预测模型组合开关

不同预测模型在同一时刻都存在不同的预测误差,最优组合预测的预测精度要比其他模型高,但并不是所有时刻误差都最小,有的时刻可能比单项预测误差还大,因此,在预测时就存在较大的风险.再组合预测可以降低预测风险,但方法比较复杂难以推广.针对这些问题,我们设计一种预测模型组合开关,根据数据的变化,不同时刻采用不同预测模型,使预测误差每一期都最小,或不是最大.实证结果表明,预测模型开关能够根据数据特征有效控制模型输出.

A Switchgroup for Forecast Models

The different models of the forecasts have the different errors at the same time. The optimum combination forecasts model produce more accuracy than others, but its errors are not the lowest one at every time in the forecast period for the reasons that the accuracy is the mean value. There would be forecast risk at some time. The recombination forecasts may reduce the risk of the forecast. The complexity of the method, however, blocks the popularization. For the reasons above, we design the system of the switchgroup for forecasts which can use the different model at the different time according to the variation of data and it makes the errors the lowest or not the highest. The empirical result reveals that the system can select the out-performance model or the non-inferior one.