An empirical research of forecasting model based on the generalized regression neural network
Xinjiang Guo, Yao Xian Xiao, Jinglun Shi · 2008
In this paper, we study the theory of generalized regression neural networks, a kind of radial basis network that is often used for function approximation, and apply it for the forecasting of the Shanghai composite index of the Chinese stock market. The raw data consists of 4245 observations of daily closing values of the Shanghai Composite Index spanning the trading dates December 19, 1990 to April 25, 2008. Each group of the training set is composed of 130 observations of daily closing values of the Shanghai composite index. The neural network we established has four layers of neurons: the input layer, the radial basis layer, the special linear layer and the output layer. Each of the first three layers has 11 neurons and the output layer has one neuron. Five statistics of forecasting error, including ME, MAE, RMSE, MAPE, and SE, are used for the evaluation of the forecasting results. The simulation results show that the generalized regression neural network we constructed is able to forecast the daily closing price of the Shanghai composite index and the effectiveness and high performance are demonstrated by the simulation results and five statistics. Therefore the forecasting model based on the generalized regression neural network is able to result in good prediction and has research value to the reality.