Abstract: For the reason that the extreme values have not been considered in traditional weighting methods of optimal combination forecasting models, there exist certain drawbacks. In order to improve the accuracy of combination forecasting, the concept of tolerable error is developed. The concept is then applied to effect analysis of each single forecasting model and the combination forecasting model, a novel optimal combination forecasting model with tolerable error is developed, in which the weighting vector of single forecasting models is determined. Finally, taking the exchange rate data between America and Japan in the period of 2016-01-04 to 2018-06-12 as an example, the feasibility and validity of the proposed model are illustrated. The numerical result shows that the developed combination forecasting model can effectively improve the prediction accuracy.

Key words: tolerable error, forecast accuracy, combination forecast, weight coefficient

摘要： 鉴于传统的最优化组合预测模型确定权重的方法未考虑到数据中极端值的影响，存在一定的局限性。为了更好地提高组合预测的精度，提出了预测结果可容忍误差的相关概念。将其应用到每种单项预测结果及组合预测结果的效果分析中，提出了一类基于可容忍误差的最优化组合预测模型，从而确定组合预测中各单项预测方法的权重。以2016年1月4日至2018年6月12日日元兑换美元日汇率收盘价的数据为例验证了模型的可行性和合理性。结果表明，所提出的最优化模型能够有效提高组合预测的精度。

关键词: 可容忍误差, 预测精度, 组合预测, 权系数