Collaborative Research: Monetary Policy in a Data-Rich Environment

本项目计划通过合作和个人实证研究，明确考虑到货币政策是在数据丰富的环境中制定的，货币政策制定者可以利用数百甚至数千个宏观经济数据系列中的信息，分析货币政策的决定因素和效果。 我们的研究与以往的大多数研究不同，以往的研究通常假设联邦的信息集包含10个或更少的变量，我们还将考虑到货币政策是在真实的时间内制定的;也就是说，我们不假设（与事实相反）美联储手头有最终修订的数据，而是只使用美联储在每个日期实际可用的数据。为此，我们使用了其他人构建的实时数据集，其中我们添加了来自各种联邦储备文件和其他联邦机构的未经修订的数据。我们的基本任务是估计联邦的政策反应函数（PRFs），它将美联储的工具与经济状况的衡量标准联系起来。我们希望尽可能让经济状况取决于当时可获得的全部宏观经济数据。为了使估计可行，我们需要一个降维方案。我们将采用动态因子模型方法，使用其他人在预测环境中设计的方法进行估计。除了允许我们处理序列数量超过观测数量的数据集外，这种方法还允许我们系统地处理数据不规则性，包括不同频率和不同跨度的数据。如前所述，我们还能够考虑到这样一个事实，即在每个日期，美联储可能会关注同一基础数据系列的不同年份（修订版）。我们将允许在PRF的参数的时间变化，使用的方法，使这种模型的估计计算容易在这种情况下。最后，美联储的PRF可以与描述其他经济体的方程一起估计，或者以不受限制的方式估计，而不指定其他经济体。我们有兴趣通过实证分析解决几个问题。从积极的一面来看，我们想看看信息密集型的PRF是否比其他替代方法更好地描述了货币政策的历史行为。 我们还想确定美联储在历史上对哪些类型的信息做出了反应，并看看似乎影响美联储决策的信息是否可以用其对通胀和真实的活动的预测内容来解释。如果后者似乎无关紧要，那么我们将探讨什么可以解释美联储的行为。 如果我们能够对货币政策的系统性和非系统性成分进行更精确的估计，我们希望利用这些结果来完善最近关于这些成分对经济的影响的研究。 从规范的角度来看，我们想解决的问题是，在数据丰富的环境中，货币政策应该如何实施。这是一个比多变量预测问题更复杂的问题，因为美联储的决策问题是动态的，其损失函数可能与标准的计量经济学损失函数具有非常不同的性质。我们认为，解决这些问题将具有实际价值，并对美联储的实际行动提供更现实的描述。

The collaborative and individual empirical research planned in this project will analyze the determinants and effects of monetary policy, taking explicitly into account the fact that policy is made in a data-rich environment in which monetary policymakers can exploit information contained in hundreds or even thousands of macroeconomic data series. Our study contrasts with most previous work, which typically assumes that the Federal Reserve's information set is spanned by ten variables or less. We will also take into account that monetary policy is made in real time; that is, instead of assuming (contrary to fact) that the Fed has final revisions of data at hand, we use only data that were actually available to the Fed at each date. To do this we use a real-time data set constructed by others to which we add non-revised data from various Federal Reserve documents and other federal agencies. Our basic task is to estimate policy reaction functions (PRFs) for the Federal Reserve, which relate the Fed's instrument to measures of the state of the economy. As far as possible, we want to allow the state of the economy to depend on the full range of macroeconomic data available at the time. To make estimation feasible we need a dimension-reduction scheme. We will apply a dynamic factor model approach, to be estimated using methods devised by others in a forecasting context. Besides allowing us to work with data sets in which the number of series exceeds the number of observations, this method permits us to deal systematically with data irregularities, including data of different frequencies and different spans. As already indicated, we are also able to incorporate the fact that at each date the Fed may be looking at a different vintage (revision) of the same underlying data series. We will allow for time variation in the parameters of the PRF, using a method that makes estimation of such models computationally easy in this context. Finally, the Fed's PRF can be estimated jointly with equations describing the rest of the economy, or in an unrestricted manner that leaves the rest of the economy unspecified.We are interested in addressing several questions with the empirical analyses. On the positive side we want to see if information-intensive PRF's provide a better description of the historical conduct of monetary policy than other alternatives. We also want to determine what types of information the Fed has historically responded to and to see if the information that appears to affect Fed decisions can be explained in terms of its predictive content for inflation and real activity. If that latter appear not to matter, then we will explore what can account for the Fed's behavior. If we are able to obtain sharper estimates of both the systematic and non-systematic components of monetary policy, we hope to use these results to refine recent work on the effects of each of these components on the economy. From a normative point of view, we would like to address the question of just how monetary policy should be conducted in a data-rich environment. This is a more complex question than the problem of forecasting with many variables, since the Fed's decision problem is dynamic and its loss function may have very different properties from a standard econometric loss function. We believe that addressing such questions will be of practical value, as well as providing a more realistic description of what the Fed actually does.