Subjectivity in sovereign credit ratings

We decompose sovereign credit ratings into an objective and a subjective component and we identify the determinants and relative importance of each component. The determinants of the objective part of credit ratings are long known and consist of macroeconomic and political variables. The contribution of this paper lies in the determination of the subjective part of credit ratings. The subjective component of credit ratings is driven by familiarity indicators (that lead to a foreign bias in credit ratings) and a measure for the lobbying power of a country. A second contribution of our work is that we measure the economic costs of the subjective part of the credit rating as the difference between the sovereign spread associated with the actual rating and the spread for a purely objective rating. We find that these economic costs are marginal: although the subjective component in credit ratings may lead to an upward or downward adjustment in the objective rating, there is no general relationship between the subjective component in credit ratings and sovereign spreads. The third contribution of our work is methodological: we compare the traditional way of modeling credit ratings, the linear panel regression, with a more innovative method based on artificial intelligence. The random forest machine learning technique leads to a signicant reduction in prediction error both in sample and out-of-sample. Finally, we show that the size of the subjective component in credit ratings varies across rating notches and over time. We find that the regulatory changes in both US and EU law has lead to sovereign credit ratings that rely less on subjective factors than in the pre-crisis period.