Gia-Shuh Jang, Feipei Lai and Tai-Ming Parng
Dept. of Electrical Engineering
Dept. of Computer Science and Information Engineering
National Taiwan University
Taipei, Taiwan, R.O.C.
This paper presents an intelligent stock trading decision support system that can generate timely stock trading suggestions according to the prediction of short-term trends of price movements using dual adaptivestructure neural networks. We have proposed a structure-level adaptive back-propagation learning algorithm that can automatically synthesize the structure of a neural network to fit the desired problem. The structure of neural networks can also be continuously adapted if the problem domain is changed later. We show that the proposed structure-level adaptive neural networks are superior to fixed-structure neural networks due to the compact size of their hidden layers and improved generalization ability. In addition, we have shown that a portfolio of dual modules of neural networks generalizes better than does a single module of neural network. Finally, for the testing period of 1990 to 1991, the annual rates of returns by using trading decision supports generated by the proposed system were higher than those using the buy-and-hold strategy.
Keywords: dual adaptive-structure neural networks, stock trading decision support systems, modeling, trend prediction, generalization
Received October 15, 1992; revised May 5, 1993.
Communicated by Jhing-Fa Wang.