Composite Forecasting with Dirichlet Priors*
从贝叶斯视角研究组合预测，用狄利克雷分布表达用户对某方法优于其他方法的先验信念，并以美国生猪价格为例说明该方法。
ABSTRACT In this paper, composite forecasting is considered from a Bayesian perspective. A forecast user combines two or more forecasts of an operationally relevant random variable. We consider the case where outperformance is modeled as a realization from a multinomial process. The user has prior beliefs about the probability that a particular method outperforms all others, information which is summarized by the Dirichlet distribution. An empirical example with hog prices in the United States illustrates the method.