Abstract

Investors have expended enormous efforts on trying to find a useful tool or tools that could forecast stock market trends accurately, enabling them to maximize their profits. Past forecasting models, however, have two noticeable drawbacks, summarized as follows: (1) the forecasting used to produce daily forecasts method, and which is evaluated by the forecasting error, may not be particularly useful for the average investor, since trading on the stock market on a daily basis is not his or her typical modus operandi; and (2) stock chart patterns advanced in past research are fixed, and, therefore, do not represent the ideal way to depict stock patterns. To refine past models, this paper proposes a new stock trend recognition model to predict the stock market with three research objectives, as follows: (1) to provide a stock market recognition model, based on an expert’s experience to analyze and forecast a stock market accurately in order to make a profit from it; (2) to propose a reasonable method to extract, as much as possible, bull market patterns from historical data so as to improve forecasting performance; and (3) to offer several trading strategies, using different stock holding periods, to help investors make decisions. To examine the trading return of the proposed model, a 15 year period of the Taiwan stock index (TSI) was used to formulate experimental datasets. To verify the superiority of the proposed model, the buy-and-hold method and Wang and Chan's (2007) model are used as comparison models. The experimental results show that the total index return (%) of the proposed model is 10 times that of the buy-and-hold method and 2 times that of Wang and Chan's (2007) model.

Key words: Stock market forecasting, bull market pattern, stock trend recognition, template matching technique, technical analysis, cumulative probability distribution approach (CPDA).