16 Ian Babetskii
The ERM-II participants do not demonstrate higher wage flexibility compared to the CE-4 group.
The estimates are insignificant for most of the period. Similar results apply to the EMU-3.
Overall, the time-varying estimates do not support the view that wage flexibility is higher in
countries participating in the ERM-II or in the EMU members compared to the CE-4. Also, for the
few cases where estimates are significant, there is evidence of a decrease rather than an
improvement in flexibility in recent years.
5.3 Cointegration/Error Correction
First we determine the optimal lag length in the vector autoregressive representation (6). Using
the Akaike and Schwarz information criteria, the number of lags is mostly found to be one, and in
a few cases two or three. To preserve homogeneity and parsimony, we set the number of lags to
one for all countries10. Next, we test for the presence of a long-run relationship between real
wages, productivity, and the unemployment rate. According to the Johansen cointegration test,
there is one stable vector (significant at the 10% level) in the case of Estonia, the Czech Republic,
Hungary, and Lithuania. No long-run relationships between the three variables are detected for the
EMU-3 members. The absence of cointegration suggests that there is disequilibrium on the
aggregate labor market, or that there are structural changes. The cases of cointegration are
reported in Table 5.
The cointegrating vectors are normalized so that the coefficient on real wages is equal to one.
Inspection of the long-term relationships indicates that real wages closely follow productivity in
Estonia. In the Czech Republic and Hungary, real wages grow faster than productivity, while in
Lithuania productivity growth is stronger than that of real wages. Notice that the long-run
coefficients cannot be interpreted as elasticities in the strict sense, since each of the coefficients
incorporates the effect of shocks to all variables (see Lutkepohl, 1994). The long-run coefficient
on unemployment does not have any particular meaning, since equation (6) is not a structural
representation and unemployment may enter with either positive or negative sign depending on
the cycle.
Next, the error correction representation provides information about the adjustment channels. Two
contrasting examples are the Czech and Estonian cases. In Estonia, adjustment to long-run
equilibrium occurs via real wages and productivity. The unemployment channel is insignificant.
On the other hand, the real wage and productivity channels do not play a significant role in the
Czech case, and it is unemployment which closes the gap. In Hungary and Lithuania, real wages
react to deviations from equilibrium, while the productivity and unemployment channels are
insignificant. The speed of short-run wage adjustment, measured by the coefficient on the error
correction term, is the highest in Lithuania (-0.394), followed by Estonia (-0.347) and Hungary
(-0.271). Negative values mean a return towards equilibrium, i.e., the error correction mechanism
is at work. Summarizing, there is an indication of a higher magnitude of short-run wage
adjustment in Lithuania. However, real wages do not react to changes in unemployment in any of
the four countries considered.
10 Estimations with two and three lags produce little difference in the long-run elasticities, but the error
correction part becomes less clear-cut because of a substantial reduction in the number of freedoms. The results
are available upon request.