BP神经网络与符号时间序列下的金融波动研究

Forecasting Financial Volatility Based on BP Neural Network and Symbolic Time Series

针对金融市场的非线性特征,将BP神经网络与符号时间序列分析方法相结合,利用历史数据对金融波动进行预测。采用上证综指2011—2014年间隔为5 min的高频数据为样本,首先将波动序列符号化,然后利用BP神经网络对样本进行训练和检验,检验结果表明,该方法可有效预测下一时点波动变化情况,达到了其在金融波动方面的预测效果。

According to the nonlinear characteristics of the financial market, a new method of BP neural network combined with symbolic time series analysis was put forward to forecast financial volatility. High frequency data whose sampling intervals were 5 minutes from Shanghai Stock Exchange Composite Index from 2011 to 2014 was used as a sample. Firstly, the volatility series need to be symbolized. Then BP neural network was used to train and test the samples. Finally the symbol value of next time point was effectively predicted. The effect of the method of forecasting financial volatility was proved.