Neural Networks: From the Chalkboard to the Trading Room – The Best of Both Worlds
By Louis Mendelsohn
One sometimes wonders who happened to fundamental analysis in futures trading over the last decade.
The more “subjective” fundamental methods of evaluating supply and demand, assessing inventory levels, discerning seasonal factors and interpreting published reports have given way to “objective” technical analysis. An underlying assumption of technicians is that all relevant forces impacting on a market have already been factored into its price through “price discovery.” Thus, fundamental factors are of little or no value.
The proliferation of software during the 1980s has reinforced the strict technical approach. These programs include a repertoire of widely used technical studies that can be back tested and optimized on historical price data. Now, it seems, most technical traders rely exclusively on this genre of software, without regard for the underlying fundamental or economic forces behind the markets.
Analysis therein involves testing a commodity’s prices history to forecast future direction or to generate trading signals. Scant attention is paid to volume and open interest statistics, macroeconomic data, intermarket analysis or traditional fundamental information. Favoring technical, or just price, data is a nearsighted approach, though an easier one because it requires a limited amount of data: open, high, low and close.
To gain a complete picture of a market’s dynamics, both fundamental and technical inputs should be combined into an overall analytic framework. Yet, the framework should be one providing a disciplined, unemotional, mechanical basis for decision-making. To date, this could only be done piecemeal by looking first at fundamentals to see whether a market is under priced, overpriced or in line, then examining technical factors for divergence or conformation.
Now, as neural computing technology is applied to forecasting of markets, a new generation of “smart” analytic software is about to debut.
“Pre-trained” neural networks will allow traders to forecast prices, trends and signals based on both technicals and fundamentals. Neural networks can take qualitative data and classify it into discrete categories, helping to find relationships and patterns in the data that are not otherwise identifiable. Intermarket data – time series data from other markets – can be used to examine meaningful relationships between markets that are difficult to perceive, given today’s focus on single-market analysis.
Seasonal factors, even those based on the day of the week, can also be ascertained. In forecasting Treasury bond prices, for instance, statistics on the Federal Reserve and banking industry, including Fed funds rates and money supply, can be used, not to mention other available figures such as the structure of interest rates.
Combining technical, fundamental, economic and intermarket data into your own electronic “analyst” is limited only by your creativity and knowledge. Neural networks bring objectivity and discipline to bear on previously ignored information. They sift through disparate fundamental and technical data with the same mechanical objectivity traditionally found in technical analysis software, manipulating daily price information and leading to more accurate forecasts and profitable trading signals.
As neural computing becomes better understood and more widely adopted it will be relied upon increasingly to forecast major price moves ahead of traditional technical signals. Traders using this new methodology will gain a marked advantage over competitors still relying on rather limited means of analysis.
Lou Mendelsohn, President,
Market Technologies, Wesley Chapel, Fla.