Crowd-sourced Credit Transition Matrices and CECL

Transcription

1 Crowd-sourced Credit Transition Matrices and CECL 4 th November 2016 IACPM Washington, D.C. COLLECTIVE INTELLIGENCE FOR GLOBAL FINANCE

2 Agenda Crowd-sourced, real world default risk data a new and extensive research resource Developing robust, granular, high frequency credit transition matrices Using crowd-sourced transition matrices to optimize CECL impairment estimates 2

3 Credit Benchmark Overview Credit Benchmark: financial data analytics company the combined credit views of the world s leading banks Founded 2013 by Mark Faulkner and Donal Smith; backed by Balderton Capital and Index Ventures 30+ employees in London and New York Contributing banks provide (1Y TTC) Probability of Default ( PD ). These are aggregated, anonymized and published back to the contributing banks 11 banks in all 3 global regions Coverage spans multiple geographies, sectors and entity sizes, from large multinational companies to SMEs. Methodology Committee meets quarterly, attended by all contributing banks Technical Advisory Group includes academics and former practitioners Regular research and blog updates posted on website Regular engagement with regulators 3

4 Credit Benchmark Service Banks Credit Benchmark Service Contributor A Contributor B Contributor C 65 Sovereigns Consensus Calculation Engine 5% of banks PDs & ¾ of listed companies Data cleansing and LGDs Virtually validation all private companies, emerging market Entity extraction corps and SMEs Consensus calculation Virtually all hedge and mutual funds CB Service Data feeds Analytical tools Reporting suite Visualization tools Consultancy Regulatory monitoring Regular webinars News and Research Membership of CB Methodology Committee Technical advisory group Credit Benchmark Service 4

5 Credit Benchmark Universe Crowd-sourced updates are frequent (monthly), and most legal entities remain unrated by the CRAs, and updates are infrequent: Potential Credit Benchmark coverage universe Rated Entities: <10,000 67% of Sovereigns 25% 2007 of banks 25% of listed corporates Unrated Entities: 33% of Sovereigns 75% of banks 75% of listed corporates Most private corporates Most emerging market corporates Most SMEs Virtually all hedge and mutual funds A range of dimensions geographic, sector, public vs. private, rated vs. unrated - support a rich set of transition matrices 5

6 May-15 Jun-15 Jul-15 Aug-15 Sep-15 Oct-15 Nov-15 Dec-15 Jan-16 Feb-16 Mar-16 Apr-16 May-16 Jun-16 Jul-16 Aug-16 Sep-16 Entities Mapped Quorate Count Credit Benchmark Coverage Latest data production statistics (October 2016) Quorate Coverage By Geography 550,000 entity observations collated from banks 268,200 individual entities mapped 15,150 entities with 2 contributing sources 6,050 entities with Credit Benchmark Consensus CBC grades (Quorate, with 3 or more contributing sources) 2,310 with Senior Unsecured LGDs Quorate Entities: USA: 2,395 Canada: 198 UK: 1104 Germany: 140 France: 167 China: 68 Japan: 128 Australia: 79 Density of shading denotes extent of Quorate coverage the greater the coverage the darker the shade Growth of Credit Benchmark s platform Coverage of major indices 300, , , , ,000 50, ,000 6,000 5,000 4,000 3,000 2,000 1, % 20% 40% 60% 80% 100% 193 Sovereigns Top 500 Global Corps 30 GSIB Top 100 Fund Managers Top 100 Single Hedge Fund Managers S&P500 FTSE100 EMIR CCPs S&P Long Term Foreign Rated Entities Mapped Quorate PD Quorate LGD Quorate Mapped Not covered 6

7 Credit Benchmark Consensus ( CBC ) The likelihood of an obligor upgrading or downgrading depends on the volatility of PDs. If PDs are more volatile, then transitions including multiple notch moves are more likely. Midpoints of these PD ranges are the equivalent of crowdsourced ex ante Default rates 7

8 Frequent, smooth updates and trends CRAs tend to update in clusters Crowd-sourced data updates in trends Test below shows that crowd-sourced data has much more trending than would be expected by chance 8

