The Damaging Bias of Sovereign Ratings

Abstract

Sovereign credit ratings guide over USD 50 trillion in outstanding sovereign debt and set the ceiling for most corporate credit ratings. The credit rating agencies assign ratings based on a combination of objective fundamentals and the subjective judgment of their in‐house rating committees. In this paper we decompose the sovereign ratings of the ‘Big Three’ rating agencies into an ‘objective’ component (the fitted value from an OLS regression of ratings on 10 explanatory variables) and a ‘subjective’ component (the corresponding residuals) using data for advanced and emerging economies over the period 1996–2013. We then examine the explanatory power of the two components for predicting sovereign default. Our main finding is that, while the ‘objective’ component has explanatory power to predict defaults both in the short and long‐term, the ‘subjective’ component does not help to predict defaults of a horizon of 1 year or more. In particular, analysing the probability of default within 3 years, we find the ‘subjective’ component is biasing default predictions in the wrong direction with—at times—dramatic consequences. This is the ‘damaging bias’ of sovereign ratings. The biggest casualty of this was the Eurozone periphery, which was downgraded far too heavily during the 2009–2011 sovereign debt crisis as the rating committees repeatedly overruled the signal coming from fundamentals. In light of our findings, we suggest that credit rating agencies should be stripped of their regulatory powers and these transferred to an international body. Failing that, the ratings agencies should be forced to substantially increase transparency, including publishing a separate breakdown of the ‘objective’ and ‘subjective’ components of ratings, the minutes of the rating committees, and the voting records.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 50 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.