Tables 3 to 8 we report the probability forecast performance of an ARDL model which uses
as regressors principal components extracted from a dataset which lacks a specific group of
variables and we check whether and to what extent the probability forecast performance of
this new model is worse than the one corresponding to diffusion indices obtained by using
the whole dataset.
First, we can observe from Table 3 that the probability forecast performance associated
with an ARDL model with principal components obtained from a dataset excluding the total
size of external debt is not altered. Therefore, we conclude that the total size of debt itself
does not play an important role in explaining (out-of-sample) the EMP in each country.
However, the maturity and the geographical composition of external debt seem to play an
important role. In particular, given that a great deal of attention has been devoted to the
common lender channel, we examine whether, by removing the external borrowing either
from Europe, or from the US, or from Japan, alters substantially the probability forecast
performance of a principal components model. The common lender channel can be justified
on the grounds that when a common lender country is highly exposed to a crisis country, it is
likely to shift away from lending and to cut its lending to other countries in order to restore
its capital adequacy. As suggested by Sbracia and Zaghini (2000), common lender channel
effects can also operate through the value of collateral (e.g. stocks or government bonds)
provided by borrowers. In particular, one can consider a country that is economically open
but it has an underdeveloped bank based financial market. If this country has difficulties in
backing its funding by asset holdings in a neighbouring country, then the lender (a developed
country) will downgrade the borrower (the emerging market) and reduce the amount of
credit issued, and this will spread the crisis internationally. Therefore, when a crisis hits
the “collateral” economy, the lender will require a sounder backing of its claims. Given the
considerable poor probability forecast performance, shown from Table 6 to 8, of a model
with principal components extracted from a dataset which lacks this information, especially
on external borrowing from Europe, we can conclude that the exposure of European banks
towards East Asia has an important role in explaining (out-of-sample) currency turbulence
in East Asia. Furthermore, while the contribution of external borrowing from the US seem
to be important only for the Malaysian EMP index forecast, the contribution of external
borrowing from Japan seems to be important only for the Korean EMP index forecast.
These results confirm (on the basis of out-of-sample predictions) the importance of the
common lender channel as highlighted in the studies of Van Rijckeghem and Weder (1999)
and (2003), Kaminsky and Reinhart (2000) and (2001). From Table 4 and 5 we can infer the
important role played by short term external debt relative to long term debt when the focus
15