Credit Rating Methodology: Lifting the lid to the black box: Assessing the need of industry-specific models that can replicate credit ratings assigned by Moody’s

Jonathan Engvall

Student thesis: Master thesis


The methodology of credit rating agencies is somewhat secretive and can to a large extent be likened to a black box. The rating agencies claim that any attempt to replicate their ratings is doomed to fail. Previous research suggests the opposite by proving that a large part of a rating can be replicated by financial data. These studies have lifted the lid to the black box and created a better understanding of what underlies any given rating. However, no study has assessed whether, how, and why a replicating model differs depending on the industry the data sample is based upon. This thesis takes a first step in this direction. By means of multivariate logistic regression, two industry-specific models are created: one that is based on a sample of 147 issuers in the oil & gas industry and another that is based on 78 issuers in the consumer products industry. A third model that incorporates the whole data sample is thereafter created to assess the predictive power of the industry-specific models. The industrial differences are analyzed in light of Porter’s Five Forces. Results reveal that certain industrial differences must be accounted for, while fundamental factors that impact the ability to meet debt payments are similar across both industries. Accounting for all rating categories, the consumer products model is able to predict 53% of the rated firms (oil & gas 41%). Accounting for two rating categories, investment grade and speculative grade, the consumer products model is able to predict 85% of the rated firms (oil & gas 74%). The concordance rates in both models are far from acceptable for any practical usage of the models. However, the ability to predict credit ratings is higher in the industry-specific models than in the model that incorporates the whole data sample. Moreover, the variables in the final rating models are statistically significant, and they differ between the models. For that reason it is concluded that there is a need for industry-specific credit rating models.

EducationsMSc in Finance and Strategic Management, (Graduate Programme) Final Thesis
Publication date2015
Number of pages85