compute Hill's estimate

[gka]

Hello guys do you know how I compute Hill's estimate?

The routine bellow does not work and I can not find why...

Code: Select all

' compute Hill's estimate subroutine local hill(scalar hill, series x, scalar m) ' get indices stom(x, xvec) vector xranks= @ranks(xvec) vector m vector xtemp ' get first m observations of sorted data vector(m)=xtemp xtemp = 0 for !i=1 to m xtemp(xranks(!i)) = xvec(!i) next ' find m-th point !xm = xtemp(m) ' find mean of first m-1 points !xsum = @sum(xtemp) - !xm !mean = !xsum / (m - 1) ' compute hill hill = !mean - !xm endsub

[gka]

Nobody?

[EViews Glenn]

What are you trying to do?

[gka]

wach the link below:
http://www.utstat.utoronto.ca/keith/pap ... sthill.pdf (see Introduction 1), Hille estimator, in finance, modeling the
tails of the distribution of returns is important in the evaluation of risk (Embrechts et al.,1997; Brooks et al., 2005).. I find the routine in phoroum but does not work, I want to estimate the hill parametr with eviews because if I estimate it, in another software i must rewrite the total routine again.

[EViews Glenn]

Normally don't read papers for the purposes of answering forum questions, but it looked like it would be easy so I took a quick look. I only glanced at the first bit, but if I'm understanding correctly, something like

Code: Select all

!k = 30 series rnks = @ranks(cons, "d") smpl if rnks = !k+1 scalar scale = @mean(cons) smpl if rnks <= !k scalar alpha = 1 / @mean(@log(cons)-log(scale))

should do the trick for k=30.

First line ranks all of the observations in the series (in my case CONS) in descending order.
Next two lines compute the value of the series for the !k+1 rank.
Last two lines restrict sample to the first k descending order statistics, and compute the mean of the log of the scaled data.

Note that as with many things in EViews, you are far better trying to do things using series generation which allows you to avoid loops.

Hope this answers the question.

[gka]

Thanks, I will try and I will tell you the results...

[Lixir]

Are there any results?
free annual credit report government