TAIWAN 50 ETF FORECASTING AND INVESTMENT ANALYSIS USING BACK PROPAGATION NETWORK

摘要

Neural network is a popular method in forecasting using some input variables and minimizing the sum of square errors between the output values and the target values. In this study for Taiwan 50 ETF price forecasting, we collect nineteen popular indices which are classified into two aspects where the first one is basic analysis variables, and the other is technical analysis variables. We first used the nineteen variables and ETF price to fit a stepwise regression model for selecting the effective variables. Next, using the selected variables to fit the back propagation network model, we derive that the forecasting error in Taiwan 50 ETF price forecasting is better than GARCH(1, 1) model. Finally, an illustration confirms the potential benefits of the BPN model whose transaction strategies make a profitable investment in the transaction of Taiwan 50 ETF.