Message-ID: <9656335.1075856589556.JavaMail.evans@thyme>
Date: Fri, 5 Jan 2001 02:41:00 -0800 (PST)
From: alex.huang@enron.com
To: vince.kaminski@enron.com, stinson.gibner@enron.com, 
	vasant.shanbhogue@enron.com
Subject: Re: The consultant's model for Gary Hickerson's group
Cc: alex.huang@enron.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Bcc: alex.huang@enron.com
X-From: Alex Huang
X-To: Vince J Kaminski, Stinson Gibner, Vasant Shanbhogue
X-cc: Alex Huang
X-bcc: 
X-Folder: \Vincent_Kaminski_Jun2001_5\Notes Folders\C:\Technote\Mail\Projects
X-Origin: Kaminski-V
X-FileName: vkamins.nsf

The model is supposed to be a real option model to capture the value of
power plants of Gencos. It is to give trader a better insight as to whether 
the
market is overvaluing/undervaluing certain Genco stocks, and trader can
act accordingly.  I'm still trying to find out how trader is supposed to use 
it. 

Modeling details:
The model takes in all Gencos' locational power forward prices and fuel 
forward
prices, and uses GARCH model to simulate one year daily prices, and then
uses a hourly profile to convert them into hourly prices.  GARCH model 
parameters
are estimated by the consultant using and separate model and 
are updated twice a year, and it does not matter whether the simulation starts
in January or September. 

Using these prices, it will determine whether a unit at a particular location 
will be dispatched
or not depending on A) spread of power and fuel prices, and B) whether the 
start-up
cost can be recovered during 8 operation hours.  The unit can be dispatched at
minimum and peak levels. Fixed O&M, SOX and NOX (I don't know what the last 
two stand for)
are taken into consideration. 

With the simulated dispatch schedule, the model calculates the value that can 
be generated
by this unit, then sums it up across all units.

The final value is the average of 100 simulations.  And it takes about 16 
hours to run for about
200 units.  

After our conversation, the consultant promised to look into A) how to make 
the model more flexible,
say, to allow a different time horizon, B) reduce spreadsheet overhead by 
doing calculation one
unit a time and not saving all the intermediate information (as of now it 
saves everything 
on the spreadsheet).

Assuming the GARCH process is modelled correctly, I believe the methodology 
is OK, though 
it does not capture most of  the optionality. 

My concerns are:
Whether the price processes are modelled correctly. I have to get more 
details before making
any conclusion.

100 simulations are way too few. Unless we convert the algorithm to C, I 
don't see how spreadsheet
can handle more simulations. I guess that's why they contact us. But again, 
if Enron's buying the
model from the consulting company, why should Enron do their job for them?

How trader's going to use the model output. For this I phoned Jeff (the 
associate who initiated all
these) and am still waiting for his returning call.  A related questions why 
the model horizon is one year. 

We can either
Oversee the conversation, but not doing actual coding for them.
Or
Redo the model for them. (The problem still remains that how trader's going 
to use the output).  But 
in view of the Great Wall of China separating the business units, should we 
do it?

As of now I have a simulation model taking start-up cost, fixed O&M, rump-up 
delay into consideration.  
It simulates monthly prices (using GBM) and takes 2 minutes 40 seconds to run 
10,000 simulations for 
one unit for ten years (120 time steps).  It can use forward-forward vol and 
incorporate seasonality into
it (I understand this is debatable). (One interesting observation is that 
when using forward-forward vol
simulation, the standard deviation is about 0.5%, while standard deviation 
using forward vol is about 
2%. Also, incorporating seasonality increases the value by about 1.6%). Since 
most of the time-cost 
occurs in price simulation, and we are to simulate about 20 price processes,  
I hope the full model 
(if we build it) will take a couple of hours to run for 200 units. The main 
task will be interfacing, i.e., 
getting data from data base, and outputting the results. This is where I need 
most help if I am to do it.

Please advice the course of action. I am supposed to talk to Michelle 
Cisneros today. 
p.s. I never promised to oversee a programmer in our group (see the message 
below). 

Best,
Alex
---------------------- Forwarded by Alex Huang/Corp/Enron on 01/05/2001 08:58 
AM ---------------------------

	
	Jeff M Gray
	
	01/04/2001
	
To: Gary.Hickerson@enron.com, Michael.W.Bradley@enron.com, 
Michelle.D.Cisneros@enron.com, Jaime.Gualy@enron.com
cc: alex.huang@enron.com, kskinner@ftenergy.com, cseiple@ftenergy.com 

Subject: FW: Project timeline

Ken and I worked up the following timeline and refined the trading 
methodology a bit this morning.  We also met with Alex Huang from Vince's 
group, and explained the model and coding tasks.  Ken and Alex have arranged 
to speak by phone on Monday, and meanwhile Alex is coordinating within the 
research group.  Alex will oversee a programmer within his group, while 
interfacing regularly with us.

1/4 Kickoff
1/11 Complete spreadsheet, table, and database structures (RDI).
1/17 Complete software coding for the Pricemaker component of the model
(RDI and Enron Research), and begin testing (Enron Research).
1/22 Complete software coding for the Dispatch portion of the model (RDI
and Enron Research), and begin testing (Enron Research).
1/22 Complete financial trader "user" interface, within the Access
environment (RDI).
1/22 Complete collection and delivery of unverified generating-unit data from 
RDI databases (RDI).  Begin verification process (RDI).
1/29 Complete all charts and reports accessible from the user interface
(RDI).
1/29 Complete compilation of consensus EBITDA forecasts for all operations 
other than merchant generation (Enron Financial Trading).
2/9  Complete code testing (Enron Research).
2/9  Deliver verified and quality-checked generating-unit data (RDI).
2/9 Complete the model, begin testing the trading methodology, and train 
users.
2/16  Finish training, testing, and final QC.


jeff