Theory and Inference for Macroeconomic Policy

宏观经济政策的理论与推论

• 批准号: 0719055
• 负责人: Christopher Sims
• 金额: $ 15.7万
• 依托单位: Princeton University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-08-15 至 2011-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0719055&HistoricalAwards=false
• 关键词: Theory; Inference; Macroeconomic; Policy

Theory and Inference for Macroeconomic PolicyNSF SES 0719055Christopher A. Sims, Princeton UniversityThis research advances knowledge in four areas related to the use of quantitative models to guide macroeconomic policy. One component is study of the foundations of statistical inference for models with large numbers of parameters. The models in use at central banks and other macroeconomic policy institutions often involve hundreds of parameters whose values have to be determined from the data, yet much of the existing literature on inference in models of economic time series focuses on methods that apply only when the number of parameters is small. Some statisticians have argued that when there are large numbers of parameters, the Bayesian inferential methods that underlie standard decision theory perform poorly. While this assertion has been rebutted in the context of some specific models, this project's research develops a more general treatment of the issues.It is now for the first time becoming feasible to construct models of the scale needed in monetary policy analysis at central banks that are both statistically reliable and equipped with a detailed interpretation, in terms of economic behavior, of the equations that constitute the model. There remain several competing approaches, however, varying especially in the relative weight put on statistical reliability versus economic behavioral stories about the equations. The second component of this project advances a particular approach to this type of modeling. In attempting to tell behavioral stories, while at the same time preserving statistical reliability, much of the existing work in this area has ended up with behavioral models that include frictions and "costs of adjustments" that are necessary to match the data, but not fully convincing as behavioral stories. This project treats the behavioral model not as itself a description of the data, but instead as a source of approximate, probabilistic predictions about the behavior of a reliable statistical model, particularly in the long run. Specifically, the project develops methods to use a fully interpreted equilibrium model to generate a prior distribution for the parameters of a structural vector auto-regression (SVAR). The equilibrium model is a restricted version of the SVAR, and the restrictions are relaxed probabilistically, especially at high frequencies where we think the behavioral model does not allow for realistically complex frictions and inertias. This approach may make it unnecessary to rely on arbitrary-seeming adjustment costs in the behavioral model, with no sacrifice in statistical reliability. If we recognize that economic agents can process information at only a finite rate, many of the inertial and random aspects of behavior emerge as optimal, given information constraints. This project extends previous research, which has developed this insight formally only for simple, single-actor models, to more realistic models of market interaction. In a way this third part of the project complements the second, as it seeks to provide a more firmly micro-founded theory for the dynamics of behavioral models.The fourth component of the project concerns the relation between government debt and deficits and control of inflation. So long as markets are confident that deficits will be cut if debt grows large and increased if it grows very small, the usual interest-rate-setting tools of monetary policy can control the price level without reference to fiscal variables. But in many countries markets realistically do not have this confidence, which results in complex interactions between monetary and fiscal policy. These interactions are studied empirically in this project. The results could be relevant to US policy as the fiscal pressures of from an aging population increase.

宏观经济政策的理论与推论普林斯顿大学克里斯托弗·a·西姆斯这项研究推进了与使用定量模型指导宏观经济政策相关的四个领域的知识。其中一个组成部分是研究具有大量参数的模型的统计推断基础。中央银行和其他宏观经济政策机构使用的模型通常涉及数百个参数，这些参数的值必须从数据中确定，然而，关于经济时间序列模型推断的许多现有文献都侧重于仅在参数数量较少时适用的方法。一些统计学家认为，当存在大量参数时，作为标准决策理论基础的贝叶斯推理方法表现不佳。虽然这种说法在一些特定模型的背景下被反驳，但本项目的研究发展了对这些问题的更一般的处理。现在，构建中央银行货币政策分析所需的规模模型首次变得可行，这些模型在统计上是可靠的，并配备了对构成模型的方程的经济行为的详细解释。然而，仍然有几种相互竞争的方法，特别是在统计可靠性与经济行为故事的相对权重上有所不同。这个项目的第二个组成部分为这种类型的建模提供了一种特殊的方法。为了试图讲述行为故事，同时保持统计可靠性，该领域的许多现有工作都以行为模型告终，其中包括摩擦和“调整成本”，这是匹配数据所必需的，但不能完全令人信服的行为故事。这个项目不把行为模型本身作为数据的描述，而是作为一个可靠的统计模型的行为的近似、概率预测的来源，特别是从长远来看。具体而言，该项目开发了使用完全解释的平衡模型来生成结构向量自回归（SVAR）参数的先验分布的方法。平衡模型是SVAR的限制版本，并且这些限制在概率上是放宽的，特别是在我们认为行为模型不允许实际复杂摩擦和惯性的高频率下。这种方法不需要依赖行为模型中看似任意的调整成本，也不会牺牲统计可靠性。如果我们认识到经济主体只能以有限的速度处理信息，那么在给定信息约束的情况下，行为的许多惯性和随机方面就会成为最佳选择。该项目扩展了先前的研究，该研究将这种见解正式地只针对简单的单参与者模型，扩展到更现实的市场相互作用模型。在某种程度上，这个项目的第三部分补充了第二部分，因为它试图为行为模型的动力学提供一个更坚定的微观基础理论。该项目的第四个组成部分涉及政府债务与赤字和控制通货膨胀之间的关系。只要市场相信，如果债务增长很大，赤字就会减少，如果债务增长很小，赤字就会增加，那么通常的货币政策利率设定工具就可以在不参考财政变量的情况下控制价格水平。但在许多国家，市场实际上并不具备这种信心，这导致了货币政策和财政政策之间复杂的相互作用。本项目对这些相互作用进行了实证研究。随着人口老龄化带来的财政压力增加，研究结果可能与美国的政策有关。