含非参数趋势的残差MA模型的预测方法

Prediction Method of Residual MA Models with Nonparametric Trend

为了探寻具有非参数趋势的残差移动平均模型的较为合适的预测方法,考虑了基于多项式样条的三种方法:非外推法、线性外推法和非线性外推法.模拟结果表明,线性外推法和非线性外推法拟合的均方误差(MSE)和平均绝对误差(MAE)相等且最小,线性外推法预测的MSE和MAE最小.此外,还对分行业增加值构成-金融业增加值进行了拟合和预测的实证分析,得到了与模拟算例相似的结果.这说明线性外推法是一种较好的预测方法.

In order to find a more appropriate prediction method for residual moving average models with nonparametric trend, three prediction methods based on polynomial spline are considered, non-extrapolation method, linear extrapolation method and nonlinear extrapolation method. The simulation results show that the mean squared error (MSE) and mean absolute error (MAE) of linear extrapolation method and nonlinear extrapolation method are equal and minimum, and MSE and MAE predicted by the linear extrapolation method are the smallest. Moreover, an empirical analysis is carried on the composition of industry added value -- financial industry added value. The results are similar to those of simulation example. This shows that the linear extrapolation method is better.