Neuro-rough trading rules for mining Kuala Lumpur composite index

Abstract

Stock market plays a vital role in the economic performance. Typically, it is used to infer the economic situation of a particular nation. However, information regarding a stock market is normally incomplete, uncertain and vague, making it a challenge to predict the future economic performance. In order to represent the market, attending to granular information is required. In recent years, many researches in stock market prediction are conducted using diverse Artificial Intelligence approaches. These artificial applications have shown superior prediction results. As such, in this study, a prediction enhancement alleged as Neuro-Rough (NR) is proposed to forecast the Kuala Lumpur Stock Exchange Composite Index (KLCI) movements. NR hybridizes high generality of artificial neural network (ANN) and rules extraction ability of rough sets theory (RST) by demonstrating the capability of simplifying the time series data and dealing with uncertain information. Features of stock market data are extracted and presented in a set of decision attribute to the NR systems. The length of the stock market trend is used to assist the process of identifying the trading signals. A pilot experiment is conducted to discover the best discretization algorithm and ANN structure. NR is implemented in a trading simulation and its effectiveness is verified by analyzing the classifier output against the information provided in Bursa Malaysia's annual reports. The experiments using 10 years training and testing data reveal that NR achieves an accuracy of 70% with generated annual profit in trading simulation of 74.33%.