Amelia Marsh via R-help <r-help <at> r-project.org> writes:
>
> Dear Sir,
>
[snip]
> Lastly using the command rsnorm(10000, mean = m, sd = s, xi = x)
> where m, s and x are the estimated parameters obtained from loss
> data. The usual procedure is to arrange these simulated values in
> descending order and select an observation representing (say 99.9%)
> and this is Value at Risk (VaR) which is say 'p'.
> My understanding is to this value 'p', I need to apply the
> transformation 10^p to arrive at the value which is in line with my
> original loss data. Am I right?
[snip; sorry to remove context, but Gmane doesn't like it]
(1) you can probably calculate the 0.999 quantile directly from
qsnorm(0.999, [params]) rather than by simulating ...
(2) ... I believe that my original example used log(), so you
would need to use exp() (not 10^x) to get back to the original scale ...
(3) ... if you're concerned about extreme events it would be
a very good idea to use the skew-t rather than the skew-Normal
(4) you should certainly consider Boris Steipe's concerns about
non-independence (although I have to admit that without more
information and further time/effort/thought I don't have any
simple suggestions how ...)
cheers
Ben Bolker
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
