significance level. The change may suggest that the function of the current account as
an indicator for foreigners’ willingness to cover the current account gap has declined.
Finally, for Moody’s the increase in value and significance of the coefficient on external
reserves may point to a higher importance attached to this variable after the Asian crisis.
Taken together with the reduced importance of the current account, this could suggest a
move towards a broader view on foreign financial relationships, which includes capital
flows in addition to the current account movements.
Looking at the individual agencies, for Fitch coefficient values remain remarkably
stable over the sub-periods. An exception is the negative (though insignificant) value for
GDP growth in the early sub-sample. For S&P, sign reversals between sub-periods
occur for the explanatory variables unemployment, government effectiveness and
external debt, but there are no significant coefficient estimates with opposing signs. For
Moody’s, the models point to a sign reversal for the insignificant estimate of the
coefficient on inflation.
4.2.3. Differentiation across ratings levels
As a further test of the robustness of the results derived above, the sample was split into
two groups according to the ratings level: regressions were run separately for high-rated
countries with grades BBB+ and below and those above this grade. The choice of the
threshold reflects practical considerations. While market participants generally divide
bond issuers into investment-grade and non-investment grade at the threshold of BBB-,
this threshold would result in a relatively small number of observations for low ratings
making inference problematic. From the estimation of the ratigns above BBB+, we
removed the variable external debt because there were too few observations of non-
industrial countries.
The results for the separate regressions according to ratings levels confirm the overall
results from the full sample (see Tables 10-12). Looking at the random effects
estimation for each agency, the variables that were found to be significant across
agencies in the full sample also show up consistently with the correct sign in the
individual regression models for high and low ratings. Most of the coefficients are
statistically significant. Notably, the importance of average external reserves appears to
20