Aggregate Wage Flexibility in Selected New EU Member States r7
One popular econometric tool for the estimation of time-varying parameters is the Kalman filter.
Note that the focus on short-run adjustment (thus ignoring long-run relationships), multiple
causalities among the variables, and endogeneity bias (especially in the panel estimates) could be
serious drawbacks of the above methods.
(iii) The third method, the Johansen cointegration and vector error correction approach,
explicitly accounts for non-stationarity of the series and incorporates both short-run and long-run
dynamics4. The method identifies whether there exists such a linear combination of non-stationary
variables which turns out to be stable over time.
c0 +cwwrt +cqqt +cuut =εt (5)
where wrt is the aggregate real wage (CPI-deflated), qt is average productivity, and ut is the
unemployment rate.
Moreover, the cointegration method reflects potential multiple causalities between the variables.
The vector error correction mechanism allows joint determination of real wage, productivity and
unemployment adjustment:
p-1
Yt =(m0+m1t+(1+αβ')Yt-1)-∑Φi∆Yt-i+et (6)
i=1
where Yt is the vector containing real wages, productivity and unemployment, t is the time trend,
α, β, Φ are matrices, and p= 3 is the number of variables; the lag structure of the model is
determined using the information criteria and by an analysis of the residuals, which should be
white noise. A link to the exchange rate policies can be established by comparing the process of
short- and long-run wage adjustment for countries participating in the ERM-II versus those with a
flexible exchange rate arrangement (the CE-4). Selected EMU members will be used as
benchmarks. As for the drawbacks, the cointegration method is more demanding with respect to
data length (ten years of quarterly data may not be sufficient for robust testing of long-term
relationships). Also, the issue of the parameters’ stability may impose estimation problems.
Our empirical strategy can be summarized as “from simple to more complex models”. In the first
step, a country-by-country analysis is performed. Univariate time-series estimates could give an
idea of the (dis)similarities of wage/labor market adjustment. A potential drawback is that the
typical sample length (ten years of quarterly data) may not be sufficient to provide robust
econometric testing. To achieve higher power, panel estimates are typically used. However, it is
crucial to have some homogeneity in the panel. Otherwise, the interpretation of common slopes
loses economic meaning. Hence, in the second step we focus on panel estimates, conditional on
homogeneity tests. The robustness of the results is assessed by confronting the time-series and
panel estimates.
Finally, the measures of aggregate wage flexibility obtained are confronted with institutional
micro-foundations of labor market flexibility, with the objective of assessing whether any
common pattern emerges from the micro- and macro-based points of view.
4 See, for example, Enders and Dibooglu (2001), Marcellino and Mizon (2000), and Tyrvainen (1995a,b) for
applications of the cointegration method to the analysis of aggregate labor market adjustment.