Message-ID: <4917845.1075840797203.JavaMail.evans@thyme> Date: Mon, 16 Apr 2001 13:07:00 -0700 (PDT) From: winston.jia@enron.com To: tanya.tamarchenko@enron.com, jaesoo.lew@enron.com Subject: RE: clustering for power Cc: vladimir.gorny@enron.com, vince.kaminski@enron.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Bcc: vladimir.gorny@enron.com, vince.kaminski@enron.com X-From: Winston Jia X-To: Tanya Tamarchenko , Jaesoo Lew X-cc: Vladimir Gorny , Vince J Kaminski X-bcc: X-Folder: \vkamins\VAR X-Origin: KAMINSKI-V X-FileName: vincent kaminski 1-30-02.pst Tanya, As we discussed in the last meeting, to simulate secondary power curve we need correlated jump sizes. This is totally different from the current secondary price curve simulation which assume the perfect correlation and also totally different from the secondary GAS basis curve simulation which is based on the hedging ratio. There are two more issues on my side I need to resolve: 1. I want resolve the power basis curve issue. Currently all power position on these basis curve are actually price positions. We are hard coding this: if power basis we add basis to corresponding region curve. I am trying to remove this hard coding by asking loading the price curve for all these basis locations. 2. Same is true for all those Power F curves. These curves looks similar to those basis curves. Currently we just directly map these F curves to the corresponding region curves. I would also prefer to load the price curves instead of the price differences. From research, I need those jump size correlations. Clearly, all these involve many new development, unless we want to use simpler way to simulate secondary power curves. Regards, Winston -----Original Message----- From: Tamarchenko, Tanya Sent: Monday, April 16, 2001 9:17 AM To: Lew, Jaesoo Cc: Gorny, Vladimir; Jia, Winston; Kaminski, Vince Subject: Re: clustering for power Jaesoo, as we discussed last week on Wednesday meeting can you, please, implement clustering for power curves by geographical region. This involves the following: 1. Deciding together with Risk Control how many geographical regions we want to use and which Enron's curves belong to each region. 2. Deciding together with Risk Control how to choose core curves for each region. This decision can be maid based on the a) position size; b) statistical analysis. There might be other considerations. 3. Doing regression analysis for each curve versus the corresponding core curve. Winston, can is it possible to run VAR for the clustering results obtained by Jaesoo with clustering done by SAS? Should we wait for the stage re-fresh and what is the status on this? Tanya.