Page 101 - Invited Paper Session (IPS) - Volume 2

IPS184 Celestino G. et al.
−1.80 −2.79 −1.21 −0.58
−1.60 1999-2008: EE_1, [ −0.08 0.16 0.05 −0.04 ]

4 [ 0.02 ] LV_1, LT_1, ES_1 1.48 0.47 0.38 0.78
0.78 −1.16 −1.38 −0.75 −0.39
−0.92
2009-2018: None

0.86
0.21
1.07 1999-2008: GR_1 [
]
−1.47
5 [ 0.39 ] −1.81
0.78 1.28
−0.92 0.57
2009-2018: GR_2
[
]
−2.50
−1.54
−0.39
−0.39 −4.98
[
]
0.02
6 [ −4.98 ] 1999-2008: CY_1

0.02
3.89
3.89
2009-2018: None

0.61
The neuron in cluster 1 ([ 0.01 ]) represents countries with high NFC net
1.19
0.51
lending (0.61; it may be recalled that values are standardised, i.e. expressed as
differences from the mean divided by the standard deviation), average
financial sector net lending (0.01), large government surplus (1.19) and
households net lending/ borrowing below average (-0.51). The countries
clustered here broadly follow that pattern (BE, FI, IE, LU, NL before the crisis),
although not exactly, presenting variations in particular in the financial sector.
In any case, the methodology followed ensures that the countries grouped
under the clusters are closer to the corresponding neuron than to any other
neuron. Note that an increase in the number of clusters would deliver lower
differences between countries and neurons, but also note that too many
clusters would result in an impractical outcome for classification purposes (in
the limit, having as many clusters as countries would deliver perfect neuron-
country matches, but this would not be of any use). As stated above, our
choice for six clusters tries to balance out the need to have sufficiently low
resolution in the clustering and at the same time to avoid grouping countries
that are too different from each other.
0.32
Cluster 2 ([ 0.08 ]) encompasses countries with relatively high NFC
−0.49
0.24
lending (0.32), although not as high as in cluster 1, and average financial sector
lending, just as cluster 1. However, the neuron shows a dramatic difference for
the fiscal situation compared with cluster 1, with government deficit above the

88 | I S I W S C 2 0 1 9