|
PureBytes Links
Trading Reference Links
|
Gary Funck wrote:
> Has any brave soul tried coding the cubic least squares fit mentioned
> in this month TASC, in Dennis Meyer's article? He references
> Numerical Recipes in C, and states that the exercise of coding
> this in EL is left to the reader ...
Don't convert, instead figure a way to write connecting code so EL
can communicate with the c routines. Numerical Recipes is on line
at Harvard, or was the last time I looked. Alternatively use the
routines from A Practical Guide to Splines by C. de Boor. The code
in that book is Fortran and is available at the Mathematics
Research Center at the University of Wisconsin, Madison.
>
> Anyone tried experimenting with cubic splines, or higher order
> polynomials as a way of getting a better handle on the trend?
Not for trading applications, but for modeling other physical
processes. Higher order polynomials are notoriously bad for
extrapolation. If I was to try something similar, I would begin
with a slightly later technology known as splines under tension.
In a loose, non mathematical, language, these splines are specified
so as to fairly represent the data while minimizing the curvature
of the spline. There is also a class of splines which are
guaranteed to be concave (convex) if the data to be fit is
concave (convex). In my opinion, you are going to need something
"non-standard" to achieve any reasonable extrapolation.
Rod
|