Interfaces with Other Disciplines
Bank-sourced credit transition matrices: Estimation and characteristics

https://doi.org/10.1016/j.ejor.2020.06.024Get rights and content

Highlights

  • The study introduces bank-sourced credit transition matrices (CTMs).

  • Bank-sourced CTMs are substantially influenced by the choice of aggregation method.

  • The CTM differences can lead to 7.3% higher 99% credit value-at-risk estimates.

  • Bank-sourced CTMs are more dynamic than those of credit rating agencies.

  • Industry-specific CTMs indicate existence of industry-specific business cycles.

Abstract

This study proposes and analyses a novel alternative to credit transition matrices (CTMs) developed by credit rating agencies - bank-sourced CTMs. It provides a unique insight into estimation of bank-sourced CTMs by assessing the extent to which the CTMs depend on the characteristics of the underlying credit risk datasets and the aggregation method and outlines that the choice of aggregation approach has a substantial effect on credit risk model results. Further, we show that bank-sourced CTMs are more dynamic than those of credit rating agencies, with higher off-diagonal transition rates and higher propensity to upgrade. Finally, we create a set of industry-specific CTMs, otherwise unobtainable due to the data sparsity faced by credit rating agencies, and highlight the implications of their differences, signalling the existence of industry-specific business cycles. The study uses a unique and large dataset of internal credit risk estimates from 24 global banks covering monthly observations on more than 26,000 large corporates and employs large-scale Monte Carlo simulations. This approach can be replicated by regulators (e.g., data collected by the European Central Bank in the AnaCredit project) and used by organisations aiming to improve their credit risk models.

Introduction

Credit risk captures the loss resulting from a counterparty failing to meet its obligations in accordance with agreed terms and it is linked mainly to loan exposures and fixed-income securities. As such, it is one of the core risks for financial institutions and it is closely monitored by regulators and researched by academics (recent studies include Augustin, 2018, Fernandes, Artes, 2016, Brigo, Francischello, Pallavicini, 2019 and Altman, Esentato, & Sabato, 2020). A key measure of credit risk is probability of default, represented by percentage or a list of credit rating categories, quantifying the likelihood of a default event over a particular time horizon (usually one year). The time dynamics of credit risk can then be captured using credit transition matrices (CTMs) which indicate the probabilities of moving from one credit rating category to another in a given time period. CTMs are an essential component of credit risk modelling (Jarrow, Lando, Turnbull, 1997, Israel, Rosenthal, Wei, 2001, Boreiko, Kaniovski, Kaniovski, Pflug, 2019) with practical applications in portfolio risk assessment, modelling of credit risk premia term structure, pricing of credit derivatives, bank stress-testing and life-time credit loss estimation under IFRS9 and CECL accounting standards.

The existing industry standard is to source CTMs from credit rating agencies (CRAs). However, CTMs estimated using CRA data are based on a limited set of rated entities typically representing only a small proportion of counterparties in a financial institution’s portfolio (especially in case of non-US entities and smaller enterprises), potentially causing modelling inaccuracy. Equally, it is generally not possible to estimate industry- or country-specific CTMs, even though both of the dimensions have been shown to affect CTMs (Frydman, Schuermann, 2008, Nickell, Perraudin, Varotto, 2000), and the resulting annual CTMs are considered inferior to long-term average of transition matrices adjusted for business cycle phase (see e.g. Wei, 2003), as they may show abnormal behaviour such as non-monotonic transition rates when a change across multiple rating categories is more likely than a one-category change (Kreinin & Sidelnikova, 2001). Last but not least, credit rating agencies face a potential conflict of interest as they are compensated by the rated company (De Haan, Amtenbrink, 2011, European Commission, 2010, Strier, 2008).

Our paper analyses an alternative approach to CTM estimation: bank-sourced CTMs based on aggregation of internal credit risk estimates pooled from multiple banks. This has multiple benefits with the potential to overcome the aforementioned issues of CRA-sourced CTMs. Firstly, the resulting entity portfolio, which can be multiple times larger than in case of CRAs, provides better representation of the economy and the increased sample size allows for estimation of country- and industry-specific CTMs. Secondly, bank-sourced data inherently reflect the phase of business cycle (see below) and the resulting annual CTMs can be directly used in risk modelling. Finally, the bank-debtor relationship avoids the potential conflict of interest risk faced by CRAs. These may lead to higher accuracy of the resulting CTMs. Bank-sourced CTMs can be particularly useful for regulatory purposes and stress testing, as various regulators are collecting increasing amounts of data from banks (e.g., the recent AnaCredit project run by the European Central Bank involves collection of the internal probability of default estimates from all of the Eurozone’s credit institutions1).

Unfortunately, banks’ internal credit risk estimates are not publicly available and have therefore not been extensively researched in relation to CTM estimation, with the existing studies focusing on limited subsets of data and not discussing performance of alternative aggregation mechanisms (see e.g. Gavalas & Syriopoulos, 2014 for European central bank data; Gómez-González & Hinojosa, 2010 for Columbian commercial loans; and Lu, 2012 for Taiwanese data). Bank-sourced CTMs can be significantly affected by specifications and overlap of individual bank portfolios; such dynamics must be considered when designing a model for CTM estimation in order to maximise its accuracy.

This study contributes to the literature on CTMs in the following three ways, none of which has been discussed in the literature yet. Firstly, we propose and analyse the three aggregation approaches – the observation-based method, entity-average-based method, and method based on the average of bank-specific CTMs – and assess the resulting transition rates and value-at-risk estimates, providing an overview of the trade-offs to be considered when developing a bank-sourced CTM aggregation model. Secondly, we estimate a series of bank-sourced CTMs and compare their characteristics to those provided by CRAs. Finally, we produce a set of novel, industry-specific CTMs possibly indicating existence of industry-specific credit cycle.

