Dissecting the Information Value of Sovereign Credit Rating Reports

We dissect the information content of sovereign credit rating reports issued by Moody's in 62 countries for the period 2003–2013. Using the Naïve Bayesian machine learning algorithm, we classify all sentences in each report into positive and negative tone, as well as six informational categories. We find that the negative tone related to “debt dynamics” affected sovereign credit default swap (CDS) spreads the most, indicating Moody's specific skill in assessing sovereign credit risk. Moreover, we use a dozen conventional country-level default predictors to separate the tone of each report into “predicted” and “surprise” tone. We find that the negative “surprise” tone caused a bigger market reaction while the negative “predicted” tone is superior in predicting a future downgrade, reflecting different aspects of credit risk assessment. Using the 2009 Eurozone debt crisis as a natural experiment, we find that public confidence in Moody's financial risk assessment dropped after the crisis afterward. Overall, our study provides new evidence that sovereign credit rating reports contain valuable credit-related information beyond sovereign rating actions