MC =
(hits * Correctrejections — falsealarms * misses)
(12)
( (hits + falsealarms) * (hits + misses)*
(correctrejections + f alsealarms) * (correctrejection + misses)
This indicator, which is bounded between -1 and +1, has the natural interpretation of
the Pearson correlation coefficient between the predicted and realised binary outcomes (see
Baldi et al., 2000, for a formal proof). Both the KS and the Matthews correlation coefficient
combine all the information contained in the contingency table above in a single value. Zero
values for the KS and the Matthews correlation coefficient correspond to the performance of
a naive predictor. Therefore, positive values for both the indicators are an indication of an
improvement, in terms of forecasting performance, upon a naive predictor.
4 Empirical analysis
4.1 The Data
As explained in section 2, given the important role of the total external debt (not only its size,
but also its geographical composition and its maturity structure) in explaining the financial
soundness of a particular economy, we need to retrieve disaggregated data on external debt.
In particular, to construct these indicators, we use the consolidated statistics on external
debt obtained from the Bank of International Settlements (BIS) on a bi-annual basis from
1983:2 to 2004:1 in millions of US dollars, for a total of 42 time series observations6 . These
data measure, on a worldwide consolidated basis, the foreign claims of banks headquartered
in the reporting area. Beyond the total external (banking) debt measures for each country,
we use the following disaggregate data on external borrowing from developed countries banks.
First, an important component of the consolidated banking statistics are the foreign
claims of BIS reporting banks vis-a-vis individual countries. As explained above, it is im-
portant to gauge information on the distribution of bank claims by nationality of bank, in
order to measure potential contagious effects operating through a common creditor chan-
nel. We concentrate on external borrowing from: Belgium, France, Germany, Italy, Japan,
Netherlands, Sweden, Switzerland, the UK, and the US. Secondly, in light of the discussion
above it is also important to have information on the external debt maturity structure. The
consolidated banking statistics provide data on the total external debt with maturity: up to
and including one year; over one year up to two years and over two years.
6 These data are also available on a quarterly basis from 1999
12