Enron email

Algorithmic Indexing

5 Oct 2000, 14:48 UTC

Re: implementing term-structure of correlations for power

From
Naveen Andrews naveen.andrews@enron.com
To
Tanya Tamarchenko
Cc
Kirstee Hewitt, Debbie R Brackett, Wenyao Jia, Vince J Kaminski, Vladimir Gorny
Bcc
Kirstee Hewitt, Debbie R Brackett, Wenyao Jia, Vince J Kaminski, Vladimir Gorny
Date
Thu, 5 Oct 2000 07:48:00 -0700 (PDT)
Tanya,
        While there is seasonal correlations in power, especially for Np-15 
and SP-15 (same region), the term structure of correlations can be input.  
However, the same correlation structure with similar periodicity may not hold 
between Np-15 and, say, R1B (Neepool), though one would imagine that  
relationship would still be seasonal (summer/winter), with greater noise.  
Even if the correlational term structure is to be done for Power, different 
rules would have to be inputted for different regions.
Naveen
Quoted or forwarded from earlier messages
Tanya Tamarchenko@ECT
10/05/2000 10:42 AM
To: Vladimir Gorny/HOU/ECT@ECT, Naveen Andrews/Corp/Enron@ENRON
cc: Kirstee Hewitt/LON/ECT@ECT, Debbie R Brackett/HOU/ECT@ECT, Wenyao 
Jia/HOU/ECT@ECT, Vince J Kaminski/HOU/ECT@ECT 

Subject: Re: implementing term-structure of correlations for power  

Vlady & Naveen,
I am sending you the factor loadings for power calculated based on the prices 
from 25-jun-00 to 25-sep-00. They look rather messy. 


Thus joint estimation of factors for a few regions does not look realistic.

Instead we might implement the term structure of correlations. 
Instead of heaving 1 correlation matrix based on prompt month prices and 
applying this matrix to all maturities,
we can have a number correlation matrices each corresponding to a particular 
prompt month.
I attached the term structure of correlations between R10 and R11 calculated 
using 5, 4, 3, 2, 1 months
of historical forward prices prior to 30-sep-00 with and without decay factor.

You can see that:
1)  the results are close for 5, 4, and 3 months of history;
2) correlations are periodic with a period of 1 year (this means we can use 
12 correlation matrices calculated from 
first 12 forward contracts and apply these matrices to other forward months);
3) Using decay factor makes the curves a little smoother.

Implementation of multiple correlation matrices will not affect the speed of 
calculations in VAR model significantly.

Please, give me your response,

Thanks,

Tanya.