Testing irregularities - Can we actually trust this software?

New comment
 
Hi all,

Happy new year, may it be profitable to all !

Thanks for looking at my post, I need some help to explain this anomally please.
I have downloaded 1 minute data from alpari to run my backtests on, the data reads as 90% acurate.
My EA calculates the pivot, S1, S2, R1 and R2 based on the previous days data and I use these to enter trades and work out my stops and targets.
Here is the anomally - I have run the backtest on the same period from 1st Jan 2005 to 1st Jan 2006 but on different time frames, namely the daily, 4H, 1H, 30min, 15 min and 5 min - and each returns a different result !?!

How is this possible, the calculations are tied to the daily ( PERIOD_D1 as opossed Period() ) and as far as I know
the tester goes through the 1M data to build all the higher time frame data.
Please, please help.

Thanks;
Ruf
 
Dear Ruf,

If you enclose the code, we can have a look at it...in the meanwhile, you can do the following:

1. Calculate by hand, day high, day low and day close from individual charts (1m, 15m, 60m, 240m and daily)...see if these values are same or not
2. If these values are same, ask the code to print the high/low/close values per timeframe when you run it (you can use print command and then check journal and log files)
3. If there is a discrepancy, check your clock/time settings to ensure that every timeframe uses same day start and day end
4. If all above fails, try to quantify how much the difference is...if it is 5-10 PIPs, I would ignore it as noise (keep in mind that the way MT calculates higher timeframes data may not be ideal). If it is more than 10 PIPs on average, then this might be a data discrepancy
5. In such case, try forward testing your system for a week and then backtest it to see how much variance you see in system performance and values. If this is within +/-10%, you can believe the system, if it is not, then there is abug in data and/or code.

Bottomline: Though backtesting in MT is accurate enough for backtesting from statistical point of view, relying on the results exactly might not be right. I always treat backtesting results with +/- 10% noise level.
 
there are quotes in different timeframes differences
after download 1min quotes erase all quotes in higher timeframes and use on 1min chart period_converter script to recalculate quotes for all higher timeframes
for precise testing I do not recommend Alpari data :)
it is demo data... crap
New comment