Causal Structure Analysis of Economic Time-Series Data : Method and Application

这一时期的研究成果包括经济时间序列因果度量的统计推断方法的理论发展及其计量经济学应用。研究者已经给出了平稳时间序列的因果度量理论。本文的研究重点是单向效应度量，给出了Wald检验和基于Johansen非平稳协整模型的置信集构建方法，并开发了相应的计算机算法。该方法应用于日本宏观经济的季度数据集，即近20年的GDP、隔夜拆借利率、出口和进口。通过测试和评估这些时间序列的各种组合的单向因果强度，日本宏观经济结构的特征是相互因果互动。本研究的推论方法的一个优点是，它不仅可以检验，而且可以估计因果强度，提供置信区域，而传统的格兰杰检验只是检验单向效应的存在。具体而言，本方法能够通过Wald检验构建多重单向效应测量的同时置信区域。本研究的另一个贡献是发展了在存在第三系列的情况下消除单向效应的新理论，并为此开发了计算机算法。消去法的优点在于避开了Granger和Geveke等传统方法的难点。最后，为此研究开发了一种改进的有理谱密度矩阵分解算法。

The results of this research period consist of the theoretical development of statistical inference method for the causal measure between economic time-series and its econometric application.The researcher has already given a theory of causal measure for stationary time-series. The present research, in particular, is focused on the one-way effect measure and gives the Wald test and the confidence set construction method based on the Johansen nonstationary co-integration model, developing also the accompanying computer algorithms. The method is applied to the quarterly data set of Japanese macroeconomy, namely the GDP, the call rates, exports and imports over the recent 20 years. By testing and evaluating the one-way causal strength of various combinations of those time-seires, the Japanese macroeconomic structure is characterized in terms of mutual causal interaction. A merit of the inferential method of this research is that it enables not only testing but estimating causal strength, providing the confidence region, whereas the conventional Granger tests merely test the presence of one-way effect. Specifically, the present approach enables constructing simultaneous confidence-regions of plural one-way effect measures by means of the Wald test. Another contribution of this research is the development of a new theory of eliminating the one-way effect of a third series in case of the presence and the development of an computer algorithms for it. The merit of the elimination method is in that it can evade the difficulties of the conventional methods by Granger and Geveke. Lastly for this purpose, the research developed an improved factorization algorithm of rational spectral density matrices.