is supposed to decline, if the Okun coefficient rises in absolute value. In addition, a re-
gressive trend growth rate can contribute to a reduction. Unemployment dynamics de-
pend on the threshold according to
(2.11a) (ut -u*) = δ1(yt - yU,GAP) ,
(2.11b) ∆ut =γ1(yt - yU,FD),
that is, unemployment will remain at its previous level if actual growth is just as high as
the threshold for unemployment. For a drop of the unemployment rate output growth
must exceed this level.
3. Spatially filtering and spatial SUR models
As the thresholds for employment and unemployment are estimated with regional data,
dependencies between the cross-sections have to be taken into account. They may stem
from common or idiosyncratic (region specific) shocks, which may generate spillovers
among the cross-sections. Eventually, variables are spatially autocorrelated over the
entities, and the particular pattern can bias the results, see Anselin (1988, pp. 58).
Therefore, appropriate filters have to be employed in order to separate the spatial and
non-spatial components of the series that enter the regression model.
At present, two approaches are available to identify spatial effects in the data, see Getis
and Griffith (2002) for a recent survey. Getis and Ord (1992) have proposed a spatial
distance statistic. It requires that all variables are positive and can be measured from
natural origins. These conditions are not met in this study, as growth rates and changes
of variables are involved. Thus, the eigenfunction decomposition approach suggested by
Griffith (1996, 2000) is preferred. Here, filtering relies on a decomposition of Moran’s I
(MI) statistic
(3.1) MI =
as a measure
x ' Wx
x' x
of the global spatial autocorrelation structure for a given variable. In par-
ticular, x holds the n observations of the variable under consideration, measured in de-
viations from the mean. W is an nxn-matrix of spatial weights, where the elements of
each row sum up to 1, and n the number of regions, see Anselin (1988, pp.16) for a dis-