Beyond this set of core variables, the agencies appear to employ a limited number of
additional variables. For Fitch the analysis finds the smallest set of additional variables,
comprising government effectiveness as a deviation from the average and foreign
currency reserves also as the short-run deviation. By contrast, the analysis finds more
significant explanatory variables for Moody’s and Standard and Poor’s, with a large
degree of homogeneity between these two agencies. In particular, on the real side
inflation is found to have a significantly negative impact. In the fiscal area, the average
debt level exerts an additional negative impact on the ratings level, whereas the fiscal
balance has a strong positive impact. With regard to the external sector, the current
account balance has a negative impact.
The findings regarding the current account effect may appear surprising as it suggests
that countries with high current account surpluses would tend to be rated lower than
otherwise equal countries without such surpluses. However, this result is quite recurrent
in the literature (Monfort and Mulder, 2000 or Eliasson, 2002). A possible explanation
is that a current account deficit could in fact serve as an indicator for the willingness of
foreigners to cover the current account gap through loans and foreign investment. In this
situation, a higher current account deficit would be associated with either higher credit-
worthiness or good economic prospects of the economy and consequently a higher
sovereign rating.
Finally, the impact of the unemployment variables appears not entirely clear cut. While
the average level of unemployment is found to have a significant negative impact on the
rating by Moody’s, the short-run deviation from the average enters positively and
significantly in the S&P model. Structural reforms that raise unemployment in the short
run but improve fiscal sustainability in the long run could provide an explanation for
this latter finding, but further research would be necessary to validate this hypothesis.
One can also assess how successfull and important our specification is. First, most of
the time averages of the explanatory variables are significant, which proves that if we
did not include them we would be mispecifying the model.3 Second, the models pass the
Hausman test, which sugested that the problem was entirely corrected. Furthermore, if
3 This is in fact the cause why without including time-averages the models would not pass the Hausman
test.
18