Learning With Misspecified Models

使用错误指定的模型进行学习

• 批准号: 0004315
• 负责人: In-Koo Cho
• 金额: $ 23.09万
• 依托单位: University of Illinois at Urbana-Champaign
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2001
• 资助国家: 美国
• 起止时间: 2001-02-15 至 2005-01-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0004315&HistoricalAwards=false
• 关键词: Learning; Misspecified; Models

This project investigates the agent who estimates a model that fits the observed data well, although he may never learn the "true" equilibrium because his model is "misspecified" in the sense that it misses one or more key parameters of the environment. Even with a misspecified model, the decision-maker's belief can be self-fulfilled and he may end up using the model to choose a policy. Although the idea of self-fulfilling belief is closely related to rational expectations studied under equilibrium paradigm, it is called a self-confirming equilibrium to emphasize that the two different models interact to generate the observed outcomes: the agent's model, which is misspecified but used to select the action, and the true model that generates the actual data. The interaction between the true and the perceived models is the central part of this research that differentiates this project from the existing learning models and the equilibrium models of economy. In an earlier collaboration with Thomas J. Sargent, the principal investigator examined the monetary policy model of Kydland and Prescott in which the government chooses a target inflation rate based on theestimated short-term Phillips curve. A high inflation rate is realized and remains stable under various conditions. Although widely used in making monetary policy in real world, the short-term Phillips curve is a misspecified model of an economy, because it does not recognize the fact that the expectations of A private agent changes in response to the government's policy. The key finding is that any slight suspicion ofthe government about the stationarity of the underlying economy can generate an abrupt and significant drop of the target inflation rate, which the existing literature has proven stable under various conditions. Thisapparently paradoxical result can be explained through an endogenous switching mechanism of two fundamentally different dynamics within the same model: namely, the "mean dynamics" that dictates the stability of the self-confirming equilibrium, and the "escape dynamics" that pushes the economy away from the equilibrium. While the existing literature has concentrated mainly on the "mean dynamics," the proposed research project investigates the "escape dynamics," which has been overlooked but can influence the dynamics of an economy significantly. The project generalizes the endogenous switching mechanism to a wider class of learning models in order to apply this idea to important economic problems. The project examines a class of misspecified learning models that have a stable self-confirming equilibrium that is different from "true" equilibrium. This research starts by building a general theory that can be applied to a different class of models. Applications include a monetary policy known as the Taylor rule that has attracted considerable interest from policymakers as well as economists. An active Taylor rule changes the nominal interest rate more than one to one in order to control inflation. Although the active Taylor rule is known to implement the inflation target under general conditions, a recent study indicated that the same policy might unintentionally lead to a "liquidity trap" equilibrium, in which the government can no longer stimulate the economy by lowering the nominal interest rate as was the case in the recent episode of Japanese economy. The proposed approach can offer a simpler, yet more realistic, model to explain the potential limit of the active Taylor rule. The key idea is that even though the government is fully committed to a policy, a slight suspicion by the private sector about the government's commitment will prompt the agent to learn about the government's policy rule, and this learning process alone could lead the economy to the "liquidity trap."The second application uses the idea of escape dynamics to capture abrupt, and possibly recurrent departures from the normal exchange rate in financial crises while maintaining the stability of the economy. The monetary policy model is used as the foundation, while incorporating foreign investors who might have a misspecified model about the government's policy rule.

这个项目调查的代理人谁估计模型拟合观察到的数据很好，虽然他可能永远不会学习“真正的”均衡，因为他的模型是“错误的”，在这个意义上，它错过了一个或多个关键参数的环境。 即使是一个错误的模型，决策者的信念也可以自我实现，他最终可能会使用模型来选择政策。 虽然自我实现信念的概念与均衡范式下研究的理性预期密切相关，但它被称为自我确认均衡，以强调两种不同的模型相互作用以产生观察到的结果：代理人的模型，被错误指定但用于选择行动，以及产生实际数据的真实模型。 真实模型和感知模型之间的相互作用是本研究的核心部分，它使本项目区别于现有的学习模型和经济均衡模型。在早些时候与托马斯J.萨金特的合作中，首席研究员研究了基德兰德和普雷斯科特的货币政策模型，其中政府根据估计的短期菲利普斯曲线选择目标通胀率。 通货膨胀率很高，并在各种条件下保持稳定。 虽然短期菲利普斯曲线在真实的世界中被广泛用于制定货币政策，但它是一个错误的经济模型，因为它没有认识到私人代理人的预期会随着政府政策的变化而变化。 关键的发现是，任何轻微的怀疑，政府对平稳的基本经济可以产生突然和显着下降的目标通货膨胀率，现有的文献已经证明在各种条件下稳定。 这一看似矛盾的结果可以通过同一模型中两种根本不同的动态的内生转换机制来解释：即，决定自我确认均衡稳定性的“均值动态”和推动经济远离均衡的“逃逸动态”。 虽然现有的文献主要集中在“平均动态”，拟议的研究项目调查的“逃逸动态”，这一直被忽视，但可以影响经济的动态显着。该项目将内生转换机制推广到更广泛的学习模型，以便将这一想法应用于重要的经济问题。 该项目研究了一类错误指定的学习模型，这些模型具有不同于“真实”均衡的稳定的自我确认均衡。这项研究首先建立了一个通用的理论，可以应用于不同类别的模型。 应用包括被称为泰勒规则的货币政策，该规则引起了政策制定者和经济学家的极大兴趣。 主动泰勒规则以一比一的方式改变名义利率以控制通货膨胀。 虽然在一般情况下，积极泰勒规则可以实现通货膨胀目标，但最近的研究表明，同样的政策可能会无意中导致“流动性陷阱”均衡，在这种均衡中，政府不再能够像最近的日本经济那样通过降低名义利率来刺激经济。所提出的方法可以提供一个更简单，但更现实的模型来解释潜在的限制主动泰勒规则。 其核心思想是，即使政府完全致力于一项政策，私人部门对政府承诺的轻微怀疑也会促使代理人学习政府的政策规则，而这一学习过程本身就可能导致经济陷入“流动性陷阱”。“第二个应用程序使用逃逸动力学的概念来捕捉金融危机中突然的、可能是经常性的偏离正常汇率的情况，同时保持经济的稳定。 本文以货币政策模型为基础，同时考虑到外国投资者可能对政府的政策规则有错误的模型。