基于多重分形谱的神经网络建模及股价指数预测

Neural Network Modeling and Stock Price Index Forecasting Based on Multifractal Spectrum

基于多重分形理论,对上证指数进行实证研究,确认了多重分形谱参数与股价指数及股指收益率之间的统计关系,以此确定神经网络的输入、输出变量来构建以多重分形理论为依据的神经网络模型,并将其应用于股价指数的预测中。结果表明,该神经网络模型能够取得比较好的预测效果,预测的平均准确率达98.9%,而且该模型能够较好地模拟股市的短期走势,对防范和控制风险具有现实意义。

Based on the multifractal theory, this paper presents an empirical research on the data of Shanghai Stock Price and analyzes statistically the correlations between the parameters of the multifractal spectrum and the stock price index and the logarithmic return. Through this correlation, this paper determines the input and output variables of neural network model based on multifractal theory and applies it to forecasting the stock price index. Results of prediction experiments with real data prove the efficiency of the prediction method based on multifractal spectrum. The average veracity gets to 98. 9%, indicating that this model can simulate stock market trend of short term. It is useful to prevent and control risks.