A Hybrid Model Based on Neural Networks for Financial Time Series

Because of their fuzzy and non-stationary nature, financial time series forecasting is still a challenge. In this paper, we propose and implement a hybrid model by combining the Maximum Entropy (ME), Support Vector Regression (SVR) and Trend model based on Artificial neural networks (ANNs) for forecasting financial time series. The approach contains three steps: feature and time alignment in data preprocessing, adopting ME, SVR and Trend model for different features as the input for ANNs, and obtaining the final predicted value using Back Propagation algorithm. The feature selection flexibility of ME and global optimality of SVR make the input model better because of its different features, which helps to have a better forecasting accuracy of ANN sin proposed model. Experimental results clearly show that the accuracy of prediction for Chinese closed-end fund net value can achieve 98.3% using the hybrid model, which is more accurate than some institutions or known financial websites in China, and we provide the prediction of real time fund net value for free in our Hai tianyuan knowledge service platformhttp://www.haitianyuan.com.