I think the direction we are going to go is to first increase the number of factors, then "tear" down the gas and power silos and jointly estimate.

Frank
PS Immediate feedback was that patches can become legacy issues and it is best to solve issue correctly?

 -----Original Message-----
From: 	Lew, Jaesoo  
Sent:	Friday, June 01, 2001 2:11 PM
To:	Hayden, Frank
Cc:	Jia, Winston; Tamarchenko, Tanya; Kaminski, Vince J; Andrews, Naveen; Ramesh, Ganapathy; Yu, Jin; Gupta, Sanjay; White, Stacey; Gorny, Vladimir; Port, David; Stock, Stephen; Brackett, Debbie R.; Basu, Nilay
Subject:	RE: Project X - Outside the box idea

Frank,

In technicality wise, I don't think it is an obstacle and the term correlations among power and NG can be obtained to use hedge purpose.  But in simulation point of view, the POWER-NG joint factors may be misleading since the "blurred" factors can be generated if the NG cores and power regions have low correlations.  

However, if the power desk uses NG as a hedge, definitely the concerns on the joint factor loadings in VAR calculation are natural.

Jaesoo 

 -----Original Message-----
From: 	Hayden, Frank  
Sent:	Friday, June 01, 2001 1:33 PM
To:	Jia, Winston; Lew, Jaesoo; Tamarchenko, Tanya; Kaminski, Vince J; Andrews, Naveen; Ramesh, Ganapathy; Yu, Jin; Gupta, Sanjay; White, Stacey; Gorny, Vladimir; Port, David; Stock, Stephen; Brackett, Debbie R.; Basu, Nilay
Subject:	Project X - Outside the box idea
Importance:	High

Since we are potentially very short of time, I wanted to "float" an idea out there for discussion and investigation.  If you believe it to be good, and we find ourselves in a jam, I want everyone to have an understanding of the proposed process.  Please think about this and realize that we may not seriously discuss this solution until the week of June 11th, when I return from vacation, at which time we can reassess the need for such a patch.  (THIS IS NOT A REQUEST TO IMPLEMENT, BUT FOR DISCUSSION ONLY.)

Background:
Factor loadings are currently done for three separate groups, natural gas, power and liquids.  There are many discussions regarding increasing the number of factors in the model and implementing PCA across the groups.  It is my understanding that joint estimation is done for each group, and the correlations in that group are embedded in those factors.  One critical problem we have in power, is that the desk hedges using NG, and that in the VAR simulation, the NG factors are "pulled" in from another factor group, i.e. the natural gas factor group, possibly causing VaR measurement issues. 

Additionally, in the factor loading process, one has to pre-identify potential core curves for each group.  If one were to include in each group's core curve list other curves that have been previously included in other groups, one would most likely create "factor-overkill", thereby destroying the value-at-risk engine and reducing it to nothing. (maybe I'm a little dramatic here, but you get the point?)  

Idea:
Introduce to the power core curve list and group, natural gas curves, so that risk engine would only use factors generated appropriately for power, and not have to borrow them from another group.

Solution:
To do so, Enpower would have to create (and copy 60 days worth of history) and introduce, unique but also identical, natural gas curves.  These curves would be inserted to the curves utilized for creating power factors -so that the factors used for power VAR, include the appropriate NG curves (used by power traders to hedge their exposure).  These NG-PWR curves would be renamed so that they are "unique" to that commodity group and system.  Upon factors being generated, positions would have to be moved to the NG-PWR curve(s) so that Value-at-risk would pick up the new factors. (If positions are not moved, the VAR engine would pull in the factors generated in the natural gas factor loading process, not the power factor loading process...)  

Additionally, factors generated in other commodity groups would not be utilized in power VAR because they wouldn't have "delta" on them,  rather the delta would be on the new NG-PWR curves, utilizing the factors appropriately generated respective to power.  

(For those joint calculated factors, where we assume perfect correlation, we would have to "ensure" that those "curves" stand out and are not done jointly..)


I recognize that this sort of patch has the potential to become a "legacy"?.

Let me know your thoughts,

Frank