![]()
Visiting guest. Why not sign in?
|
|
Robotrading: Beating Wall Street Neural Networks, Genetic Algorithms, and Expert Systems | |
Robotrading: Beating Wall Street
US News has an interesting new feature entitled Robotrading 101: Sophisticated Computer Programs Take the Human Element Out of Picking Winners on Wall Street. They discuss multiple ways that computer agents (they're not really robots, but hey, that's the press for you!) can attempt to beat the stock market prediction. One Neural Network based solution was fed a history of 70 years of trading, learning from its experience. I assume the network must have been pretty big! The final product can do fairly well, sometimes beating the trend by 3%. An expert system approach is briefly mentioned, but the genetic algorithm based solution seems the most interesting. However, this is just a dreamed up solution by 'sci-fi' author Ray Kurzweil. We're yet to find out how well it performs. These agents are only doing well since they compete against humans. Against other agents, they'll struggle, and no amount of experience will prepare them for that. Wether they can learn from feedback and adapt quick enough is to be seen. |
|
Black - Scholes Equation
The Black - Scholes Equation is a partial 2nd order differential equation that is related to the wave equation through a change of variables transformation. It can be used to model the pricing of stock futures. I bring it up because in the 80s there was this group (wish I could remember their name) that put together a mutual fund where they made their purchases and sales based upon the results of the model. Signing up for the fund required two things: 1. a minimum of a million dollar investment and 2. the investor had to keep their money in the fund for a minimum of 3 years. This group made trillions of dollars. (Yes, I got the amount correct.) However, there was one minor (sarcasm) problem, the model could not adapt to rapid changes. When the warning signs of the Asian Economic meltdown started to appear, the model said don't worry it will rebound. The mutual funds managers ignored their intuition and went with what the model told them. Had they started to pull out at that point, the group would have had just over 4 trillion dollars. However, as the Asian economy continued to tank, they lost it all. In fact, there was so much money tied up in the fund that after a point they couldn't pull out without further precipitating the crisis. On the small scale, I think models/agents are great, but, at least for now, you still have to look over their shoulders and make sure that they still apply. |
|
Documentary
Yeah, I saw a program on that too! They were billions in debt at the end, but the initial investors are still well off and have left the insurances to pay for the 'damage'... Quite funny, as all they really needed was a bit of human supervision and sense to stop the endless buying to compensate the drop! Funny... and only in America ;) |
|
Genetic Algorithms for Portfolio Management Patented
For anyone interested, we have a patent in this area (Patent # 6,188,992)available for licensing. The patent can be seen at www.uspto.gov. |
|