SI
SI
discoversearch

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: alan von weiler who wrote (14)12/29/1997 9:53:00 AM
From: Optim  Read Replies (1) | Respond to of 871
 
I also own the commercial (100 Neuron, Genetic Capable) version of Neurostock. I created nets for about 20 tech stocks, and have been testing the nets since the last week of November. So far, in closed trades they have made about a 3% return, and the open trades are currently down about 18%. Unfortunately I think this is due to a general sell-off in the tech sector. I also haven't tried shorting a stock when a sell signal is given, which I think in this case would yield some really good returns. I agree that Neurostock is a good program, but it takes forever to train compared to some of the other programs I have used. It is however very simple to use and the batch option automates training and prediction reports. Overall I am happy with it...

But, being the type of person that likes to tinker with a system, I bought the Neuroshell Trader in the hopes of designing a custom-tailored system to each stock. I have now been using it for about a month, and I am pleased with the results, although I don't think I have a truly tradable system yet. The best results I have managed has been a solid 250% annulaized return. But the same net managed a -66% return during the subsequent 6 month out-of-sample test. So it pretty useless...

Still, I have come up with a few nets that achieve about 60-70% annualized returns, and have been consistant on 3 out of sample tests. Its just that it takes a lot of time (and work) to get a decent system going for each stock. I am happy with the NS Trader package though, as it only takes anywhere from 10-60 seconds to fully train a net. This allows various setups to be tried quite painlessly and allows you to come up with good combinations. I have also tried creating 3 nets at once, predicting a 1,2 and 3 day forward % change in price. When the trading strategy is set to buy or sell only when all three are in synch, it yielded the best returns so far, testing at 220% annualized and 80% out of sample on CSCO.

I am confident that I will have a great trading system up by the end of January, and hopefully one that will generalize well on a number of different stocks. I will post a few results as I go along...

Optim



To: alan von weiler who wrote (14)12/29/1997 10:20:00 AM
From: Optim  Respond to of 871
 
>They seem to be more versatile in usage. I would like to try some
>shareware other than NeuroStock. Especially ones with a genetic
>algorithim. They are available but are not compiled (C++) and
>therefore beyond my expertise.

You may want to try Genehunter from Ward Systems, wardsystems.com as it is an MS Excel plug in Genetic Algorithm. I bought this about a week ago, and have been pleased with it so far. It works pretty quick, although not as fast as a compiled C or VB program.

The way it works is pretty simple. You set up you spreadsheet with a row of cells that are your chromasome. These could be your values for indicators for example. Then you set up you worksheet formulas to use these chromasome values, and set up a 'fitness function' to evaluate how the genetic algorithm is doing. In my case the fitness function was the correlation of the indicators to the predicted output. This allowed me to optimize the values of my indicator inputs to the neural net, and reduced the amount of error produced by the nets. This is probably the closest you can get to a full-fledged GA without programming.

I think there is also another company that offers a GA Excel add-in, but it doesn't work under Windows NT. They are located at iea.com

Optim