Re: diffpara

[Gerd Bürger]

I have the impression that the diffpara function from the signal package 
is flawed. For example, it is obvious that

-------------
  D = zeros (b - a + 1, k);
  for l = 1:k
    for m = a:b
....
      D(m-a+1) = - x(1:m) * y(1:m) / sumsq (x(1:m));
    endfor
  endfor
-------------

doesn't make much sense as D is 2-dim. Apart from similar mistakes the 
fractional-d estimation does not seem to use the formula suggested by 
the reference (Brockwell, GPH regression).

I don't know who is responsible for the signal package. The author of 
diffpara, F. Leisch, was unresponsive, so I wonder what should be done 
in this case. I could provide patch that at least restores the intended 
algorithm in its simplest form.

Gerd