Subject: re : estimating tail of distribution and additional risk measures
naveen ,
the " analytical var " approach is working for equity portfolio .
it gives us the tool to examine the tails ' behavior for this portfolio and
calculate " expected tail loss " .
the same should be done for commodities portfolio as well .
meanwhile , as we discussed , we can give some rough estimates of the losses
corresponding to percentiles other than 5 th .
look at the figure below . you can see var numbers for 5 % , 1 % , 0 . 5 % and 0 . 1 %
calculated with
1 ) simulations ( 100 thousand simulations ) ;
2 ) analytical var ( gamma - delta positions representation )
1 ) and 2 ) are very close because there are not many options in equity
portfolio .
3 ) simulations ( 1000 simulations ) to calculate 5 % var . then in order to
approximately estimate var for 1 % , 0 . 5 % and 0 . 1 %
i scaled 5 % var with factors corresponding to normal distribution ( for
example : norminv ( 0 . 001 , 0 , 1 ) / norminv ( 0 . 05 , 0 , 1 ) for 0 . 1 % ) .
the result of such extrapolation in this case is quite good ( just 5 %
different from the correct number ) .
we probably can use such rough estimates of tail for commodities portfolio
until we have proper methods implemented .
tanya tamarchenko
02 / 28 / 2001 01 : 17 pm
to : wenyao jia / hou / ect , debbie r brackett / hou / ect @ ect
cc : vince j kaminski / hou / ect @ ect
subject : re : " analytical " var implementation in risktrac
debbie ,
i am forwarding to you a 2 page document describing implementation of
" analytical " var in risktrac .
here is why this effort is very important :
1 . we need to calculate var for other percentile but 5 ( 1 % or even 0 . 2 % as
mentioned by rick buy )
and our simulation model can not handle required number of simulations ;
2 . we need to present additional risk measures ( such as mean tail loss ) to
the board .
the analytical approach is implemented in a spreadsheet and fully tested
already so there will be no problems
with the algorithm itself .
we need to get together and discuss it implementation .
what do you think ?
tanya