基于信息准则的最优组合预测小样本改进模型

在对样本量小且波动大的变量进行预测时,最优组合模型往往容易出现过拟合问题而导致预测效果不佳.借鉴信息准则中对过拟合问题的处理方式,结合等权组合在预测实践中的良好表现,通过在最优组合模型的目标方程中增加向等权收缩的惩罚项,建立了最优组合预测小样本改进的二次规划模型.最后通过实例,用最优组合预测和其他常用组合预测方法结果的比较,说明了该方法的可行性和有效性.

Small Sample Improvement of Optimal Forecast Combination Model Based on Information Criteria

When forecasting the variable with small sample and huge volatility,the optimal forecast combination may usually appear the problem of overfitting easily.And the overfitting problem often goes with bad forecast performance.Aiming at this problem,the process of dealing with the overfitting problem in information criteria can be borrowed.Since the equal weights combination performance good in forecast practice,the penalty term that shrinks the weights towards equality is added to the object equation in optimal combination model.Then the quadratic programming model of improving optimal forecast combination in small sample is established.Finally,an example is given to show the feasibility and availability of this method by comparing with optimal forecast combination and usual forecast combination methods.