Robust structural analysis when theoretical information is scarce

动态系统中结构关系的识别需要外部信息(例如，基于理论的限制、工具变量)。为了在缺乏或很难获得这种信息的情况下支持结构分析，出现了一种基于数据的识别的可行文献，这种识别依赖于正交冲击的随机特征，例如非高斯冲击的异方差或独立性。为了检测独立分量，已经提出了几个不同的建议，例如，在参数严格性方面。Gourieroux、Monfort和Renné(2017)的伪最大似然方法特别有吸引力，因为它允许在规范错误说明的“程度”的条件下对结构关系进行稳健估计。然而，事实仍然是，分析师总是不知道实际的错误说明。在这个项目中，我们首先发展了结构参数的核最大似然估计，作为参数伪最大似然估计的替代，它保证了在更弱的条件下的稳健估计。其次，新的方法将被用来揭示从潜在冲击到可见冲击的传递过程中确定性(结构突变)和随机变化(马尔可夫转换)的范围。第三，我们建议核最大似然法作为一个概念框架来解释广义系统，在广义系统中，可观测值是由较少的(独立)冲击驱动的。第四，除了计量经济学方面的进展外，该项目还对当前宏观经济研究的重要课题做出了贡献。一方面，每一种发达的计量经济学工具都被用来更好地了解全球石油市场的冲击及其影响。另一方面，我们使用计划的计量经济学进步的集合来识别中央银行货币政策声明的信息效应。最后，整体方法将作为开发一种新型软件包的核心标准，该软件包将(几乎)实时总结各国央行公告中包含的相关信息，以揭示货币政策(包括美国和欧元区)的状况及其潜在的宏观经济影响。

The identification of structural relations in dynamic systems requires external information (e.g. theory-based restrictions, instrumental variables). To support structural analysis in cases where such information is scant or hardly available, a viable literature has emerged on data-based identification that relies on stochastic characteristics of orthogonal shocks, e.g. heteroskedasticity or independence of non-Gaussian shocks. To detect independent components, several suggestions have been made that differ, for instance, with regard to parametric rigour. The pseudo-maximum-likelihood approach of Gourieroux, Monfort and Renné (2017) is of particular appeal, as it allows robust estimation of structural relations under conditions that regulate, for instance, the “extent” of misspecification. Yet, the fact remains that actual misspecification is always unknown to the analyst. In this project we first develop kernel-maximum-likelihood estimation of structural parameters as an alternative to parametric pseudo-maximum-likelihood that promises robust estimation under much weaker conditions. Second, the new approach will be used to unravel the scope of both deterministic (structural breaks) and stochastic changes (Markov switching) in the transmission from latent shocks to observables. Third, we suggest kernel-maximum-likelihood as a conceptual framework to account for singular systems, in which observables are driven by a smaller number of (independent) shocks. Fourth, in addition to econometric advances, this project contributes to important topics of current macroeconomic research. On the one hand, each developed econometric tool is employed to achieve an improved understanding of global oil market shocks and their effects. On the other hand, we employ an ensemble of the planned econometric advances to identify information effects of monetary policy announcements of the central bank. Finally, the ensemble approach will serve as a core yardstick for the development of a novel software package that summarizes relevant information contained in central banks announcements in (almost) real-time to unravel the state of monetary policy (both in the US and the Euro-Area) and its potential macroeconomic implications.