Johansen’s cointegration tests confirm that both inflation and inflation uncertainty can
be treated as stationary in levels (see Table 3). Therefore the remainder of the analysis
is carried out under this assumption, and a VAR in levels was estimated.
Standard selection criteria suggested choosing lag length 14 for the bivariate VAR in
levels. The deterministic component was specified to include a constant and a trend,
as well as seasonal dummies and a shift dummy in 1998m12 to capture the
introduction of the euro. Standard diagnostic tests indicate that the model is
statistically adequate. Granger-causality tests imply uni-directional causality running
from steady-state inflation uncertainty to inflation at the 5% confidence level,
consistently with the Cukierman-Meltzer model, and bi-directional causality at the
10% level (see Table 4).
Table 4. Test for Granger-causality, dummy 1998m12
H0: “infl ss var” does not Granger-cause “infl” - whole sample_____________________ | |
__________Test statistic l = 1.8757__________ |
pval-F( l; 14, 342) = 0.0279 |
H0: “infl” does not Granger-cause “infl ss var” - whole sample_____________________ | |
__________Test statistic l = 1.6509__________ |
pval-F( l; 14, 342) = 0.0644 |
The model was also estimated including a sequential intercept shift dummy in order to
test endogenously for possible structural breaks. Chart 10 shows the sequential t-value
of the corresponding coefficient in the equation for steady-state inflation uncertainty.
As can be seen, a sizeable downward shift is apparent at the end of 2000. This is
consistent with the earlier finding of declining and more stable steady-state inflation
uncertainty in the euro years.
Chart 10. Sequential t-value of the dummy coefficient
14