Page 1 of 1
Difference when testing Tick Data Vs 1min forexrite
Posted: Mon Jun 03, 2013 7:42 am
I recently purchased premium tick data from my FT2 - and problem is, that when I ran the systems I built with the tick data, the result vs the 1min historic data (forexrite) is completely different. So much so that a winning system becomes loosing. I am concerned because I am live trading 2 systems that I have built over a year ago each.
My systems are not sensitive , i.e stoplosses = 20 pips. so that shouldnt be the problem.
I am confused as to why this would happen - anyone else had this experience.
Posted: Mon Jun 03, 2013 8:58 pm
Ok - SO I wanted to quantify the difference for you:
First of all - I used FXDD tick data (FXDD)/ Forexrite 1 min data (FXR) / Dukascopy historical tester tick data (DUK).
The system is basic logic - on FXDD and FXR the system code is identical. The logic of the DUK code is the same, different code (java).
All three were set to Jan 2011 - Jan 2012 (365 days)
NUMBER OF TRADES
FXDD - 84
FXR - 78
DUK - 141
FXDD - -54.93%
FXR - 1124.93%
DUK - -56.9%
So what I can see is that the forextester data is completely different to the broker backtest. Also the 2 forextester backtests are totally different - same system settings.
The forexrite data was just imported (no timeshift, just download and generate ticks). This is the same for the FXDD data (no timeshift, just download and generate ticks).
I am expecting slight differences between data & platform - but not HUGE differences. This completely invalidates my LIVE systems. Also this creates trust issues with the forextester software - I no longer know if I can use your tool to develop trading systems accurately. It looks as though the FXDD and DUK results are similar - however there are 60 TRADES difference!!!!!!
If this is an issue with how I am downloading the data - let me know (should I timeshift & ignore weekends?)
Please guide me!
Posted: Tue Jun 04, 2013 7:32 am
Now I have imported 1m Dukascopy data, same year - same system - same setting. Different Results!
The variance between data is too big. I am confused as to what I can do to fix this issue.
So.. I guess the question is - what do I have to do to get a reliable result I can trust?
Posted: Tue Jun 04, 2013 9:40 am
Can you send us your strategy and data which you have from another source so we can test it and investigate a problem?
Posted: Tue Jun 04, 2013 5:53 pm
I'll send you my strategy via PM - regarding the Data - for a start use the FXDD VIP tick data (year 2011-2012) vs the standard forexrite 1min data (year 2011 - 2012).
Also is there a way I can import tick data from dukascopy? This is most probably the data I want to use as this is my broker
Posted: Wed Jun 05, 2013 10:25 am
It seems that the difference was caused by the number of digits autopoint in historical data. In you your code you use "point" value which equals 0.0001 for 4-digits data (Forexite) and 0.00001 for 5-digits data (FXDD). Please try to place point value with some constant value like 0.0001
Posted: Wed Jun 05, 2013 7:52 pm
Thank you for your response - I have changed the system so that it isnt using POINT but definined as 0.0001 - I have tested again across FXD and FXR and the results are still polar opposite. This is the same case with MANY of my systems - hence why it is such a big issue.
I will PM you the modified system code - please help explain why there is a huge difference between different data sets.
Posted: Sat Jun 08, 2013 5:53 pm
Hello Support - have you had a chance to look at this, can you replicate the issue?
Posted: Mon Jun 10, 2013 10:40 am
Thank you for your updated version of your strategy. We will test it tomorrow and let you know about results.
Posted: Mon Jun 10, 2013 7:17 pm
Hello Support - I look forwards to hearing how to fix my issues, Thanks again!
Posted: Thu Jun 13, 2013 7:49 pm
Hello Support - Any update?
Posted: Fri Jun 14, 2013 10:04 am
Your strategy is too sensitive to the small price changes and small difference between price and Envelopes indicator. Here is an example (on the screenshot below):
Forex Tester with tick data opened an order because the price nearly crossed Envelopes_bottom_line and crossed in terms of your strategy (current bar Close was smaller then previous Envelopes_bottom_line value). And with Forexite data order wasn't opened because the price didn't cross the Envelopes_bottom_line. And there are many such differences during the long test and this is the reason why you see big difference in testing results.
So there is not a bug of Forex Tester but this is how this particular strategy works. Forex Tester just performs the algorithm and there isn't its fault.
Posted: Sun Jun 16, 2013 6:24 am
Yes - agree with this system it is too sensitive. I've just ran through another system - Forexrite vs Dukascopy 1m data - very large difference again. This system is not so sensative - 15min data, 20 pip stoploss.
Do you want me to send that to you also - or..
I guess the alternative is - to build everything in dukascopy data, and stay away from anything which is towards scalping or fast indicators?
Posted: Mon Jun 17, 2013 4:01 am
Yes, you can send us your strategy, but i don't think that we can help you with such strategies because they can work "a bit randomly" depending on very small factors. And this is not a fault of Forex Tester.
Posted: Mon Jun 17, 2013 4:45 pm
So the only way to check if a system is truely valid - is to run on multiple sets of data and crosscheck the results?
Posted: Thu Jul 04, 2013 2:00 pm
Sorry for the delay, we missed this thread somehow
Yes, this idea seems to be good but it do not guarantee you that you will get similar results in the real trading, strategies of this kind are a bit random