Price Index Forecasting Using BP Neural Network and Wavelet Neural Networks

Price index time series is nonlinear, nonstationary and chaotic, which has structural breaks, and is influenced by many factors. Price index forecasting using classical econometric methods is not possible, because of restrictive and unrealistic assumptions. Therefore, by using recent achievements in applied mathematics and intelligent methods, we try to overcome these problems. In this paper, back propagation neural network and two types of wavelet neural networks are employed to predict the price index where wavelet neural network takes the framework of neural networks and combines it with wavelet transform. The study population consists of Tehran exchange price index for the period of ten years from 2004 to 2013. The research hypotheses have been examined by using two measures: root mean square error (RMSE) and mean absolute percentage error (MAPE). The results of this study shows that loose wavelet neural network (LWNN) has better forecasting accuracy than back propagation neural network (BPNN) and compact wavelet neural network (CWNN). The forecasting error of LWNN is less than BPNN and CWNN in two data sets of train and test. Therefore, they are better methods in forecasting price index.