erc/metu
INTERNATIONAL CONFERENCE IN ECONOMICS  IV
September 13-16, 2000, Ankara

The Effect of Deterministic Trend on Model Selection Criteria in the Nonstationary VAR Process

Nezir Köse (Gazi University)
Nuri Uçar (Bilkent University)

Abstract

Alternative testing procedures have recently proposed in the literature for economic hypotheses that can be expressed as coefficient restrictions of a nonstationary VAR. Although the determination of the lag structure is usually not the final objective of a VAR modelling, it might have a great impact on subsequent inferences about the persistence of shocks, impulse responses, variance decomposition, forecasting, etc.

The objective in this study is to compare the finite sample performance of some model selection criteria for the nonstationary VAR processes through a Monte Carlo study. We also analyse the effect of deterministic time trend on model selection criteria. Data generating processes (DGP) for the nonstationary VAR are distinguished into two classes as trend stationary (TS) and difference stationary (DS). When the trend variable is omitted from TS models, the criteria have the tendency to overestimate the correct lag order. This observation can be attributed to the omission of a relevant variable on the model selection criteria. The simulation results obtained from the DS model with the inclusion of time trend represents that all the criteria selects the true lag length. We can conclude that adding the trend to the DS model does not have any influence on the determination of true lag. Moreover, Schwarz criterion seems to be a reasoable choice as a lag selection criterion for the correctly specified VAR model generated by either inclusion of trend variable and unit root.

Economic Research Center
Middle East Technical University
06531 Ankara Turkey
Phone: + 90 312 210 3044, 210 2003
Fax: +90 312 210 1244
e-mail: metuerc@metu.edu.tr