|
PureBytes Links
Trading Reference Links
|
Er, I got lost three posts back, guys... (:-)
Anyone have some code to play with that might plot the central 'trend
line'...?
Ian
> Gary Fritz wrote:
> >The only way I know of to get results like that is to curve-fit
> >to the entire data series. That is, each point is calculated
> >with full knowledge of past AND future points. A polynomial fit,
> >as Audrey mentioned, is one possibility.
>
> Looks that way to me too, come to think of it.
>
> >Of course, you couldn't use that to trade, since you have no
> >knowledge of future data points (bummer!). If that's what those
> >charts really are, then they're just dummied-up pretty sales
> >tools that couldn't actually be used for trading.
>
> Not necessarily. You don't need future points to fit a polynomial
> to data. Furthermore, you don't need a polynomial with many degrees
> of freedom either, if you're concerned with seasonal tendencies as
> seems to be the case here. A 3rd or 5th degree polynomial should be
> sufficient for getting trends on a year's worth of data. Simply fit
> it to 250 days (approx 1 year) and re-fit on every new bar, plotting
> only the most current endpoint of the polynomial. This is no more
> valid or invalid than using a SMA or EMA or linear regression
> projection to identify a trend.
>
> >he couldn't be using a polynomial-fit curve for his trading
> >decisions.
>
> I think he could in this context. Think of it this way: for markets
> with seasonal tendencies, the tendencies average out to a sinewave
> having a period of 1 year. A sinewave for a single period can be
> well-approximated by a polynomial.
>
> --
> ,|___ Alex Matulich -- alex@xxxxxxxxxxxxxx
> // +__> Director of Research and Development
> // \
> // __) Unicorn Research Corporation -- http://unicorn.us.com
>
>
|