Determining a Portfolio of Linear Time Series Models
提出一种基于后验概率比和证据等级的非机械方法，用于选择线性时间序列模型的阶数，并构建多个合理模型的组合，以辅助预测等后续任务。
The paper applies selection criteria to order determination in linear time series models in a less mechanical fashion than is often found. This is achieved via posterior odds ratios and by applying concepts of grades of evidence advanced by Jeffreys (1961). It is shown how alternative specifications may be compared and this leads to the development of a practical method for determining a portfolio of different models, each of which may be regarded as reasonable for the data and which could be useful in subsequent tasks such as forecasting. To complement the theory, some simulation experiments are reported which illustrate the practical efficacy of the procedure.