Monetary transmission in three central European economies: evidence from time-varying coefficient vector autoregressions

Abstract

We study the transmission of monetary policy to macroeconomic variables with structural time-varying coefficient vector autoregressions in the Czech Republic, Hungary and Poland, in comparison with that in the euro area. These three countries have experienced changes in monetary policy regimes and went through substantial structural changes, which call for the use of a time-varying parameter analysis. Our results indicate that the impact on output of a monetary shock changed over time. At the point of the last observation of our sample, the fourth quarter of 2011, among the three countries, monetary policy was most powerful in Poland and not much less strong than the transmission in the euro area. We discuss various factors that can contribute to differences in monetary transmission, such as financial structure, labour market rigidities, industry composition, exchange rate regime, credibility of monetary policy and trade openness.

Appendix: setting initial conditions

Since the time-varying parameters are assumed to follow random walks, they do not have means that could be used as initial conditions. For this reason we estimated a fixed parameter VAR, equation by equation, with OLS for the first sixteen (effective) observations and used these estimates to form initial conditions for the time varying parameters and their covariance matrix, which is assumed to be block diagonal. For example, when the lag length is 1 and an intercept is not included, the first equation is:

$$ y_{1,t} = \phi_{11}^{(1)} y_{1,t - 1} + \phi_{12}^{(1)} y_{2,t - 1} + \phi_{13}^{(1)} y_{3,t - 1} + \phi_{14}^{(1)} y_{4,t - 1} + v_{1,t} $$

Let $\hat{\Upphi }_{1}$ denote the estimated coefficient vector, that is, $\hat{\Upphi }_{1} = \left[ {\begin{array}{*{20}c} {\hat{\phi }_{11}^{(1)} } & {\hat{\phi }_{12}^{(1)} } & {\hat{\phi }_{13}^{(1)} } & {\hat{\phi }_{14}^{(1)} } \\ \end{array} } \right]^{\prime }$ and $\hat{\Upomega }_{1}$ the OLS covariance matrix of this estimation. Let us denote the parameter estimates and their covariance matrices similarly (with different subscripts) for the other three equations.

a ₀, the initial condition for the time-varying parameters is set to:

$$ a_{0} = \left[ {\begin{array}{*{20}c} {\hat{\Upphi }_{1}^{\prime } } & {\hat{\Upphi }_{2}^{\prime } } & {\hat{\Upphi }_{3}^{\prime } } & {\hat{\Upphi }_{4}^{\prime } } \\ \end{array} } \right]^{\prime } . $$

P ₀, the initial condition for the covariance matrix of a ₀ is set equal to the block diagonal matrix formed from the OLS covariance matrices of the four equations:

$$ P_{0} = \left[ {\begin{array}{*{20}c} {\hat{\Upomega }_{1} } & 0 & 0 & 0 \\ 0 & {\hat{\Upomega }_{2} } & 0 & 0 \\ 0 & 0 & {\hat{\Upomega }_{3} } & 0 \\ 0 & 0 & 0 & {\hat{\Upomega }_{4} } \\ \end{array} } \right], $$

where 0 indicates an appropriately sized null-matrix, e.g a 4 × 4 matrix when the lag length is 1 and an intercept is not included.