🚀 go-pugleaf

SOFTWARE/THEORY WANTED FOR REGRESSIONS ON EXTREMES

I want to fit the model Y_i= b.x_i+e_i , i=1...n, given {x_i}

where e_i are iid observations from a Generalised Extreme Value distribution as
defined by Hosking et al (1985) Technometrics p252. This is a 3-parameter
model. Given b one can estimate these 3 parameters say a, using their L-moment
(PWM) method  which for their range of k and n (to 100) is more efficient than
the MLEs -  (though not for very large samples.)

As their L-moment method does not appear easily extended to this regression
problem I propose to start with b=0, use their estimate for a, compute the MLE
for b, estimate the residuals as e_i =Y_i - b.x_i, and iterate.

I haven't worked out any theory for variances etc yet.

Any comments welcome; eg variances, references, software.

As I don't read most of the news groups I'm posting this too
(namely comp.misc, comp.sources.wanted, aus.archives,sci.math.stat )
an e-mail reply would be appreciated.

Kia ora!   

Kit Withers	email srwmcsw@wnv.dsir.govt.nz