rise for low ratings in the models for Fitch and S&P. In the cases of statistical
significance external reserves always have a positive impact on the ratings.
[Insert Tables 10, 11 and 12 here]
The explanatory power of the individual regressions is somewhat lower than that found
for the full sample as well as for the sub-periods. This reflects the fact that splitting the
sample in this way reduces the number of rating categories for each estimation, so that,
with a discrete dependent variable, estimated rating errors become relatively larger.
Beyond the core variables, the results for Moody’s and S&P suggest a significant
difference in the importance of inflation for high and low ratings, respectively. For both
agencies the (significant) coefficient on inflation as an average and the deviation from it
is much higher for high ratings than for low ratings. (For Fitch, this finding is supported
by the pooled OLS and the fixed effects estimations, but not for the random effects
specification.) This suggests that for high rated countries inflation has a much more
important impact on the rating. A possible explanation is that for this set of countries
price level stability may be taken as an indicator for sound economic and in particular
monetary policies which support the long-run sustainability of government finances.
Turning to differences across agencies, the results point to a relatively high level of
consistency in the approach to high and low ratings for Fitch and Moody’s. For these
two agencies signs and (mostly) significance levels of coefficient are generally
consistent for high and low ratings. A somewhat higher degree of variation in this
regard can be observed for S&P where the sign of the estimated impact of some
variables switches between high and low ratings, although in most instances the
comparison involves statistically insignificant coefficients. Additionally, one notices
also a higher R-square for low ratings for S&P and Fitch.
4.3. Ordered probit results
In view of the discussion of econometric issues above, ordered probit models should
give additional insight into the determinants of sovereign ratings. In particular, this
method allows to relax the rigid assumption on the shape of the ratings schedule.
21