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: Optim who wrote (650)	9/1/1999 4:36:00 PM
From: LastShadow	Read Replies (1) \| Respond to of 871

Years ago I had a lot of success with fourier analysis, but since I mostly concentrate on tech stocks, that hasn't proved too useful in the last 30 months. Tuning wavelets tends to end up being stock specific, and a whole lot of retuning has to occur to be cnsistently profitable. But it isn't necessarily trial and error as you can usually overlay the major cycles and look for patterns from that point down. I do apply some technical indicators after smoothing data, but they are generally vvery short cycle indicators rather than 7 or more days - the reason is that smoothing will desentize your data to important reversal days and make them late in showing up on your screens - adding a nother filter on to p of that, such as a MACD or Stochastic or whatever, adds another smoothing factor and more importantly another delaying factor. you can't use the lead/lag relationship (as that gives an after the fact "enter yesterday..." signal), but you can use the close lag relationships with the wavelets and no other smoothing or technical indicators longer than 3 to 5 periods

To: Optim who wrote (650)	9/1/1999 6:59:00 PM
From: LastShadow	Read Replies (1) \| Respond to of 871

Everyone here subscribe to this? Very good coverage. To add a new email address or remove your email address from this newsletter, visit: ai.about.com and click the subscribe or unsubscribe button.

To: Optim who wrote (650)	9/3/1999 1:58:00 PM
From: Tim Krieg	Read Replies (1) \| Respond to of 871

Short FFT ramps and spike analysis FFT & Wavelets can be a help, however they can be as tricky (useless) as a 12 handed sprocket gizmo. Some of the wavelet filters are not as well suited as others. Don't get me wrong, by going from just frequency to time- frequency domain they are extremely useful, however they should carry the label don't try this at home. I enjoy working with wavelet components and filters, however the fft components have worked out better to build a better $$$ model. Try ramping up the fft components over time, summing fewer components on older data and using all of the components on the most recent data. It will allow fewer spikes to get through from special market conditions that may no longer be valid. Take Y2K, it may cause some spikes now that will have little impact 2002. Those spikes now are important, and may not want to be smoothed out in the near term. You may also try comparing the SFFT results to the SFFT input and filter out completely the error of various portions of your training set when the difference is above a threshold. The approach has been used in an opposite way in other domains, focusing or training to look for the spikes. Tim Short FFT (SFFT) - a wavelet wannabe