Since stock market is a nonlinear system with internal structural complexity and external factors variability, a neural network forecasting model based on stock market sensitivity analysis is proposed. Input and hidden layer neurons sensitivity are computed and insensitive neurons are pruned to optimi e network structure and improve forecasting accuracy;Significance and feedback mechanism of input layer neurons are analyzed by sensitivity to solve black-box problem of neural network. The empirical results with different time spans of Shanghai Composite Index indicate that the proposed model with corresponding network structures can analyze timeliness of macroeconomic policy and attention of stock market. Compared with other neural network models, the proposed model can improve predictive performance with higher forecasting accuracy and more concise structure.%股票市场是非线性系统,具有内部结构复杂性和外部因素多变性,建立基于股票市场灵敏度分析的神经网络预测模型.针对神经网络结构设计问题,计算网络输入层与隐层神经元的灵敏度,并修剪网络中不敏感的神经元,在保证模型泛化能力的同时,实现网络结构精简;针对神经网络黑箱问题,根据输入层神经元灵敏度解决各输入变量对股票市场的重要性和反馈机制.以上证指数为例,在不同的时间跨度下对股票市场运行规律进行学习,并分析不同结构修剪模型的适用性和市场意义.最后,通过与其他神经网络预测模型比较,验证本文模型的有效性.
展开▼
