22
From Table 4.1, it can be seen that all explanatory variables of the optimal degree, except
NAIRU, have the expected sign.19) Only one explanatory variable (SLOPE) is significant
at a 90% confidence level. The other explanatory variables have relatively low t-values.
Nevertheless, we have calculated the optimal degree on the basis of the ultimate determi-
nants for each country and, after normalization of the optimal degree and the legal indices,
the average difference between these variables. Positive average differences - of 0.20 or
higher - are found for Germany and Switzerland, implying that the legal degree of central
bank independence exceeds the optimal degree and that the legal degree should be
decreased. Negative average differences - of 0.20 or lower - are observed for Australia,
Norway, Sweden and the United Kingdom, meaning that the optimal degree exceeds the
legal degree and that the legal degree should be increased.
For the other countries - Austria, Belgium, Canada, Denmark, Finland, France, Ireland,
Italy, Japan, the Netherlands, New Zealand, Spain and the United States - the average
differences are relatively small, indicating that there is no reason to adjust the central bank
law in these countries from the perspective of the ultimate determinants. In some countries
- notably France and Spain - the central bank has, recently, been made more independent
from government which can be explained by another argument: a prerequisite for entering
the third phase of Economic and Monetary Union in Europe is, among others, the
independence of the national central banks of the participating countries.
The relatively low t-values for the explanatory variables in Table 4.1 could, probably, be
attributed to the many severe restrictions imposed on the model by LISREL and the two
additional restrictions made by us (λy3 = 1 and Θγ is diagonal) to identify all parameters
of the model. Relaxing some of these restrictions might improve the t-values of the
explanatory variables.20)
Table 4.2 gives the empirical results, if we relax only two restrictions on the covariances,
for the sample period 1960-1993 (for NAIRU: 1960-1988). First, the restriction on the
covariance of [γ2, γ3] between the GMTT- and ES-index is eliminated. This implies that
the disturbances of these indices may be correlated. Second, the restriction on the
covariance of [γ2, ζ] between the GMTT-index and the regression equation - equation
(4.2) with ξ =x-islifted. This means that the disturbances between the GMTT-index and
the regression equation can be correlated. All other restrictions on the model remain
imposed.
19) The negatieve coefficient for the variable NAIRU may, however, be explained by the existence of reverse
causation: a high degree of central bank independence leads, apparently, to a low NAIRU in the long
run. Moreover, there is empirical evidence for an increase of NAIRU in the OECD countries during the
last decades.
20) See in this respect: Aigner, Hsiao, Kapteyn and Wansbeek (1984, p. 1371). The relaxing of restrictions
could imply that, although the latent variables method is still used, the assumptions of LISREL are not
valid anymore.