Forecasting Using Non-linear Multivariate Time Series Models with Bayesian Stochastic Search Variable Selection Method and its Application to Macroeconmics

2020 Fiscal Year Final Research Report

• Project/Area Number: 17K03661
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Economic statistics
• Research Institution: University of the Ryukyus
• Principal Investigator: Katsuhiro Sugita 琉球大学, 国際地域創造学部, 教授 (50377058)
• Project Period (FY): 2017-04-01 – 2021-03-31
• Keywords: 計量経済学 / 時系列分析 / 多変量時系列 / ベイズ統計学 / MCMC

Outline of Final Research Achievements

A VAR (Vector Auto Regression) model is often used for empirical studies for macroeconomic analysis or forecasting macroeconomic variables. However, one of the problem of using a VAR model is that VAR model often contains too many variables of which are insignificant. In this research, I examine the forecasting performance of Bayesian SSVS (Stochastic search variable selection) method to remove insignificant variables in the model for model selection.I showed that the SSVS method improve the performance of the time series forecasting by using aritificially generated stationary or non-stationary data.

Free Research Field

計量経済学

Academic Significance and Societal Importance of the Research Achievements

研究成果の学術的意義として、多変量時系列モデルにおいてより高い予測精度をもたらすSSVS法の利便性を示したことにある。このメソッドは汎用性があり多くのモデルに応用できるので、これからの時系列予測や計量分析に役立つという意義がある。