Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Strategies & Market Trends : Neural Nets - A tool for the 90's -- Ignore unavailable to you. Want to Upgrade?

To: xu, b. who wrote (715)	3/28/2000 9:54:00 AM
From: Optim	Respond to of 871

Hi Bai, >I found Ruggiero's book very informative and full of SCIENTIFIC ideas.< I agree. The best thing a person can do is work though his book chapter by chapter and try things for yourself. My copy of the book is worn from overuse! >Am wondering if anyone could point me to a right direction to develop a NN trading system.< I can offer a few guidelines, but in the end it is through your own work that you will be successful. My best stock models have used a combination of intermarket inputs, combined with seasonal influences. It's funny, but technical indicators don't contribute much to the end result, and often models won't even have any inputs based on the traded issue! I obtained my best results by mixing timeframes (weekly and daily) and normalizing inputs to be level independant. This can be done by using defferences, subtracting a moving average, using a z-score, etc. Watch for outliers, which are much more prominent in stock data than most of the indexes. Also, don't model on too short a timeframe. A percent change in close shifted one day forward is wildly profitable, but also very difficult to model. I tend to use a 3-5 day momentum smoothed by a 3-5 average shifted a few days into the future. What you'll find is that if you shifted it forward or backward by a day or two it won't affect the profitability by very much, which means the system won't break down if the neural network is off by a little. Try modeling on weekly data to get your feet wet. It is easier to develop some decent models. Good luck! Optim