Message-ID: <11651410.1075858480371.JavaMail.evans@thyme>
Date: Fri, 18 May 2001 14:36:06 -0700 (PDT)
From: kaminski@enron.com
To: vkaminski@aol.com
Subject: Var Issues
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-From: Kaminski, Vince J
X-To: 'vkaminski@aol.com'
X-cc: 
X-bcc: 
X-Folder: \Vince_Kaminski_Jun2001_10\Sent Items
X-Origin: Kaminski-V
X-FileName: vkamins.pst

A few comments on the state of our VaR models. The issues always become more
controversial during the periods when vols and positions increase and I would
like to share with you a few observations.

1. Critical importance of a  good risk management system to
Enron (and the P/E ratio).

A number of recent analysts' reports characterized Enron as a hedge fund engaged in
massive directional trading and representing huge risks to investors. One way to counter this
crirticism is to emphasize the quality of Enron's risk amnageemnt system and
our strong risk amangeemnt culture.  Our claims regarding the strong risk management
technology and culture should be, however, supported  by reality.

It is critical that we continue to improve our risk management systems and also can
convince the external parties (stock analysts, creditors,  credit rating agencies) about
the quality of our financial technology. This can be done 
through targeted disclosure of the elements of our technology, as we have done frequently in the past.
by reality.

2. Inputs from the trading desks vs. historical calibration.

The VaR system used in Enron is based on the assumption that as much input
information as possible should come from the market (i.e. trading desks), or should be calibrated to
the market observations (as opposed to historical data).

One immediate improvement: we should review inputs that are not based on the trading desks
information and move, if possible, to replace them with market based, forward looking data.

One should not assume, of course, that the information coming from the
trading desks is perfect. One problem that is persistent in some portfolios is the disconnect
between the posted forward vols curves and subsequent fluctuations in the forward prices 
curves. The forward vols curves should be relatively unbiased expectations of future realized 
forward price volati;oty. If this is not the case, back testing for some portfolios will 
systematically fail. 

One way to address the problem is to implement a module that compares systematically
fwd vols posted by the trading desks with subsequent price movements. It will allow us to identify the 
source of problems we see in some books.

3. VaR process management.

Given the size of our trading operation, we are running a VaR factory,
a system designed to produce timely risk assessment for all the portfolios.
What is missing is the industrial type process that guarantees the quality
of the results.

Specifically, I am concerned about absence of one person vested with the responsibility
for the model calibration. Right now this responsibility is dispersed in the organization
and resides somewhere between the IT, RAC and Research. As a result we have
inputs updated inconsistently and without proper quality controls. Some inputs used 
currently in the system are 2 years old.

Another aspect of this problem is prioritization of different projects. In my view 
some recent modification to the system tied up IT resources and were of marginal
usefulness (and in some case on questionable
understanding of mathematics). Of course, I can be wrong but there is was no systematic process to
bring these issues to the table, in a meeting involving all the concerned parties

and