Hello,

i have tried to re-consturct gretl's calculation code for the local whittle estimation via the mle-command and i am getting very often the failure:

---
Using numerical derivatives
Tolerance = 1.81899e-012
failed to invert OPG matrix GG'
---

Has anyone an idea how i can solve the problem or avoid it? The last value before the failure for d in log-likelihood function is normally ok...but how can i avoid gretl stopping the optimization?

My code is this:

------------

#getting a series y
series e=normal()
series y=fracdiff(e,-1)

#defining bandwith m
scalar m=ceil($nobs^(0.6))

#getting the periodogram
Y={y}
F=fft(Y)
S=sumr(F.^2)
S = S[2:(m)+1]/($nobs/2)
omega=seq(1,(m))'.*(2*pi / $nobs)

#defining starting values
scalar d=-1.0
scalar z=8.0


mle ll = ln((h + r))*(-1)
    matrix t=z*(omega).^(-2*d)
    matrix lt=ln(t)   
    matrix iota=ones(m,1)
    matrix st=S*transp(t.^(-1))
    matrix std=diag(st)
    series h = transp(iota)*lt
    series r = transp(iota)*std
params d z
end mle

------------

Thank you so much in advance!

Kindest regards

---

Alles in einem Postfach – Ich will Hotmail!