8 - A genetic-programming-based approach to the generation of foreign-exchange trading models

Summary

Of all the markets available to the international investor, the global foreign-exchange markets are perhaps the most liquid and heavily traded. The average daily turnover in currencies around the world is estimated to be up to a trillion dollars a day.

Currency markets show myriad different behavior patterns, with quiet periods, intervals of extreme volatility, steady trends, violent reactions to external news events, and positive feedback all seen at various times. Despite the apparent unpredictability of these markets, it is estimated that at least 90% of the turnover in foreign exchange is due to speculative interbank trading by large financial institutions attempting to make short-term returns.

Pressure on capital ratios within the banking sector has led to the investigation of new ways of taking on risk by trading markets in a rigorously controlled manner. As a part of such a strategy, the rest of this chapter describes a class of algorithms to assist in the determination of market direction. However, we stress that directional trading algorithms are only part of a successful investment operation. Equal weight must be given to asset allocation and risk management.

The majority of algorithmic trading systems are those in which a computer program gives directional forecasts for markets, based on recent market activity. Such models are frequently specified in terms of technical indicators or mathematical functions that compress information about an aspect of recent market behavior into a single number. These indicators are then combined into simple rules and used for trading purposes. Sample indicators include moving averages, overbought and oversold indicators, and historical volatilities.