The study uses a unique large dataset of probability of default (PD) estimates sourced from 24 global banks approved by regulators to use the advanced internal ratings-based (A-IRB) approach to credit risk estimation, allowing them to employ internal credit risk models to calculate PD estimates. The dataset consists of 1.74 million monthly observations of PD estimates covering more than 26,000 large corporates in North America, United Kingdom and the European Union (EU) over the period of 2015–2019. The data are used for analysis of the three aggregation approaches and estimation of overall and industry-specific bank-sourced CTMs. To evaluate association between differences in the three versions of CTMs and data characteristics, we utilise large-scale Monte Carlo simulations driven by relationships among credit risk level and change variables observed in the data and introduce controlled variance in 12 selected parameters.

Our analysis shows that bank-sourced CTMs are substantially influenced by the choice of aggregation method and that the differences are driven by the entity overlap among banks, size of their PD data samples, initial PD distributions, and rating changes. Using value-at-risk assessment, we estimate that the CTM differences can lead to 7.3% higher 99% credit value-at-risk estimates based on a CreditMetrics calculation. Comparing the rich bank-sourced CTMs against corporate CTMs produced by the three major credit rating agencies, covering 2,000-5,000 entities each, our analysis highlights that the bank-sourced CTMs exhibit relatively high off-diagonal transition rates and more favourable features overall, including a close to bell-shaped steady state distribution and a clear linear pattern in the relationship between transition rates and notches. Finally, the industry-specific CTMs, otherwise unobtainable due to the data sparsity faced by rating agencies, indicate existence of industry-specific business cycles which can be critical fro IFRS9 modelling.

The study is structured as follows. First, we introduce the CTM notation, the main methods for CTM estimation, comparison, and aggregation of the underlying datasets, and the bank-sourced PD estimates used in our analysis. Subsequently, we compare the three aggregation methods for CTM estimation, construct empirical, bank-sourced CTMs and compare them against CTMs obtained from CRAs. Finally, we analyse the industry-specific CTMs.

Section snippets

Concept of transition matrices

Credit transition matrices are estimated using historical data on companies’ credit risk estimates. The two most common approaches to CTM estimation are cohort (discrete time) and duration (continuous time) methods (Jafry, Schuermann, 2004, Fuertes, Kalotychou, 2007); the straightforward cohort approach has become the industry standard (Schuermann, 2008) and is used by credit rating agencies. Both approaches are based on the time-homogeneous Markov chain assumption (Jarrow & Turnbull, 1995).

Data

The unique empirical dataset used in our study is provided by Credit Benchmark and contains PD estimates from 24 global banks. This section discusses the source of the data, data characteristics, features of the bank-sourced transition matrices, banks’ internal risk systems and modelling considerations.

Comparison of aggregation methods

In this section, we use the extensive empirical dataset of bank-sourced PD estimates to analyse the characteristics of the CTMs estimated using the three aggregation principles presented in Eqs. (6)–(8). These are assessed in terms of differences in the observed CTMs and dependence of the differences on data characteristics such as portfolio overlap among banks, size of data samples and initial PD distributions. To do so, we use Monte Carlo simulations with data-derived parameters to create a

Bank-sourced vs CRA credit transition matrices

The analysis thus far was essential for understanding how the specifics of banks’ credit risk portfolios affect the results of their aggregation. In this section, we turn to focusing on a practical application of bank-sourced CTMs, comparing the 2018 bank-sourced CTM for North American and EU large corporates with the 2018 CRA corporates CTMs. The bank-sourced transition matrix is based on PD estimates on 26,000 entities and estimated using the cohort approach defined in Eq. (2) and the

Conclusion

Banks’ internal credit risk estimates can be used to create an industry standard for credit transition matrices, overcoming the issue of data sparsity and potential conflict of interest faced by rating agencies, which are currently the main source of CTMs in the field. Indeed, data from banks provide a greater level of detail than data from credit rating agencies and allow estimation of country- and industry-specific transition matrices, which may lead to improvements in the accuracy of

References (35)

  • R.Y. Rubinstein et al.

    Simulation and the Monte Carlo method

    (2016)
  • Wittmann, A. (2007). CreditMetrics: Functions for calculating the CreditMetrics risk model. R package version 0.0–2....
  • E.I. Altman et al.

    Assessing the credit worthiness of Italian SMEs and mini-bond issuers

    Global Finance Journal

    (2020)
  • Basel Committee on Banking Supervision

    Studies on the validation of internal rating systems

    Working Paper 14

    (2005)
  • Basel Committee on Banking Supervision

    Basel II: International convergence of capital measurement and capital standards: a revised framework, comprehensive version

    (2006)
  • C. Bluhm et al.

    Calibration of PD term structures: to be Markov or not to be

    Risk

    (2007)
  • D. Boreiko et al.

    Identification of hidden Markov chains governing dependent credit-rating migrations

    Communications in Statistics-Theory and Methods

    (2019)
  • Cited by (6)

    This work was supported by the Czech Science Foundation (Project no. GA 18-05244S) and the Charles University PRIMUS program (project PRIMUS/19/HUM/17). I wish to confirm that there are no known conflicts of interest associated with this publication. The analysis and conclusions are those of the author. Credit Benchmark is not responsible for any statement or conclusion herein, and opinions or theories presented herein do not necessarily reflect the position of the institution. Declarations of interest: the author is an employee of Credit Benchmark.

    View full text