Contents I. Introduction
In the literature, various theories and methods have been developed to help investors to pursue great profits in the investment. One is a computer-based trading system that produces appropriate investment decision signals (also called trading signals hereafter) on the basis of the available information to assist an investor make a sensible investment. In the past, one kind of widely used trading systems consists of two modules: prediction module followed by trading module. However, this type of trading system is optimized to some prediction criterion (e.g., mean square error), which is not the ultimate goal of a financial investment. Therefore, it often leads to suboptimal performance in the profit-achieved sense. To solve this problem, some efforts have been made along different directions. One is to use a prediction criterion more correlated with common trading strategies such as that proposed in [2]. Another direction is the return-based systems as proposed in the papers [1], [10], where the prediction module and the trading module are merged into one single system that optimizes the returns instead of the prediction criterion.
