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 : NeuroStock -- Ignore unavailable to you. Want to Upgrade?

To: Jay Hartzok who wrote (258)	10/17/1998 12:15:00 PM
From: Optim	Read Replies (1) \| Respond to of 805

Jay, I will try to get a screen capture of the results up today. I have tried it on a few other stocks this morning with poor results also. I am finding that this approach is very dependant on the way the issue trades. Try NN for example. It is a nice volatile issue which seems to work well, and I feed in the Nasdaq Composite, it's sector index, and the usual l/t, price and volume. Unfortunately I use the Canadian equivalent of NN (NNC on the TSE). But you should be able to get similar results. After experimenting a bit more (been up since 6:30) I am noticing that this simply works or it doesn't. By training a net on my fastest machine (a 450 Mhz) I find that within 5 minutes of BP training I know if it is a good net. The verification period will be dead on, as in the case of NN and a few others I have tried. The reason I like this is because I am now in the process of trying the TSE 35 Index stocks, and then I will move up to the TSE 100, 300, etc. I can go through the issues at a rapid pace. It is actually quite neat to watch the nets learn as the scatter graph converges to the diagonal line. By watching only the verification period, I can find inputs that are predictive. I simply add or subtract an input and start the training agan. If the graph converges well, then the inputs I am using have a predictive value. I have also found that 1 year might be a little short for some issues. I have started to experiment with this also, and am finding a sweet spot of 1 1/2 to 2 years on most issues. I think I might have lucked out on the first few. Again, I am still expeirmenting with this, so if someone has a tried and true method, stick with that. Also, try to use TA as a complement to the nets predications. Nothing in life is 100%. Optim