for the lagged dependent variables in order to ensure stationarity. A co-integration analysis
has not been considered due to the short sample and due to the fact that the panel is
unbalanced. In addition, we have estimated the equations in levels in an AR framework and
have checked the AR-coefficient rho in this equation. It turned out to be close to 1 in the
equations for PPI consumer goods and for the HICPX components but much lower for the
other PPI components. As we want to estimate a pricing chain in a coherent framework, we
decided not to exploit the level information from the stationary PPI series (i.e. energy and
intermediate goods) and estimate all equations in first differences.
As we want to estimate homogenous coefficients across countries, we also include a variable
for trade openness to capture any differences across countries related to the exchange rate
pass-through. Trade openness is measured by the ratio of extra-area imports of each country
to real GDP. Although this variable should, in the initial equations, be multiplied with the
coefficients on the exchange rate variable to capture such heterogeneity, it can be estimated as
stand-alone variable (i.e. homogenously across countries) when taking dlogs.6 This variable
does, however, also capture any effect of globalisation so that the expected sign of the
coefficient is not clear.
Due to the huge number of variables involved in the above set-up we do not use a panel VAR
model but rather estimate single equations. The variables and lags included in the final model
for each price variable are selected using a judgemental general to specific approach. That
means that we start from a model including most of the exogenous variables and 4 lags for
each of the variables and drop progressively variables which are not statistically significant or
counter-intuitively signed. However, for the lags of the dependent variables, we keep all lags
until the longest significant to avoid too much volatility in the simulations. This procedure is
repeated until all variables are significant and correctly signed.
Once the final model specifications have been decided, the impact multiplier of exchange rate
and commodity price shocks are calculated in order to assess the pass-through on sectoral
prices at the different stages of the production chain. The impact at early stages of the
production chain is then used as input to calculate the impact at later stages of the production
chain; i.e. the impact multiplier of, say, an energy commodity price shock on PPI intermediate
goods is calculated as the direct impact of energy commodity prices on PPI intermediate
goods plus the indirect effect of energy commodity prices on PPI intermediate goods via PPI
energy, and so on.
Pt = cNEERt aoPENt ^ ln(Pt ) = ln(c) + aOPENt ln(NEERt ) ^ d ln(Pt ) = α∆OPENt + d ln(NEERt )
ECB Щ|
Working Paper Series No 1104
November 2009∣ 11