD60.hst
GBPUSD240.hst
GBPUSD1440.hst
GBPUSD10080.hst
GBPUSD43200.hst