S&P). Moreover, after the Asian crisis, it seems there was a decline in the relevance of
the current account variable in the specifications for Moody’s and S&P.
Another relevant outcome ot the analysis ist that low ratings levels are more affected by
external debt and external reserves while inflation plays a bigger role for high rating
levels. On the other hand, the specifications for the Fitch ratings seem to be most
consistent over time and ratings categories. There was more variation for S&P and
Moody’s in the middle of the rating scales, which possibly points to a more quantitative
model-based approach for Fitch.
Finally, regarding the prediction analysis, the random effects model including the
estimated country effect turns out to be the method with the best fit. On average for the
three agencies, such specification correctly predicted 70 per cent of all observations and
more than 95 per cent of the predicted ratings lay within one notch. Moreover, the
models also correctly predicted between one third and one half of respectively upgrades
and downgrades. This is quite satisfactory for two reasons: first, the rating agencies also
have a forward looking behaviour that is absent from our models and second, other
qualitative factors not captured in our variables may play an important role.
Looking forward, further studies could investigate how to capture agencies’
expectations in empirical models as well as their views on qualitative variables.
Moreover, in our modelling approach we only use the government effectiveness
indicator in order to asses the impact of qualitative factors on the rating determination.
Therefore, other qualitative information could also be tentatively assessed as for
instance, socio-political factors.
30