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: current trend who wrote (25)	12/30/1997 12:12:00 PM
From: Optim	Read Replies (1) \| Respond to of 871

>In reading your posts it appears you have experience with several >different NN software packages. Besides price how do they differ? >Since you are presently using neuroshell, is it the best? I have used Neurostock, Neuroshell Trader, a demo Of NeuroGenetic Optimizer, a demo of Neurshell Easy, TDL and a few others that are shareware based. What I have found with most of the packages is that you require a good set of mathematical skills, logic skills and probably some statistical skills thrown in. The Neurostock package is the easiest of all the packages, but also the most limiting. You are stuck with a 10 day forward forecast. The NeuroGenetic package is great, but it is really complex, and I doubt that most would have time to develop the skill set required to use it. The packages from Ward Systems (NS Trader, NS Easy Predictor) are really easy to use, so I may be a bit biased towards them. I don't have the time to mess with Excel macros to load and convert stock data, or to sit a tweak the initial weights of inputs. The Ward packages tend to shield the user from the complications of the nets, while still yielding good results. And the Trader package can read Omega Research, Metastock, CSI, AIQ, and ASCII formatted data. It also has all the TA indicators built in, so it is as easy to use as SuperCharts or Metastock. I like the walk-forward out of sample testing also, as it can show you the forward results of the nets in a matter of seconds. Again, having used a limited number of packages, I am probably biased towards those that I use. But I have to say compared the the general category of NN software, the Trader package is the most friendly for time series (stocks, commodities, etc) forecasting. Anyone else agree/disagree? >Can you speak to the amount of time it takes to get a NN up and >running with the type of results upon which you would risk your >hard earned dollars? I presently use a system of about 25 price >based indicators to make buy/sell decisions. Is this too many >to feed into a NN? Getting a net up and running is a combination of finding a good net that works over extended periods of time, and find a good trading strategy that complements the net and makes money based on it. I would probably not trade it until it achieves a 65-70% win ratio on out-of-smple tests. If you currently have a system based on 25 indicators, using a package like Trader you could simply re-enter these indicators (chosen from the 600 or so included in the package) into a new prediction. Then you check your results. Or you could forecast the 25 (or probably part of the 25) indicators into the future, so as to allow earlier entry. The NS Trader package has a limitation of 150 inputs per net, and you should be well below that, or else you are probably over-fitting you system to the data. You want a net to generalize as much as possible, to maintain its effectiveness in the future. You can however, have more than one net, and use a sort of consensus buy and sell strategy. If you go to neuroshell.com and look at the Delta example you will see what I mean. Optim