Alexander Karminsky Vladimir Sosyurko Alexander Vasilyuk National Research University Higher School of Economics Moscow, Russia Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL June, 1-3

Agenda Problem of credit rating comparison Rating agencies in Russia Multiple mapping of rating scales. Concept & Development Data gathering & Results Conclusion Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 2

Purpose and constraints of credit ratings Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 3 Ratings are the independent estimates of: financial performance of companies, banks or financial instruments issuers creditworthiness (credit risk) admission to various market products or activity Ratings are the interest for business entities and market participants, as far as for the authorities and regulating organizations (Central Banks, Ministries of Finance, Deposit Insurance Agencies, etc.) Limitations and constraints for ratings: Low number of current relevant ratings Problem of rating comparison for different rating agencies Absence of multiplicative effect from presence of competitors rating estimations Requirement for expanded use of independent rating estimations

Problem of rating comparison Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 4 Most relevant: Possibility of comparison of various agency ratings Diversified estimations with use of rating modeling Lacks: Only pair comparisons are used, scales correspondences are incompatible, displays are linear and use of econometric potential is limited No settled approaches to rating scales comparison Conclusion: required considering all restrictions on arrangement, data accessibility, etc.

Rating agencies in Russia Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL ratings at the end of 2010 Number of bank ratings agencies = 3 international & 4 national

Concept of multiple mapping Increase of comparison reliability of scales mapping by using all available statistical information (in time, on agencies, scales and structures) Development of the database that includes ratings, financial and macro- indicators Econometric exposure of the most significant publicly accessible explanatory variables that have an influence on ratings Creation of a base scale for mapping transformation of all compared agency-ratings Building up a criteria of scales correspondence, considering the peculiarities of explained component Determination of mapping parameters using the optimization procedures. Carrying out the comparison of rating scales Verification of criteria and estimation of parameters for scales conformity. Analysis of time dynamics and trends Forming the methodical and practical basis for regular monitoring, modeling and verification of rating models Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 6

Comparison methods and base scale Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 7 Comparison methods for rating scales include: Methodology and principles of mapping of rating scales Criteria for comparison of rating scales (Mathematics) Econometric models for scales comparison Audit of the conformity table and the coordination of its structure Comparison methods are concluded to have: Choice of a base rating scale Mapping system for displaying external and internal ratings into a base scale Application to each class of rating entities (banks, companies, etc.) Allowing simultaneous use of all independent rating estimations Rating Scales Numeric Scales RS 1 RS i RS N NS 1 NS i NS N BS F1(α1)F1(α1) Fi(αi)Fi(αi) FN(αN)FN(αN) Base Scale

Rating modeling (Ordered probit models) Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 8 where x i – is a set of independent variables Rating is a depended variable y Less values of y are connected with higher agency-ratings Ratings are represented as a numeric scale: 12+ grades

Research data 10 rating scales: 4 national rating agencies 3 international agencies (3+3) Time period: 1q2006 – 4q2010 (20 quarters) 370 Russian banks with at least 1 rating during this period Total 7400 observations of banks Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 9

Numerical scales Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 10 S&PS&P (rus)Moodys Moodys (rus) FitchFitch (rus) Expert RA NRAАК&М Rus- Rating SPSP_ruMM_ruFF_ruERANRAAKMRR AAA1ruAAA1Aaa1Aaa.ru1AAA1AAA(rus)1A++1AAA1A++1AAA1 AA+2ruAA+2Aa12Aa1.ru2AA+2AA+(rus)2A+2AA+2A+2AA+2 AA3ruAA3Aa23Aa2.ru3AA3AA(rus)3A3AA3A3 3 AA-4ruAA-4Aa34Aa3.ru4AA-4AA-(rus)4B++4AA-4B++4AA-4 A+5ruA+5A15A1.ru5A+5A+(rus)5B+5A+5B+5A+5 A6ruA6A26A2.ru6A6A(rus)6B6A6B6A6 A-7ruA-7A37A3.ru7A-7A-(rus)7C++7A-7C++7A-7 BBB+8ruBBB+8Baa18Baa1.ru8BBB+8BBB+(rus)8C+8BBB+8C+8BBB+8 BBB9ruBBB9Baa29Baa2.ru9BBB9BBB(rus)9C9BBB9C9 9 BBB-10ruBBB-10Baa310Baa3.ru10BBB-10BBB-(rus)10 BBB-10 BBB-10 BB+11ruBB+11Ba111Ba1.ru11BB+11BB+(rus)11 BB+11 BB+11 BB12ruBB12Ba212Ba2.ru12BB12BB(rus)12 BB12 BB12 BB-13ruBB-13Ba313Ba3.ru13BB-13BB-(rus)13 BB-13 BB-13 B+14ruB+14B114B1.ru14B+14B+(rus)14 B+14 B15ruB15B215B2.ru15B B(rus)15 B B-16ruB-16B316B3.ru16B-16B-(rus)16 B-16 CCC+17ruCCC+17Caa117Caa1.ru17CCC+17 CCC+(rus ) 17 CCC+17 CCC18ruCCC18Caa218Caa2.ru18CCC18CCC(rus)18 CCC18 CCC-19ruCCC-19Caa319Caa3.ru19CCC-19CCC-(rus)19 CCC-19 CC20ruC20C C.ru20C C(rus)20 C

