Business Cycle Accounting with Non-linear Solution Methods

该项目的主要目的是评估线性和非线性求解方法的计算效率和精度之间的权衡。我们将分析2019冠状病毒病危机，因为经济波动很大，这将导致线性方法的近似误差很大。我们采用了Chari，Kehoe和McGravity（2007）的商业周期核算框架，因为它是将经济波动分解为代表性来源的有用方法。然而，由于计算负担，用非线性方法求解该模型具有挑战性。因此，本课题不仅对经济危机的根源，而且对计算效率和精度的权衡，都将提供新的见解。第一年，为了设计我们的经济周期核算方法，我们回顾了有关解决方法的文献。特别是，合作研究者审查了有限元法和参数化期望算法。主要研究者审查了人工神经网络的机器学习文献。我们在概念层面上共同设计了人工神经网络经济周期核算的非线性求解方法。

The main aim of this project is to assess the tradeoff between computational efficiency and accuracy across linear and non-linear solution methods. We will analyze the Covid-19 crisis because economic fluctuations are large, which will lead to significant approximation errors in the linear method. We apply the Business Cycle Accounting framework of Chari, Kehoe and McGrattan (2007) because it is a useful method to decomposes economic fluctuations into representative sources. However, it is challenging to solve this model with non-linear methods due to the computational burden. Therefore, this project will provide new insights not only on the sources of the latest economic crisis, but also tradeoff between computational efficiency and accuracy.In the first year, we reviewed the literature on solution methods to design our Business Cycle Accounting method. In particular, the Co-Investigator reviewed the finite element method and the parameterized expectation algorithm. The Principal Investigator reviewed the machine learning literature on Artificial Neural Networks. We have jointly designed the non-linear solution method for Business Cycle Accounting with Artificial Neural Networks at a conceptual level.