Stock investment decision support for Hong Kong market using RBFNN based candlestick models

Abstract

Candlestick pattern, as an efficient method in technical analysis, is widely used in decision support of stock investment. From historical data, there is a no 100% guarantee for a stock price increasing after the appearance of a bullish candlestick pattern. The main aim of this paper is to enhance the prediction ability of Candlestick Patterns using a Multiple Classifier System (MCS) consisting of Radial Basis Function Neural Network (RBFNN) trained by a Localized Generalization Error Model (L-GEM). The RBFNN classifies particular candlestick pattern to be a real bullish candlestick pattern or not based on training with past data. The MCS fusing RBFNN for different patterns makes the final prediction of the stock price trend. In this paper, stock price data of 40 stocks in Hong Kong Hang Seng Component Index is collected to carry out the investment simulation experiment. Experimental result shows that the proposed method yields statistically significant profit when compared with a random investment strategy and candlestick investment without RBFNN.