摘要
利用BP神经网络提取非线性特征,对残差运用广义自回归条件异方差模型分析波动性,最后将趋势与波动性结合给出预测区间。以2001年2月至2017年6月美元/欧元汇率的日收盘价数据为例,研究发现:BP神经网络具有很好的非线性刻画能力,但"过拟合"以及"欠拟合"均会影响预测区间的精度,只有合适的误差大小和精度标准才能得出较好的预测结果;同时也发现广义自回归条件异方差模型能够较为准确地分析波动性,且组合模型优于单一模型,适合中长期的区间预测。
In this paper,BP neural network is used to extract the nonlinear characteristics,and the variance is used to analyze the volatility using the generalized autoregressive conditional heteroskedasticity model.Finally,the trend and volatility are combined to give the forecast interval.The results of the analysis of the daily closing price data for the US dollar/euro exchange rate from February 2001 to June 2017 shows that BP neural network has a good non-linear characterization ability,but 'over-fitting' and ' under-fitting' will affect the accuracy of the forecast interval and only the appropriate error size and accuracy criteria can lead to better prediction results.It is also found that the generalized auto-regressive conditional heteroskedasticity model can analyze the volatility more accurately.The combined model is superior to the single model,which is suitable for medium and long interval prediction.
出处
《科技和产业》
2017年第11期141-147,共7页
Science Technology and Industry
