DSGE Models and Optimal Monetary Policy in a Data-Rich Environment

最近的宏观经济研究投入了相当多的精力来开发和估计动态随机均衡（DSGE）模型，这些模型内部一致，并基于第一性原理。最近一些涉及大量摩擦和各种类型冲击的微基础DSGE模型似乎在重要维度上复制了数据，并具有可观的样本外预测性能。在这些有希望的结果的推动下，这些模型现在越来越被视为有用的预测工具和决策的宝贵投入。然而，这些经验模型的一个潜在的重要限制是，它们只利用了少数宏观经济指标。这与央行和市场参与者监控数百个经济指标的事实不符，也与强调大量宏观经济指标重要性的最先进的预测模型不一致。该项目的第一步是建立DSGE模型估计的一般经验框架，有效利用大量可用数据。不是假设理论概念（例如，通货膨胀和产出）可以通过单个数据系列适当地衡量，而是通过许多嘈杂的指标将这些视为不完美的观察。来自与模型概念有未知关系的指标的信息也可以被利用。该估计基于MCMC算法的一种新变体，该算法有效地处理了估计问题的高维性，并继承了经典极大似然估计的特性。本文提出的实证框架具有以下几个优点：1)有助于更好地识别测量误差带来的结构性冲击；2)有可能提高估计效率，从而提高模型的预测性能；3)与动态因子模型的宏观经济应用不同，它产生具有明确经济解释的潜在变量；4)对数据集中的所有序列都有预测；5)它提供了一种自然的方式来记录结果对先验的敏感性。第二步将这一经验框架应用于最先进的DSGE模型，这是Smets和Wouters（2004）模型的一个变体，该模型显示出有希望的经验成功。初步结果表明，加入更多的信息会影响经济周期的重要结论，显著提高模型的预测性能。第三步推导出一个数据丰富的稳健最优货币政策目标标准。现有的衍生品对如何从大型数据集中最佳地利用信息保持沉默，这可能是一个关键的操作考虑因素。在数据丰富的经验框架内，项目的这一部分将现有的稳健最优目标标准推导程序与不完全信息下最优货币政策的现有结果结合起来。更广泛的影响：本提案的目标是促进开发一种能够产生有用的、实时的政策处方的操作模型。拟议的方法可以帮助中央银行系统地实时处理大量信息，通过一个结构模型的镜头，在各种经济发展、预测和政策处方之间保持可解释的联系。它也可能对私营部门有用，因为它提供了最佳政策立场的基准和对广泛经济指标的隐含预测。该项目的一个具体输出是开发一个基础设施，使pi能够实时地持续估计、预测和计算最佳策略设置，这些设置将在网站上发布（连同数据和各种计算机例程作为估计的基础）。最后，所提出的框架的操作便利性使其作为一种教学工具特别有吸引力。调查小组计划以此次提案的结果为基础，扩大现有的MBA案例。

Recent macroeconomic research has devoted considerable efforts to the development and estimation of dynamic stochastic equilibrium (DSGE) models that are internally consistent, and based on first principles. Some recent micro-founded DSGE models, which involve numerous frictions and various types of shocks, appear to replicate the data in important dimensions and have appreciable out-of-sample forecasting performance. In part motivated by these promising results, these models are now increasingly perceived as useful forecasting devices and as valuable inputs to policy making. However, one potentially important limitation of these empirical models is that they exploit only a handful of macroeconomic indicators. This is at odds with the fact that central banks and market participants monitor hundreds of economic indicators and is inconsistent with state-of-the-art forecasting models that emphasize the importance of large sets of macroeconomic indicators.The first step of this project develops a general empirical framework for the estimation of DSGE models which makes efficient use of the large amount of available data. Instead of assuming that theoretical concepts (e.g., inflation and output) are properly measured by a single data series, these are treated as imperfectly observed through many noisy indicators. Information from indicators that have an unknown relationship with the model's concepts can also be exploited. The estimation is based on a new variant of an MCMC algorithm, which deals effectively with the high dimensionality of the estimation problem, and can inherit the properties of classical maximum likelihood estimation. The proposed empirical framework has several advantages: 1) it helps to better identify structural shocks from measurement errors; 2) it potentially makes the estimation more efficient which could improve the model's forecasting performance; 3) it yields latent variables that have a clear economic interpretation, unlike in macroeconomic applications of dynamic factor models; 4)it has predictions for all series included in the data set; 5) it provides a natural way to document the sensitivity of the results to the priors.The second step applies this empirical framework to a state-of-the-art DSGE model, a variant of the Smets and Wouters (2004) model, which has shown promising empirical successes. Preliminary results show that adding more information affects important conclusions about business cycles and improves significantly the model's forecasting performance.The third step derives a data-rich robustly optimal target criterion for monetary policy. Existing derivations are silent about how to optimally exploit information from large data sets, which might be a crucial operational consideration. This part of the project combines, within the data-rich empirical framework, existing procedures for the derivation of a robustly optimal target criterion, with existing results for optimal monetary policy under imperfect information.Broader Impacts: The objective of this proposal is to contribute to the development of an operational model that can produce useful, real-time, policy prescriptions. The proposed approach could help central banks systematically process a large amount of information in real-time, through the lens of a structural model that maintains an interpretable link between various economic developments, forecasts and policy prescriptions. It could also be useful to the private sector, by providing a benchmark of the optimal policy stance and implied forecasts for a wide range of economic indicators. A concrete output of this project is to develop an infrastructure allowing the PIs to continuously estimate, forecast, and calculate the optimal policy setting in real time, which will be published on a website (together with the data and various computer routines underlying the estimation). Finally, the operational convenience of the proposed framework makes it particularly appealing as a teaching tool. The investigators plan to expand an existing MBA case on the basis of the results from this proposal.