International transmission of stock market movements: an adaptive neuro-fuzzy inference system for analysis of TAIEX forecasting

This study aims to examine the fundamental forces driving stock returns and volatility across the international stock markets. Logistic regression analysis is used to investigate possible highly correlated among 9 international stock markets with stock market of Taiwan. Afterward, the highly correlated stock indices with Taiwan would be selected as the input variables of adaptive network-based fuzzy inference system (ANFIS) model to predict stock prices and their direction of Taiwan Stock Exchange Capitalization Weighted Stock Index. The experimental results of the proposed model are contrasted with other models including the first-order model (Chen in Fuzzy Sets Syst 81(3):311-319, 1996), weighted fuzzy time series model (Yu in 349:609-624, 2005), simple neural network model (Huarng and Yu in Phys A 363(2):481-491, 2006), multivariate model (Huarng et al. in J Travel Tour Mark 21(4):15-24, 2007), ANFIS with volatility causality (Cheng et al. in Neurocomputing 72(16-18):3462-3468, 2009), ANFIS with AR model (Chang et al. in Appl Soft Comput 11:1388-1395, 2011), and artificial bee colony-recurrent neural network model (Hsieh et al. in Applied Soft Computing 11:2510-2525, 2011). Finally, the proposed model produces with lower inaccuracy rate and offers higher direction preciseness than above previous models. The benefit of this methodology depended on its application of a hybrid approach to predict the stock prices and direction with higher accuracy.