A quarterly crude oil price forecasting system based on optimally selected model is proposed. The system integrates forecast results from optimally selected model and extemal experts with low prediction errors,and obtains final oil price prediction values after the meta-synthesis of integrated results and information and experiences owned by internal experts. It is found that the cross-correlogram based multivariate regression model is preferred to vector error correction model and vector error correction model with exogenous variables for quarterly oil price forecasting. Better prediction results obtained from optimally selected model and extemal experts are served as the basis of integration. In practical work, forecast results of this system have been consulted by Sinopec and the Market Expectations Survey System of State Administration of Foreign Exchange (SAFE), and taken the leading position among members of Market Expectation Survey System.%构建了一个以优选模型为基础的季度油价预测系统.系统不仅依靠模型和内部专家,同时以外部专家的预测值作为参考.系统将优选模型与外部专家的预测值集成后,由内部专家根据掌握的信息和经验,对结果进行综合集成,得到油价预测值.研究发现,基于时差相关的多元回归模型比误差修正模型和带外生变量的误差修正模型更适合预测季度油价.系统通过对模型的优选和对专家的评价,使效果更好的预测值作为集成预测的基础.实际工作中已被中石化和外汇管理局市场预期调查系统所参考,预测精度在市场预期调查系统各成员单位中处于领先水平.
展开▼
