EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Statistical inference of non-Gaussian structural vector autoregressive (VAR) models Authors:  Koichi Maekawa - Hiroshima University of Economics (Japan)
Gigih Fitrianto - Hiroshima University of Economics (Japan) [presenting]
Abstract: Statistical inference in structural vector autoregressive (SVAR) models under non-Gaussian error is considered. Recent studies show that the non-Gaussian errors play an important role to identify a model and, they are useful to detect causal order of variables by using independent component analysis. This approach is gradually spreading in macro, micro, and financial econometrics. We conduct Monte Carlo experiments to see the performance of estimation methods and of testing long-run and short-run restrictions in a small SVAR model with t-distributed errors under acyclic and non-acyclic contemporaneous causal structure. The experiments show that performance of the test naturally depends on the degrees of freedom of t distribution and sample size. Furthermore, we apply this model to Japanese macroeconomic data to see whether Japans quantitative easing financial policy is effective or not. As a result, we found that SVAR model is potentially useful to detect causal order among macroeconomic variables.