I would like to test my strategy through a WFA. In that respect I wonder how Ninja deals with the so called start up overhead.
My ATR run is made over a period of (almost) 5 years whereby the in sample number of days ( or optimization period) is set at 365 days, while the out of sample number of days or test period is set at 121 days.
The WFA time frame runs from through 2 Jan 2007 thru 30 Nov 2012. So the first test period starts after an optimization of the first 365 days, i.e. from 2 Jan 2008 and runs to 1 May 2008, the second test period runs from May 2 to 30 August 2008, the third from 31 August to 29 December 2008, etc.
Now, let us assume that one of my indicators is an ATR measured over 14 days. So 14 days of data must be processed before the first trade can be taken into account, right? Or more precisely: any trades occuring in the WFA in the first 14 days should be disregarded. This is what is called the start up overhead ( by Robert Pardo). I know that in Trade Station, when running a WFA, one can enter the instruction to disregard trades made during the start up overhead period. one can simply enter the number of days of data to be disregarded. Is that possible in Ninja as well? Should this be entered in the 'general' box , under 'min. bars required'?
I order to enable me to understand the results of an WFA correctly I would like to know how WFA calculations are made by Ninja. When running above WFA, is the start up overhead period taken into account at each separate optimization run (so 15 times) and test period ( another 15 times) or only at the first 14 days of the first optimization period ( i.e. from 2 Jan 2007 thru 15 Jan 2012)? In other words: is each optimization and test period a continuation of the previous period or is a reset made prior to each calculation. The answer is important because if only the 14 days of the 5 year period ( = 0.8%) should be disregarded or 14 X 15 days (i.e. times number of optimization periods) or even 14 X 30 days ( i.e. times number of optimization periods plus number of test periods) should be disregarded, it would make a big difference for the validity of the WFA calculations. In the later case 420 days out of a 5 year period (=23%) of the days, should be disregarded.
If optimization or test periods are made shorter, while the time frame is kept the same, the effect of a reset would be even more dramatic.
Any one who can shed some light on this?
My ATR run is made over a period of (almost) 5 years whereby the in sample number of days ( or optimization period) is set at 365 days, while the out of sample number of days or test period is set at 121 days.
The WFA time frame runs from through 2 Jan 2007 thru 30 Nov 2012. So the first test period starts after an optimization of the first 365 days, i.e. from 2 Jan 2008 and runs to 1 May 2008, the second test period runs from May 2 to 30 August 2008, the third from 31 August to 29 December 2008, etc.
Now, let us assume that one of my indicators is an ATR measured over 14 days. So 14 days of data must be processed before the first trade can be taken into account, right? Or more precisely: any trades occuring in the WFA in the first 14 days should be disregarded. This is what is called the start up overhead ( by Robert Pardo). I know that in Trade Station, when running a WFA, one can enter the instruction to disregard trades made during the start up overhead period. one can simply enter the number of days of data to be disregarded. Is that possible in Ninja as well? Should this be entered in the 'general' box , under 'min. bars required'?
I order to enable me to understand the results of an WFA correctly I would like to know how WFA calculations are made by Ninja. When running above WFA, is the start up overhead period taken into account at each separate optimization run (so 15 times) and test period ( another 15 times) or only at the first 14 days of the first optimization period ( i.e. from 2 Jan 2007 thru 15 Jan 2012)? In other words: is each optimization and test period a continuation of the previous period or is a reset made prior to each calculation. The answer is important because if only the 14 days of the 5 year period ( = 0.8%) should be disregarded or 14 X 15 days (i.e. times number of optimization periods) or even 14 X 30 days ( i.e. times number of optimization periods plus number of test periods) should be disregarded, it would make a big difference for the validity of the WFA calculations. In the later case 420 days out of a 5 year period (=23%) of the days, should be disregarded.
If optimization or test periods are made shorter, while the time frame is kept the same, the effect of a reset would be even more dramatic.
Any one who can shed some light on this?
Comment