Message-ID: <3510465.1075857036140.JavaMail.evans@thyme>
Date: Tue, 15 Feb 2000 09:01:00 -0800 (PST)
From: zimin.lu@enron.com
To: brad.horn@enron.com
Subject: Re: Storage Model: Simple Issues
Cc: stinson.gibner@enron.com, vince.kaminski@enron.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Bcc: stinson.gibner@enron.com, vince.kaminski@enron.com
X-From: Zimin Lu
X-To: Brad Horn
X-cc: Stinson Gibner, Vince J Kaminski
X-bcc: 
X-Folder: \Vincent_Kaminski_Jun2001_9\Notes Folders\C:\Technote\Mail\Techmemos
X-Origin: Kaminski-V
X-FileName: vkamins.nsf

Brad,

Here are my thoughts on your concerns.

* You needs curve inputs.  This is an IT job.  I can help you for the curves 
before the system is properly set up.

* Intrinsic value vs Time value: 
  The intrinsic value depends on how you allocate the volumes.  If you have a 
rough idea about the allocation as
  you did in the spreadsheet, we can calucate the intrinsic value within the 
model.    The difference between the 
  total and the intrinsic will be the (option) time value. However, by 
pre-allocating volumes, you killed some options.
  
  In the storage model, volumes are allocated dynamically, therefore it is 
hard to distinguish the intrinsic vs. time value. 

* Factor of loading:  Factor of loadings are used to give historical 
correlation matrix. The three factors correspond to
  paralle shift, slopping and curveture.  The covariance matrix in the model 
is expressed in the form
     
  COVAR = row( vol_{i} ) *( Correl(i,j) )* Colum( vol_{j} ) where vols are 
the implied volatilities from the vol curve.
  
  (Correl (i,j)) =L*L' +residue (small)
  
where L is the factor of loading matrix.  So in a simple words, the factor of 
loadings (say, 60x3) are a simplier way for us to 
remember the historical correlation matrix (say, 60x60).

Let me know if I can offer further help.

Zimin






   Brad Horn                02/15/2000 07:15 AM

To: Zimin Lu/HOU/ECT@ECT
cc: Sandra Henderson/HOU/ECT@ect 
Subject: Storage Model: Simple Issues

Zimin:
 Thanks for your time with the revised storage valuation.  Your right to 
point out the similarity to market bids.  Here are some basic questions tied 
to implementation and calibration:

Model Infrastructure/IT Support: I obviously need to re-build my link to the 
forward curves as the model is not working in my new location.  Short-term 
(aprox 1 month), I'd like to establish a link to the ENA database EGSPROD32 
in order to fetch the long-dated price and volatility curves.  My link to ENA 
forward curves would then be quickly severed in favor of the curves generated 
by the new Bridgeline entity (database name and data structure yet to be 
defined).  However, its not clear to me what is required in this two stage 
process to support your model.  Any definition of model input or minimum 
support requirements you provide is appreciated.  I'll then work with Sandra 
Henderson, an Enron employee providing our IT support, to ensure the model 
continues to work regardless any downstream system changes that may take 
place as we build and establish our separate trading systems or databases.  
Meanwhile, is there anything you think you can do to ensure Im up and running 
quickly?

SANDRA: linking EXCEL spreadsheets to Bridgeline forward curves will be key 
to all our pricing projects, not just the storage model supplied by Research.

Intrinsic vs Extrinsic Value: it would be helpful to decompose the model's 
calculated storage price and to distinguish intrinsic vs extrinsic (time or 
option) value.  I could easily link a new spreadsheet tab to your model 
inputs and to calculate the intrinsic value,  and then through a simple 
difference I could  determine the extrinsic value.  Ive included a simple 
spreadsheet calculation for the intrinsic value for review.  I wanted to 
share this with you to ask the following:
does the nature of the model define intrinsic and extrinsic value differently 
than the simple difference proposed?
do you think it would make sense to do the simple value decomposition in the 
backcode C-code via .dll in order to ensure run-time is faster?



My goal here is straightforward: a) to better understand the model and its 
sensitivities.; and b) to determine if and when the option approach is 
associating significant value above and beyond the simple present value of 
the time spreads.  

Factor Loadings: What are some of the thoughts or insights you can offer with 
regards to factor loadings and how I should interpret the graph of the 3 
factors calculated?  Factor loadings have always been a mystery to me.  For 
example, what problems should I be looking for as a warning against 
mispricing?  What, if anything, is implied about 1 day price change or 
expected curve re-shapings (after all, curve-reshapings are key to storage 
valuation!!!)?

Calibration: we are preparing a simple summary of descriptive statistics 
which should allow me to refine some of the model inputs.  I'll share the 
data when we are and model results once Im up and running.

