time-invariant regressors. This setting will allow us to make the constructive distinction
between immediate and long-run effects of a variable on the sovereign rating.
Moreover, we also use a limited dependent variable framework by estimating the
augmented-model using ordered probit and random effects ordered probit specifications.
The latter is the best procedure for panel data as it considers the existence of an
additional normally distributed cross-section error. This approach allows both to
determine the cut-off points throughout the rating scale as well as to test whether a
linear quantitative transformation of the ratings is actually more appropriate than a
possible non-linear transformation. Furthermore, we perform robustness check by
allowing for a sub-period analysis and for a differentiated high and low rating analysis.
We find that in particular six core variables have a consistent impact on sovereign
ratings. These are the level of GDP per capita, real GDP growth, the public debt level
and government effectiveness, as well as the level of external debt and external
reserves. A dummy reflecting past sovereign defaults is also found significant as well
as, in some cases, the fiscal balance and a dummy for European Union countries. It is
noteworthy that fiscal variables turn out to be more important than found in the previous
literature.
The paper is organised as follows. In Section Two we give an overview of the rating
systems and review the relevant related literature. Section Three explains our
methodological choices, specifically regarding the econometric approaches employed.
In Section Four we describe the dataset and report on the empirical analysis, notably in
terms of the estimation and prediction results. Section Five summarises the paper’s
main findings.
2. Rating systems and literature
2.1. Overview of rating systems
We use sovereign credit ratings by the three main international rating agencies,
Moody’s, Standard & Poor’s (S&P) and Fitch Ratings. Although these agencies do not
use the same qualitative codes, in general, there is a correspondence between each
agency rating level as shown in Table 1. S&P and Fitch use a similar qualitative letter