Message-ID: <15158563.1075852727272.JavaMail.evans@thyme> Date: Wed, 3 Oct 2001 16:29:06 -0700 (PDT) From: ravi.thuraisingham@enron.com To: john.arnold@enron.com Subject: RE: Neural Networks Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-From: Thuraisingham, Ravi X-To: Arnold, John X-cc: X-bcc: X-Folder: \JARNOLD (Non-Privileged)\Arnold, John\Inbox X-Origin: Arnold-J X-FileName: JARNOLD (Non-Privileged).pst John, here is a power point slide that provides a draft outline of the problem at hand. It is very draft in nature but I wanted to get the working version over to you ASAP. I wanted to get this discussion going via written format so that I (or others who may implement this) can stay focused on what you want and not get into broader research, etc.... I think I can put a model together if we can define what parameters, the interface (levers) and model (transfer functions) to use that would be useful for you as phase I product. I will fire off updates to this document as I make them. If you have a spec doc or ideas that you want to hand write on a print out, please do so and I will update the document. I will like to make sure that I am on the same page before beginning to code the program and to start linking to additional moths (month 2, 3, ..) and different curves, etc. My Job search has gone well with the Crude desk and I am waiting compensation indication from them to make my decision. Also, Kevin Presto feels that I could potentially help his power group with new market or spread trading. John Suarez is a director who came from the power desk to EBS is going back to work for Kevin Presto. I may have an opportunity to build the southeast market by supporting John. I am very grateful that I met you and other key Enron traders during this job search. Ravi Thuraisingham, CFA Director, Storage Trading Enron Broadband Services p 713.853.3057 c 713.516.5440 pg 877.680.4806 ravi.thuraisingham@enron.com -----Original Message----- From: Arnold, John Sent: Monday, October 01, 2001 3:47 PM To: Thuraisingham, Ravi Subject: RE: Neural Networks not necessarily looking for predictive power. that's a 3 year project. just for market making skillset. The work that Dave Forster did was just for front month. That creates month 1. Then, similar logic has to create a month 1/month 2 spread to create a month 2 outright market. Same for month 2/ month 3 to create month 3. There might be 24-36 individual markets to create a forward curve. Sometimes month 1 has correlation to the month 1/ month 2 spread. Sometimes it does not. Must create a system that is mechanical but very easy for a human to add bias. -----Original Message----- From: Thuraisingham, Ravi Sent: Monday, October 01, 2001 10:20 AM To: Arnold, John Subject: Neural Networks John, I just wanted to give you heads up that I did look into the subject system to learn from what your trading activities and then figure ways to automate some aspects of you daily activities. I will try to send you a few power point slides showing my initial thoughts on the system. It appears that neutral network (AI is a subset of this class of learning systems) type of model that takes input from all available sources (including actual market feedback, weather and other fundamentals) and uses curve building functions and other existing tools as transfer functions, along with your own thinking (your processing functions that your neurons are wired up to do), could help the neural network to learn and eventually provide the necessary predictive power. Ravi Thuraisingham, CFA Director, Storage Trading Enron Broadband Services p 713.853.3057 c 713.516.5440 pg 877.680.4806 ravi.thuraisingham@enron.com