PureBytes Links
Trading Reference Links
|
Gary is quite right... I concur in all points. (And I learned an elegant test for the beginning of a new trading week!)
(Incidentally, I did originally mean for the inputs to be manually updated weekly. The thought of automatic updating based on an initial set of two inputs was a later thought intended to allow for backtesting. However, as Gary points out, this suggestion doesn't really work...)
At any rate, we've pretty well licked this question...
---- Gary wrote:
> OM wrote:
> > You set the inputs to whatever the appropriate values would have
> > been at the start of the chart 33 weeks ago (13000 5-min bars =~
> > 33 weeks). Then the 52-week high and low can easily and
> > automatically be adjusted for any changes that take place during
> > the time covered by the chart:
>
> Ah. I assumed you meant to manually input the CURRENT 52wk H & L on
> a weekly basis. That's the only way you could do it with only two
> inputs -- but it wouldn't work for backtesting.
>
> You could do what you suggest, but you couldn't just give it one H
> and one L. To properly update the 52wk H/L forward from the 33-wks-
> ago starting point of the system, you'd have to "prime" the system
> with the H & L for *each* of the previous 52 weeks and let it move
> forward from there. Otherwise you wouldn't know if the 52wk high
> happened last week or 51 weeks ago, and therefore if it should drop
> out in the 2nd week of your 33wk strategy test.
>
> (And given TS's limitations, I'm not sure how you'd feed the system
> 104 inputs, short of putting it right in the system code.)
>
> > If DatetoJulian(Date) - DatetoJulian(Date[1]) > 1 and AHHold > AH
> > then AH = AHHold;
> > If DatetoJulian(Date) - DatetoJulian(Date[1]) > 1 and ALHold < AL
> > then AL = ALHold;
> > {Above two statements adjust 52-week H & L (AH and AL) each week.
>
> Close, but not quite. They adjust it any time Date skips more than
> one day between bars. That would happen over a weekend, but it would
> also happen for any midweek trading holidays. I think testing for
> DayOfWeek(Date) < DayOfWeek(Date[1]) is a safer test. The market
> would have to be closed for more than 5 consecutive weekdays before
> that test would fail.
>
> Gary
>
|