Why Doesn't Everyone Use Neural Networks?

If neural networks are so powerful and so obviously useful, why doesn't everyone use them for market prediction?

The Problem

The answer to this is preprocessing, the Achilles heel of neural network trading system development. Neural net trading system developers spend 85% to 95% of their effort on preprocessing input data. Preprocessing is the collection, selection, manipulation, detrending and smoothing of neural network input data. In the case of financial markets, these are the input indicators employed to predict the market. Proper preprocessing is the KEY to successful neural network market prediction. The amount of effort employed is frequently beyond the ability of all but institutions to afford. This is largely due to the fact that almost everyone employs spreadsheets for this part of the process. Spreadsheets work well for research on a small number of problems, but require significant effort each time the net must be changed during development. In addition once the system has been developed it must be updated with new data over time. We have often heard that a good network was generated but the onerous task of keeping it up to date was overwhelming. We heard of one sad case where an excellent network had been developed but one day during spreadsheet manipulation, an error was made and the network never worked again. The problem with a spreadsheet requiring many buttons, clicks and drags is that it is hard to do it the same way twice.

The Solution

This is where Stock Prophet comes to the aid of the trading system developer. Stock Prophet performs the preprocessing functions with a minimum of decisions and keystrokes without any sign of a spreadsheet page. Minimum keystrokes translate into minimum chance for input error. In addition a keystroke recording MACRO capability allows you to record keystrokes in a way that allows updating your network consistently, without error, with new data just by selecting a MACRO name in Stock Prophet and about 10 more keystrokes in the utility program. A typical neural net can be updated with new data, and the buy/sell indicator generated, in about 2 minutes. This process has been made so efficient by Stock Prophet, that often the manager of a small money management firm will run Stock Prophet on a regular basis in addition to his regular duties. Stock Prophet provides cost effective access to neural network decision making technology.

Indicators available for preprocessing