Causality in EU macroeconomic variables

ABSTRACT

This article investigates the interdependence of macroeconomics, financial and other variables for European Union countries using a multivariate vector autoregressive (VAR) approach for quarterly data. The VAR analysis is applied to all bivariate cases, and the best fitted models are selected in order to conduct Granger causality testing and impulse response functions. Contrary to the existing literature, this study reveals evidence of a unilateral direction between several cases and ambiguous results regarding Impulse response functions analysis.

KEYWORDS:

• Vector autoregressive analysis
• causal relations
• impulse response functions
• macro and financial variables

JEL CLASSIFICATION:

• C22
• C59
• G10
• E60

Acknowledgement

We thank Dr. Michalis Polemis from University of Piraeus for his useful comments and suggestions.

Notes

¹ The 11 Countries and members of EU that are used in this study are Austria, Belgium, Netherlands (Dutch), Finland, France, Germany, Greece, Ireland, Italy, Portugal and Spain.

² The data for macroeconomic variables is obtained from Eurostat, the Yahoo Finance data base for data on European stock indices, the US Energy Information Administration for Brent oil spot prices and data on euribor from the European Central Bank.

³ The use of ln(P_t/P_t−1) as a proxy variable of returns is not considered since the data set consists of quarterly observations, and therefore, it would not have been an appropriate proxy variable.

⁴ The ADF test is implemented by the proposed methodology of Agiakloglou and Newbold (1992).

⁵ In most cases, the same best-fitted models are also selected, by the Schwarz (SC) criterion, since it selects models with small number of parameters. In cases where the LR test was not applicable, the best-fitted model was selected only by the SC criterion; Table 1 reports only the bivariate cases that produced statistically significant results for VAR models and were supported by most countries.

⁶ Table 1 also reports Vector Error Correction (VEC) models for some cases that VAR models were not appropriate to estimate. VEC models, which are different representations of VAR models, are used in cases where two nonstationary processes are co-integrated to capture the long run effect, see Johansen’s see Johansen and Juselius (1990).