9 aaa aa+ aa aaa+ a a- bbb+ bbb bbbbb+ bb bbb+ b b- ccc+ ccc cccaaa aa+ aa aaa+ a a- bbb+ bbb bbbbb+ bb bbb+ b b- ccc+ ccc ccc- Credit Transition Matrices ( CTMs ) The observed distribution of obligors across credit categories reflects the transition matrix* *[A full matrix would also include the rate of formation of new companies] 9

10 Obligors by Credit Category The observed distribution of obligors across credit categories reflects the transition matrix: Low number of obligors may distort CTMs Low number of obligors may distort CTMs Source: All Credit Benchmark quorate obligors, averaged across 12 months. 10

11 Defaults & Emergences S&P long run Corporate default rate shown below is used for cumulative PD term structures. Transition matrix comparisons use the same default rates to isolate transition effects. S&P long term matrix also assumes no obligors emerge from default bottom row is zeros in off-diagonals. (Source: S&P) 11

12 Diagonals ( Stable ) Crowd-sourced data shows a higher propensity to transition in all credit categories Crowd-sourced and CRA data show a high propensity to transition in lower credit categories because the PDs are proportionately more volatile in these categories S&P: Average = 87.2% Crowd-sourced: Average = 81% (Corporate only; both adjusted for S&P default frequencies; AAA excluded due to small sample. 12

13 Off-Diagonals ( Up / Down ) Crowd-sourced data shows a tendency to more activity (up or down) and in particular a bias towards the upward revisions. Crowd-sourced 2 nd diagonal elements (two notch transitions) are higher than the S&P equivalent. 1-year transition frequency, averaged across credit categories S&P: All corporates, past 30 years Crowd-sourced: Global corporates, last 12 months Leading Diagonal ( Stable ) 87.2% 81% 1 st Off-diagonal Downgrade 6.8% 7.2% 2 nd Off-diagonal Downgrade 0.6% 1.3% 1 st Off-diagonal Upgrade 7.0% 8.3% 2 nd Off-diagonal Upgrade 0.3% 1.8% 13

14 Crowd-sourced Comparisons 1) US Corporates transition more often in all categories 2) Upgrade frequency slightly higher than downgrade for both US and EU 3) Corporate & Financial: very similar transition propensity, except for one notch upgrade 1-year transition frequency, averaged across credit categories Crowd-sourced Corporate: US vs EU % Crowd-sourced: Global Corporate vs Financial % US EU Corp Fin Leading Diagonal ( Stable ) st Off-diagonal Downgrade nd Off-diagonal Downgrade st Off-diagonal Upgrade nd Off-diagonal Upgrade

15 PD Volatility & CTMs (1) PD volatility determines the CTM structure. calibrated to Crowd-sourced data: Results below are based on simulations, Frequency of Downgrade (+) / Upgrade(-) PD Monthly Volatility % 6.2% 87.4% 6.4% 15% 0.40% 9.4% 79.8% 10.2% 0.16% 25% 0.63% 2.46% 14.1% 68.2% 13.7% 0.95% 0.00% 1) Higher PD volatility leads to more frequent and larger notch transitions 2) Trending ( Path dependency ) also leads to more frequent and larger notch transitions 3) Volatility approach removes noise from CTM much faster and allows systematic monitoring of forthcoming changes in CTM behaviour. 15

16 PD Volatility & CTMs (2) PD volatility determines the CTM structure: 1) Lower credit categories have proportionately higher volatility, so transition more often 2) Higher general PD volatility leads to more frequent and larger transitions 3) Regional, Sectoral and Temporal CTM differences are due to PD volatility differences Obligor Set Monthly Standard Deviation of PDs European Corporate 14.5% North America Corporate 13.7% Asian Corporate 11.9% Global Corporate 11.8% Global Financial 12.2% 16

17 21 x 21 Transition Matrix Crowd-sourced dataset is broad and deep enough to begin supporting regular, large CTM updates. Still shows effect of sampling variation, especially smaller samples in tails; but signal is rapidly growing vs noise. Plans for 42 x 42 in Source: Credit Benchmark; US Corporates, past 12 months. 17

18 Cumulative PD term structures Powering up the CTM gives the cumulative term structure probabilities of default. Choice of CTM is critical: S&P Corporate Crowd-sourced For AA, A and BBB equivalent credit categories, Crowd-sourced data gives significantly higher cumulative default probabilities (due to the higher propensity to transition) 18

