Page 100 - Invited Paper Session (IPS) - Volume 2

IPS184 Celestino G. et al.
code and the suffixes 1 or 2 for the two consecutive periods. They are shown
for each cluster together with two 4xn matrices that contain the standardized
5 4-dimensional vectors with the medians of sector net lending/ net borrowing
6
for n countries included in the cluster for each of the two time periods. For
each cluster, the corresponding “neuron weight vector” is provided, i.e. the
closest node in the “trained” neural network, which can be interpreted as the
(standardised) net lending/ net borrowing configuration of a notional
representative country for that cluster.
Figure 1. SOM country clustering on the basis of domestic sector net lending/
net borrowing

Cluster Neuron Countries

1999-2008: BE_1, FI_1, IE_1, LU_1, NL_1

0.23 1.62 0.29 0.32 1.75
11 0.61 [ 0.10 −0.29 0.62 −0.28 0.22 ]
1 [ 12 ] = [ 0.01 ] 0.85 2.43 1.42 1.67 0.73
13
1.19
0.50 −0.74 −1.96 −0.44 −0.66
1 −0.51 0.12 1.08 0.86 −0.06
2009-2018: −0.15 −0.27 −0.42 0.09
EE_2, FI_2, [ 0.89 0.14 0.98 1.39 ]
DE_2, LU_2 0.29 −0.82 0.72 −024
0.11 0.31 0.36 0.06
1999-2008: −0.26 0.01 −0.11 0.07
[ 0.14 −0.13 −0.40 −0.38 ]
AT_1, FR_1,

0.32 DE_1, IT_1 0.21 0.81 0.18
0.61
2 [ 0.08 ] 2009-2018: AT_2, BE_2, FR_2, IE_2, IT_ 2, LV_2, LT_2, NL_2,
−0.49 PT_2, SK_2, SI_2, ES_2
0.24

−1.63 −1.00 −1.17 −1.15
1999-2008: 2.54 −0.05 −1.14 −0.06
−1.26 MT_1, PT_1, [ −1.03 −0.79 −0.98 0.01 ]
3 [ 0.46 ] SK_1, SI_1 0.61 0.01 0.66 0.61
−0.62
0.50 2009-2018: CY_2, −0.66 −1.97
0.34
1.30
MT_2 [ −0.76 −0.14 ]
−0.51 1.58

5 i.e. expressed as differences from the mean divided by the standard deviation.
Sectors codes are those of the SDMX standard, i.e. S11 NFCs, S12 financial corporations, S13
6
general government and S1M households and non-profit institutions serving households.
87 | I S I W S C 2 0 1 9