Criteria for choosing the mapping function Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 11 Q – set of all observations {t – period of time, j – bank, R i1jt – Moodys rating (base scale), R i2jt – rating of another agency } t = 1, …, T j = 1, …., K F i1 : R i R base F i = α i1 f i (R i ) + α i2 f i – linear, polynomial, logarithmic function that transforms rating into a base scale

Mapping function Multiple mapping into the base scale: linear logarithmic polynomial (up to 5th power) Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 12 Moodys – Moodys (rus) Moodys – S&P

Logarithmic model of multiple mapping Moodys credit ratings (R) and default probabilities (PD) of banks are approximated by a logarithmic dependence during the years Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 13 PD = 0,000218×R 3,8 PD (%) R M = constR a Ln(M) = aLn(R)+b VariableCoefficients a,bp LOG(M_RU)*D_M_RU0,2540,000 D_M_RU2,2020,000 LOG(SP)*D_SP0,9160,000 D_SP0,1460,029 LOG(SP_RU)*D_SP_RU0,2650,000 D_SP_RU2,1130,000 LOG(F)*D_F0,7490,000 D_F0,5940,000 LOG(F_RU)*D_F_RU0,2130,000 D_F_RU2,1620,000 LOG(AKM)*D_AKM0,2690,000 D_AKM2,4910,000 LOG(ERA)*D_ERA0,3730,000 D_ERA2,3290,000 LOG(RR)*D_RR0,6740,000 D_RR1,0160,000 LOG(NRA)*D_NRA0,1630,000 D_NRA2,4740,000 Number of Observations3432 R2R2 0,902 Logarithmic model for years:

Rating comparison (logarithmic scales) Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 14 Moodys S&P Fitch Fitch (rus) Moodys (rus) S&P (rus) Rus-Rating Expert RA AK&M NRA

Comparison of international banks 3639 pairs (Moodys – another agency) Bank data 1995 – different banks Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 15 Moodys S&P Fitch Credit rating comparison for scales of international agencies (logarithmic model)

Conclusion Econometric models for ratings play significant role due to IRB Approach and other Basel II recommendations and should be developed Scientific and practical basis of using econometric rating models for bank risk management is discussed Comparison method of ratings of different agencies lies in the basis of Unified Rating Space modeling system Scales Mapping Concept and methods are built Including the criteria for choosing the function of transformation of rating value into the base scale Comparison of credit ratings has been performed Models were verified by international bank data and other mapping approaches The main problems are DATA, MONITORING and VERIFICATION of models Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 16

Q & A Alexander Karminsky, Prof., Dr. Vladimir Sosyurko Alexander Vasilyuk Higher School of Economics (HSE) Moscow, Russia Thank you for your attention! Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 17

Stages for building a comparison method The choice of the base scale for ratings Displaying all ratings in the numerical scales Determining the most informative functions of transformations of scales Determining the distance between two ratings of different agencies for the same bank at the same time (Euclidean distance) Formation of a proximity criteria in the form of an integral function of pairwise distances (adopted by the sum of squared pairwise distances for all pairs of comparable ratings) Comparative analysis of modifications of the methods due to language restrictions on the set of compared pairs (for comparison between pairs of ratings from the selected base scale, symmetrical setting without isolation of the scale) Determination of statistical characteristics of the compared approaches, and selection of the main Solution to the problem of minimizing the distance for estimating the parameters of the maps for all scales Formation compliance scheme Choice accuracy of representation for the table (adopted accuracy of a quarter grades) Formation of concordance and the presentation of its regulator Karminsky, Sosyurko, Vasilyuk - Comparison of Bank Credit Ratings Assigned by Rating Agencies EBES 2011 CONFERENCE - ISTANBUL 18

EBES 2010 CONFERENCE - ATHENS Karminsky etc. 19 / 20 References Altman E. and A. Saunders, (1998). Credit risk measurement: Developments over the last 20 years. Journal of Banking & Finance. Altman E. and H. Rijken, (2004). How rating agencies achieve rating stability. Journal of Banking and Finance. Basel Committee on Banking Supervision, (2009). Enhancements to the Basel II framework. Jacobson T., J. Linder and K. Roszbach, (2006). Internal ratings systems, implied credit risk and the consistency of banks risk classification policies. Journal of Banking & Finance. Karminsky A., (2009). Rating model opportunities for emerging markets. Proceedings of the Challenges for Analysis of the Economy, the Businesses, and Social Progress International Scientific Conference. Morgan D., (2002). Rating Banks: Risk and Uncertainly in an Opaque Industry. The American Economic Review. Peresetsky A., A. Karminsky, (2008). Models for Moodys bank ratings. Bank of Finland, BOFIT Discussion Papers, 17/2008.

