Computational social choice theory, convex analysis, memetic algorithms, econometric modeling and data visualization based on cutting-edge research papers. Numerical computations were carried out using the Matlab software.
Empirical base of research
Long-term credit ratings assigned by seven credit rating agencies (CRAs) to Russian banks in national scales over the period from 2010 to 2016, data on defaults of Russian banks. Price data on Russian government and corporate bonds.
Results of research
We develop an algorithm of studying and assessing a quality of arbitrary aggregate rating in order to determine if aggregate rating is a valid credit scoring tool and what limits of its applicability are. The algorithm is based on comparison of discriminatory power and ordering agreement among aggregate rating and existing models of credit risk. The algorithm also includes building a predictive model in order to map aggregate rating grades to companies’ individual characteristics and study if such model can describe aggregate rating fairly well and be used for extrapolation of aggregate rating to non-rated companies without significant loss in discriminatory power and/or ordering agreement. Developed algorithm is applied to aggregate rating of Russian banks constructed as a Kemeny median-based consensus of ratings assigned by seven credit rating agencies during the period 2010-2016. It is shown that such aggregate rating demonstrates practical quality on its own domain, but can be hardly extrapolated to the set of non-rated banks using econometric model. We also study if rating-like indicator can be constructed from bond credit spreads. Using different spread conventions and methods of building risk-free yield curve, we show that such indicator can be obtained for Russian market and it is practically valid. Obtained results can be used for the purposes of credit analysis, mapping of credit ratings and benchmarking IRB models.