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 (717)	3/28/2000 10:18:00 AM
From: Optim	Read Replies (2) \| Respond to of 871

Hi Bai, <It's an informative forum and has been a little bit dead recently. > I'll say! I just about fell out of my chair when I saw 4 new posts! :) Seriously, in order to possibly spark up a few ongoing topics, here is one observation I have made over the last year. I wouldn't mind people offering there opinions if they have a moment. Most of the inputs that have worked well for me have also worked well when tested as a simple trading system. For example, in Ruggeiro's book he offers an intermarket test that is used by many trading systems. It is simple and often robust if the relationships are valid. The method is to trade X by watching to see if Y is above its moving average. I have used this (or bettr preprocessed variations of it) with great success to model smaller cap stocks by looking at larger cap stocks, often in the same or related sectors. Many time these models will make huge profits when tested in Metastock, although drawdowns are often high. I optimize the inputs by shifting them back 1 to 50 days and by optimizng the periods used in the moving average. Then I export the optimization run to Excel or ScatterPlot (a free prg, try hotfiles.com to search for it) and build a 3D surface plot or scatter plot. I make the shift the X axis, the moving average period the Y axis and Profit the Z axis. What I look for is a good sized cluster of values that are reasonably profitable. If there is none, then I discard that input as the relationship between the traded issue and the leading issue is too unstable and will probably breakdown over time. Make sure if you try this that you leave out a period of data as an out of sample period. I usually test up to 12/31/98 and then leave the rest of the data to be used a NN out of sample periods. This method does require a lot of work to build a model, but the end results tend to hold up and I have much more confidence in them as I know they are built on solid relationships in the data. A good article of this type of analysis is in the current Technical Analysis of Stocks and Commodities magazine. I think it was the April issue, but I'm not sure. Optim