Subject: re : var calibration issues
we are proposing the following changes to the calculation of ng correlations :
1 . weight the data set ( 3 calendar months ) used in calculating correlations
( most recent data weighed heavier )
2 . use respective contract prices , instead of prompt month prices ( i . e . for
nov - 00 correlations use nov contract prices for the last 3 months , as opposed
to prompt month prices for the last three months .
tanya ,
i have confirmed with ted and he gave us green light to make both changes .
did we get an opinion from vince ?
winston ,
it is my understanding , that this changes apply to ng correlations only , not
the correlations between commodities . we will test the changes in gas and
then decide on world - wide implementation . any estimate on timing of this
implementation ?
cassandra ,
ted suggested that you and veronica should document this as a change in var
parameters and inform all commercial desk heads of these changes . we intend
to make them for na gas first , but ultimately make these changes consistent
across all commodity groups . let me know if you have questions .
thanks , vlady .
wenyao jia
10 / 13 / 2000 03 : 43 pm
to : vladimir gorny / hou / ect @ ect
cc : tanya tamarchenko / hou / ect @ ect , jin yu / hou / ect @ ect
subject : re : var calibration issues
vlady ,
also in the meeting , we identified that there are still some issures
regarding to the correlation matrix calculations .
since different commodity has different expiration dates . when calculate
correlation between two commodities , the two may have different prompt
months . are we going to use prices on two different prompt months or are we
going to use the prices on the same month disregarding prompt months .
because above issues , jin is not going do any changes on the correlation
matrix calculation until above issures can be solved .
thanks !
winston
tanya tamarchenko
10 / 13 / 2000 03 : 16 pm
to : vladimir gorny / hou / ect @ ect
cc : wenyao jia / hou / ect @ ect , jin yu / hou / ect @ ect , jin yu / hou / ect @ ect
subject : re : var calibration issues
vlady , we met with winston and jin today regarding var calibration issues .
the outcome on this discussion is :
1 . jin will put weights into calculation of factor loadings ;
2 . jin will change the way factor loading are calculated . for each commodity
the prompt month contract will be selected for the effective date of vatrfacs
run .
then the historical prices will be collected for 3 month for all forward
contracts starting from
selected prompt month contract . the variance - covariance matrix will be
calculated
based on these data , it will be converted into correlation matrix , then
factor loadings
analysis will be performed on the correlation matrix .
tanya .