Examining stock index return with pattern recognition model based on cumulative probability-based granulating method by expert knowledge

In this paper, we apply an advanced pattern recognition model based on cumulative probability-based granulating method (Chen and Chen Inf Sci 346:261–274, 2016) using expert knowledge to examine its model efficiency further. The patterns of a bull market are selected from historical stock by a stock analysis expert. The study examines the trading returns of the model using different trading strategies and compares the returns with the original model (Wang and Chan Expert Syst Appl 33(2):304–315, 2007) and 1-year period buy-and-hold method. By using the 15-year period of the TAIEX (from 1995 to 2009) as experimental datasets, we have verified the predictive accuracy and profitability of the proposed model from the experimental results, and discovered three major findings as follows: (1) this research has tested several trading strategies with various stock holding periods and trading criteria with the total index return percentage, and the optimal trading strategy is the “20-day” holding period, which maybe brings better stock index return; (2) one-month investments based on bullish patterns can bring much better profit return than 1-year investments; and (3) although artificial intelligence algorithms have been widely applied in financial forecasting, it is found in the proposed model that expert knowledge still trumps slightly automatic mechanism in recognition “accuracy” but “efficiency”.