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 : Systems, Strategies and Resources for Trading Futures -- Ignore unavailable to you. Want to Upgrade?

To: F Robert Simms who wrote (15526)	2/15/1999 2:10:00 PM
From: Robert Graham	Read Replies (3) \| Respond to of 44573

So in broad terms what you are saying is that neural net based trading breaks down when the market changes and requires a different approach to trading. At this point in time, a new model needs to be created. However, when a neural net has been built and proven for a given market, then good profits can be made. This period of profit making can last for months. Finally, not all types of models work with the neural net approach to the market. By the way, thank you for your up front and honest description of your experiences with neural nets. Bob Graham

To: F Robert Simms who wrote (15526)	2/15/1999 4:53:00 PM
From: Brian Hornby	Read Replies (1) \| Respond to of 44573

Thanks very much. I guess I would advise sticking with your own system, most traders I know who have been successful long term have their own system that they trust and follow. Have you considered just trading an Emini with the system? Trading options seems to stack the odds against you, however I guess the loss limit feature is attractive. I read on the Profit site discussion on the GRNNs. They do give best back tested results, more or less by definition. However I believe the consensus is that they fall apart with out of sample data. Have you tried using a large interval for the hold back date? My procedure is to do the model building with a hold back date early 1998 (say March), then optimize with the pre-hold back date, then finally evaluate the models with the out of sample (post hold back date) data, without any further optimization. Perhaps that is the standard routine for most people and I am sure I am not saying anything you don't know. I am also finding that sometimes adding more variables to a system completely screws it up. For example, today I added inputs based on the SP close to my bond model, and most of the models the system built were negative versus 100% profitable models from the old system. Next thing I will do is start at basics, develop SP models based just on the SP price (and perhaps volume), ditto for Bonds, then add other variables one by one, for example Advance-Decline to the SP model and interest rates to the Bonds, and re-evaluate. In that manner I will know exactly the effect of each new variable and move to complex models only if they add profitability.