Using Neural Nets For Intermarket Analysis by John Kean
Changes in Treasury bill yields, Treasury bond yields, gold and the U.S. dollar affect the stock market in subtle but predictable ways. These are complex and challenging relationships to decipher - tasks well suited to the abilities of neural networks, since neural nets are designed to detect patterns in relationships. Here, systems analyst John Kean presents his work on utilizing a neural network for researching intermarket relationships.
Watching changes in market prices and interest rates may suggest that hidden somewhere in these
seemingly . chaotic fluctuations, useful patterns that are discernible may be used to predict price changes
in the stock market. It's easy to speculate on such things, but how does one take an intuitive notion such
as market interdependence, test it and perhaps turn it into a lucrative trading method?
One possible strategy would be to devise a computer program that through trial and error would produce
a sequence of conditional statements. These statements could define certain intermarket conditions that,
if satisfied, would signal the trader to be long, flat or short the stock market. For example, suppose that
during the preceding week gold prices fell, the U.S. Dollar Index rose and Treasury bond and bill rates
fell. Our computer program would signal that it might be profitable to be long in stocks the coming week.
Obviously, the potential for subtle variations in any set of conditions abound, making trial and error
analysis arduous and limited. Neural networks, a form of artificial intelligence that is rapidly gaining
recognition, is adept at handling this type of problem and gives superior results.