19 Through-the-Cycle and Point-in-Time Crowd-sourced data is 1Y TTC/Hybrid, so longer time horizons embedded Point in Time PD estimates contain risk premiums which may be significant vs PD Risk Premiums over time Cumulative Real World PD term structures Vary by credit class and asset type; market implied >> real world Can be compared with market-implied PDs and can be used to derive synthetic yields for comparison with observed yield curves 19

20 Probability of Default, % Probability of Default, % Probability of Default, % CECL application: Cumulative PDs 0.20% 0.15% 0.10% 0.05% 0.00% aaa Transition Approach Survival Rate Approach % 24% 18% 12% 6% 0% bb Transition Approach Survival Rate Approach % 75% 50% 25% 0% ccc - c Transition Approach Survival Rate Approach Transition Matrices vs Simple extrapolation (Survival rate): Differences can be material depending on credit category Years Years Years The 5-year loss is 6bp using the survival rate approach, 18bp using the transition matrix The 5-year loss is 2,450bp using the survival rate approach, 2,280bp using the transition matrix The 5-year loss is 8,400bp using the survival rate approach, 5,900bp using the transition matrix S&P Corporate Crowd-sourced CRA vs Crowd-sourced: Difference in cumulative PDs at least 20% 20

21 CECL (and IFRS9) Impairment 1. Market implied PDs contain significant risk premiums, depending on credit category: will tend to overstate impairments 2. Simple PD extrapolation may overstate or understate impairment, depending on credit category and comparison CTM structure 3. CTMs derived from CRA long run averages may understate impairment if the current PD volatility level is above the long run average; they will overstate impairment if the current PD volatility level is below the long run average 4. Crowd-sourced CTMs provide: a) Real World PDs (undistorted by risk premium) b) Cumulative PDs which reflect the CTM, (ignored by the survival approach) c) Frequently updated PD volatility and transition estimates d) Region and Sector specific PD volatilities and CTMs 21

22 Conclusions The Credit Benchmark dataset is a large and frequent source of credit risk updates. The dataset tracks all such changes across the mapped universe (which exceeds 250,000 names) for a variety of time periods (one month to one year). The resulting transition matrices can be used to estimate cumulative PD curves and provide benchmarks for Point-in-Time risk estimates derived from market prices. These can then be used to estimate monthly risk premiums for different obligor types, credit classes and time periods. Through-the-Cycle data is well-suited to the calibration of transition matrices with multiple time horizons and provides fresh insights into the path dependency issue. These benchmarks have applications for compliance with IFRS9 and CECL accounting standards, and the choice of CTM and PDs may have significant financial implications. 22

23 Harry Chopra Mahim Mehra David Carruthers Chief Commercial Officer Global Head Contributor Relationships Head of Research 23

24 DISCLAIMER We have prepared this document solely for informational purposes. You should not definitely rely upon it or use it to form the basis for any decision, contract, commitment or action whatsoever, with respect to any proposed transaction or otherwise. You and your directors, officers, employees, agents and affiliates must hold this document and any oral information provided in connection with this document in strict confidence and may not communicate, reproduce, distribute or disclose it to any other person, or refer to it publicly, in whole or in part at any time except with our prior consent. If you are not the recipient of this document, please delete and destroy all copies immediately. Neither we nor our affiliates, or our or their respective officers, employees or agents, make any representation or warranty, express or implied, in relation to the accuracy or completeness of the information contained in this document or any oral information provided in connection herewith, or any data it generates and accept no responsibility, obligation or liability (whether direct or indirect, in contract, tort or otherwise) in relation to any of such information. We and our affiliates and our and their respective officers, employees and agents expressly disclaim any and all liability which may be based on this document and any errors therein or omissions therefrom. Neither we nor any of our affiliates, or our or their respective officers, employees or agents, make any representation or warranty, express or implied, that any transaction has been or may be effected on the terms or in the manner stated in this document, or as to the achievement or reasonableness of future projections, management targets, estimates, prospects or returns, if any. Any views or terms contained herein are preliminary only, and are based on financial, economic, market and other conditions prevailing as of the date of this document and are therefore subject to change. We undertake no obligation to update any of the information contained in this document. 24

25 COLLECTIVE INTELLIGENCE FOR GLOBAL FINANCE London: Eagle House 167 City Road London EC1V 1AW New York: 205 East 42 nd St 20 th Floor New York NY (0)