Analysis of Panel Vector Error Correction Models Using Maximum Likelihood, the Bootstrap, and Canonical-Correlation Estimators

Abstract

In this paper, we examine the use of Box-Tiao*s (1977) canonical correlation method as an alternative to likelihood-based inferences for vector error-correction models. It is now well-known that testing of cointegration ranks based on Johansen*s (1995) ML-based method suffers from severe small sample size distortions. Furthermore, the distributions of empirical economic and financial time series tend to display fat tails, heteroskedasticity and skewness that are inconsistent with the usual distributional assumptions of likelihood-based approach. The testing statistic based on Box-Tiao*s canonical correlations shows promise as an alternative to Johansen*s ML-based approach for testing of cointegration rank in VECM models.