EBES 2010 CONFERENCE - ATHENS Karminsky etc. 20 / 20 Conclusions The main problems of econometric rating modeling are DATA, MONITORING and VERIFICATION Comparison method of ratings of different agencies lies in the basis of rating modeling system Rating models can be used for a remote forecast of credit ratings on the basis of open information. Good predictive power of the company rating models in the mixed scale. Forecast power has increased to 99% with no more than 2 notch divergence Significant factors in regressions that are crucial to the rating agencies methodologies include bank financial and structural factors, country- specific and market indicators Comparison of credit ratings for three rating agencies has shown that S&P is the most conservative rating agency in the banking sector. Moody's credit ratings are the highest and Fitch have has an average level of rating estimations. Further research is to include the analysis of panel data, introduction of new financial indicators to the model and estimation of non-linear dependences, analysis of sensitivity of banks to the market, dependency on the bank type, influence of ownership structure, etc.

А. Карминский В. Сосюрко Система моделей рейтингов 21/27 Этапы реализации регулярного метода Выбор базовой шкалы для отображения в нее всей совокупности рассматриваемых рейтинговых шкал Отображение всех шкал в числовую ось путем сопоставления градациям рейтингов последовательных целых чисел Определение наиболее информативных классов преобразований шкал за счет исследования аппроксимаций вероятностей дефолтов для рейтинговых агентств с устойчивой историей ( отобраны линейное, логарифмическое и степенное (до уровня 5) преобразования) Определение расстояния между двумя отображениями рейтингов различных агентств для одного и того же банка в одно и то же время (принято обычное эвклидово расстояние) Формирование критерия близости отображений в виде интегральной функции от попарных расстояний (принята сумма квадратов попарных расстояний по всем сравнимым парам рейтингов) Сравнительный анализ модификаций методов за счет формулировки ограничений на множество сравниваемых пар (рассмотрены сравнения для пар рейтингов с выбранной базовой шкалой, симметричная постановка без выделения шкалы) Определение статистических характеристик сравниваемых подходов и выбор основного Решение задачи минимизации расстояния для оценивания параметров отображений для всех шкал Формирование схемы соответствия Выбор точности представления для таблицы (принята точность в четверть градации) Формирование таблицы соответствия и представление ее регуляторам

А. Карминский В. Сосюрко Система моделей рейтингов 22/27 Эконометрический подход к мэппингу Шаг 1. Для каждого наблюдения строятся модели ordered logit, на основании данных по рейтингам банков и данным их финансовых и других показателей Шаг 2. Для каждой пары рейтингов рассчитываются соответствующие оценки значения «непрерывного» рейтинга для каждой из рейтинговых шкал у* и z*. Далее, строится преобразование одной непрерывной рейтинговой шкала в другую, т.е. подбирается нелинейная, монотонная функция f, такая, что Методом наименьших квадратов оценивается регрессионное уравнение: Шаг 3. Полученное преобразование позволяет найти образ диапазона непрерывного рейтинга z, соответствующего рейтинговой градации r, в непрерывной шкале первого рейтинга y*. Далее положение этого интервала сопоставляется с точками отсечения первого рейтинга, и производится соответствие дискретных рейтингов.

Применение эконометрического подхода Временной диапазон гг., поквартально Финансовые данные Агентство «Интерфакс» Данные по собственности иностранный, государственный В качестве объясняющих факторов были выбраны следующие показатели: l_ta логарифм от совокупных активов; npl_ltr Просроченные кредиты/Кредиты НБС- резидентам; d_ta Депозиты НБС/Совокупные активы; llp_ltr Резервы под кредиты НБС/Кредиты НБС- резидентам; pe_ta Расходы на персонал/Совокупные активы; c_ta Собственный капитал/Совокупные активы; n1 Норматив достаточности капитала (Н1); n7 Норматив максимального размера крупных кредитных рисков (Н7); for_dum Индикатор иностранного владельца; st_dum Индикатор государственного банка; n4 Норматив долгосрочной ликвидности банка (Н4); кроме того, были опробованы варианты с показателями: n3 Норматив текущей ликвидности банка (Н3); dfe Депозиты предприятий, % от депозитов НБС lte Кредиты предприятиям, % от кредитов НБС- резидентам dfe_ta Депозиты предприятий/Совокупные активы; lte_ta Кредиты предприятиям/Совокупные активы; А. Карминский В. Сосюрко Система моделей рейтингов 23/27 Отображение шкалы S&P в шкалу Moody's. 2007:1–2010:4 Отображение шкалы Fitch в шкалу Moody's. 2007:1–2010:4