Stocks market modeling and forecasting based on HGA and wavelet neural networks

A method for stocks market modeling and forecasting is proposed based on hierarchical genetic algorithm and a wavelet neural network with continuous parameters. Design of the wavelet neural network, different from the existing one that determines structure of the network and parameters of the wavelet separately, is completed by a well-designed hierarchical genetic algorithm proposed. Thus, based on the AIC Criterion a fitness function is set up and the proposed hierarchical genetic algorithm is then used to train the wavelet neural network, with the structure of the network and parameters of wavelets, including connection weights, stretching parameters and movement parameters, all determined at the same time. A case study is finally carried out with practical data sets acquired from Shenzhen stock market composite index and Wanke Stock price, respectively, showing a good performance of the new method. It can then be concluded that the proposed hierarchical genetic algorithm and wavelet neural networks can be widely applied to model and forecast uncertain systems such as stocks markets.