14
fifth spatial component comprises a north-east trend as well as and a trend subsiding
from an elevation at the northern border with France in all domestic directions. The
sixth spatial component is marked by a central depression. This pattern also holds for
the eighth component actually having two additional depressions at the north-eastern
and south-eastern border. In the seventh component the central depression disappears,
while two depressions are located at the northern and western border.
6. Estimation and testing results
As indicated by Moran's I, in almost all years the spatial correlations of the residuals
from the spatial SUR model lie near-by zero (see Table 6.1). Only in one year the z val-
ue for Moran’s I exceeds the quantile z0.975 of the normal distribution that is usually
used in a two-sided test of significance.4 In contrast, the non-spatial GLS residuals are
highly correlated. Revisions of the employment statistics led to a structural break in the
late 1990s. Since April 1999 the minor employed are liable to the social security sys-
tem. Their registration has brought a noticeably upward revision of the total number of
employed. This is captured by means of time dummies.
Both coefficients of Verdoorn's law turn out to be significant with the correct sign. The
spatial SUR model implies a threshold of 1.2% for employment. This level is in line
with previous findings of Logeay (2001) and Walwei (2002), who report estimates in
the range of 1-1.5%, on the base of time-series methods. The value of 2.9% implied by
non-spatial GLS estimation overrates the threshold considerably.
The output elasticity of employment of 0.2 implies a rather weak reaction of employ-
ment to GDP growth. Factors like labour hoarding and adjustments of labour intensity
have played a prominent role in the 1990s. It should be noted, however, that the elastic-
ity has to be regarded as a result of variations occuring in the sample period (Figure
6.1). Both the intercept and the slope of the Verdoorn relation tend to vary from year to
year. In 1993 the data exhibit even a negative relationship between the growth rates of
GDP and employment for the entire regions.
4 For some difficulties with Moran’s I in space-time models see Hooper and Hewings (1981).