Solution, Analysis and Control of Simultaneous, Nonlinear, Dynamic System of Equations on CYBER 205 and IBM 3090

这个项目将向量处理器(超级计算机)的能力用于解决美国经济的大型模型。超级计算机在经济分析中发挥着越来越重要的作用。在过去，为分析易操纵性而构建的模型往往包含不切实际的行为假设。超级计算机由于其速度，降低了对如此高度易操纵性的需求，并允许使用更逼真的模型。然而，试图使用这台超级计算机的研究人员正在付出代价。用于在标准大型机上求解模型的编程代码不能很容易地移植到向量处理器，并且通常没有充分利用计算速度。目前，一些经济分析师正在进行开创性的工作，为向量处理器开发算法和代码，以解决经济问题。这个项目是这项工作的一部分。从技术上讲，安藤教授比较了CyBOR 205和IBM 3090超级计算机在求解大型宏观经济模型方面的计算能力。他在过去的工作中表明，CyBOR 205在执行必要的计算方面慢得令人失望。有证据表明，IBM 3090的速度要快得多，执行相同的计算只需较少的代码修改。这个项目通过估计一个非常大的美国经济模型的随机非线性方程，比较了这两台机器的速度和分析精度。如果计算速度、设置成本和编码可移植性方面的这种差异是机器本身固有的，这项工作将为其他进行经济分析的研究人员提供非常有用的信息。

This project brings the power of the vector processor (supercomputer) to bear on solving large models of the U.S. economy. The supercomputer is playing a more and more fundamental role in economic analysis. In the past, models which were built for analytical tractability often incorporated unrealistic behavioral assumptions. The supercomputer, due to its speed, has decreased the need for such a high degree of tractability and allowed for more realistic models. However, a cost is being paid by researchers trying to use the supercomputer. The programming code used to solve models on a standard mainframe is not readily transportable to the vector processor, and typically does not make full use of the computational speed. At present some economic analysts are doing pioneering work on developing algorithms and codes for the vector processor to solve economic problems. This project is part of that work. Technically, Professor Ando compares the computational power of the CYBOR 205 and the IBM 3090 supercomputers for solving large macroeconomic models. He has shown in past work that the CYBOR 205 is disappointingly slow in carrying out the necessary computations. There is evidence that the IBM 3090 is much faster and requires less modification of code to carry out the same computations. This project compares the speed and analytical accuracy of the two machines by estimating stochastic nonlinear equations of a very large model of the U. S. economy. If such differences in computational speed, set up costs, and coding transportability are inherent in the machines themselves, this work will provide very useful information to other researchers doing economic analysis.