Modelling Economic Time Series Under A Bayesian Frame of Reference

利用时间序列分析的贝叶斯方法和协整系统的半参数规范检验，开发了分析经济时间序列的新程序。这些方法的实证应用包括分析美国经济的宏观经济时间序列数据，韩国、澳大利亚和新西兰的宏观经济数据，以及一些长期股票价格和股息序列。进行了大量的仿真实验来评估新程序的性能。该项目开发的新的估计方法将提高对广泛问题的实证经济研究的质量。这项工作尤其及时，因为最近对将贝叶斯方法（本项目使用的框架）应用于实证经济研究的兴趣有所增加。该项目的主要活动是关注时间序列分析的客观贝叶斯方法。特别注意经济时间序列，其行为表明可能的非平稳特征。确定基于模型的参考先验的问题，以适应非平稳将被详细考虑。贝叶斯时间序列分析中数据调理的影响一直是人们关注的焦点。在前一笔赠款中制定的概念框架扩展到贝叶斯模型似然检验、后验几率检验和模型选择标准。模型选择标准提供了一个广泛使用的标准的泛化。详细探讨了贝叶斯推理的所有这些特征，并为一般类型的时间序列问题建立了渐近理论。协整半参数规格检验的工作依赖于拉格朗日乘子原理。LM方法为结构系统的长期运行元素提供了模型规范测试，并针对规范不足（长期运行关系太少）和过度规范（长期运行关系太多）进行测试。项目的两个部分将通过开发贝叶斯模型规范测试相关联，在协整系统的情况下，该模型规范测试将与LM测试程序密切相关。

New procedures for analyzing economic time series are developed using Bayesian methods of time series analysis and semiparametric specification testing in cointegrated systems. Empirical applications of these methods include analysis of data on macroeconomic time series for the United States economy, macroeconomic data for Korea, Australia and New Zealand, and some long stock price and dividend series. Extensive simulation experiments are conducted to evaluate the performance of the new procedures. The new estimation methods developed by this project should improve the quality of empirical economic research on a wide range of problems. The work is especially timely because of the recent increase in interest in applying Bayesian methods, the framework used by this project, to empirical economic research. The main activity of the project is concerned with objective Bayesian methods of time series analysis. Specific attention is given to economic time series whose behavior indicates possible nonstationary characteristics. Issues of determining model-based reference priors that accommodate nonstationary will be considered in detail. The effects of data conditioning in Bayesian time series analysis is the major focus of attention. The conceptual framework developed in the previous grant is extended to Bayes model likelihood tests, posterior odds tests and model selection criteria. The model selection criteria provides a generalization of a widely used criterion. All of these features of Bayesian inference are explored in detail and an asymptotic theory is developed for a general class of time series problems. The work on semiparametric specification testing in cointegration relies on the Lagrange multiplier (LM) principle. The LM approach delivers a model specification test for the long-run elements of a structural system and tests against both underspecification (too few long-run relations) and overspecification (too many long-run relations). The two parts of the project will be related by developing a Bayes model specification test which, in the case of cointegrated systems, will be closely related to the LM test procedure.