|
PureBytes Links
Trading Reference Links
|
Alex,
my system is trading hypthetically about 600 trades in that 25 years that are
24 trades per year (looks reasonable to me for an EOD-system). But my system
calcs the 'whole' program to determin the new stop- and revers-prices on
every new bar (= new day). I think that most systems do that. So only the
admin. of the results differ if you have only very few trades.
But your argument that using ema() or sma() would make a measurable difference
of the speed is not correct for me. I'm using 7 functions to smooth data
either sma() or ema() or others.
Of course I know, that different systems may be faster or slower, but the
difference should not be too big, because I think the main time-consuming
parts are the data-server (reading from hd) and the adminstration of the data
received. Only hereafter the individuality of the system comes in with the
complexity of the system itself and the admin. of the results.
So I still think that it is meaningfull to compare even different systems on
different plattforms.
Do you have a EOD-system tested over 25 [or 10] years (100 [40] contracts)? If
yes can you remeber the time?
Carl
Am Dienstag, 25. November 2003 18:09 schrieb Alex Matulich:
> >may I ask you a question about how long you need for one turn with system
> > that is trading historically in 1 market about 100 End-of-Day-contracts
> > (= 25 years with a market that has 4 contracts per year).
> >
> >I need about 30 sec (Pentium IV; 2.7 MHZ; no Hyperthreading).
>
> It depends on what sorts of calculations happen on each bar, and how
> many trades it makes. A computer configuration like yours, with
> that much data, takes a fraction of a second if it makes just 6
> trades per year, using simple exponential calculations (which need
> only 1 bar lookback and no loops). On the other hand if I introduce
> a complex price filter that involves looping on every bar, or a
> strategy that trades 3 or 4 times per week instead of 6 times per
> year, the trade history calculation slows down considerably.
>
> If you can change simple moving averages to exponential moving
> averages, and standard deviations to exponential standard
> deviations, and otherwise make your code more efficient, you should
> see a speed improvement.
|