us current expected credit losses (cecl) industry benchmark … · gcd is working with accenture...

Global Credit Databy banks for banks

US Current Expected Credit Losses (CECL) Industry Benchmark Study

Study Kick-off

02/07/2019 1Public – CECL Benchmark Study Kick-off

Global Credit Data

Project Team Benchmarking Study

2

Soner Tunay

Head of Quantitative Analytics, Finance & Risk Services at Accenture Consulting

[email protected]

Daniela has 15 years of experience in credit risk management, modelling and reporting. Prior to Global Credit Data, she was heading the financial risk team at Delta Lloyd Bank where she was responsible among others for the Basel III migration plan and the implementation of AIRB-compliant credit risk models. Her profile is rounded by risk positions at RSU Rating Service Unit, PricewaterhouseCoopers and NIBC.

Soner has held executive level positions leading model development and validation functions in various US banks, FBOs and G-SIBs. Soner is regarded an industry thought leader in Risk Rating analytics, CCAR and CECL models. He chairs industry conferences and teaches masterclasses to industry participants on credit risk and integration of modeling with business aspects. Soner received his Ph.D. in Economics from Boston College.

Daniela Thakkar

Methodology & Membership Executive, Global Credit Data

[email protected]

Martin Boer

Director of Regulatory Affairs, Institute of International Finance

[email protected]

Martin advocates and contributes to the IIF work on regulatory consistency, impact assessment, prudential capital and liquidity standards, accounting, insurance regulation and the growing area of non-bank/non-insurance regulatory issues. Martin previously served as the Secretary General of the European Financial Services Round Table. Before that he served in various positions at ING Group, UNDP, The Financial Times and Bloomberg News.

02/07/2019Public – CECL Benchmark Study Kick-off

mailto:[email protected]



Global Credit Data

Industry Experts in this webinar

3

Dr. Calcagno is a Managing Director in MUFG Union Bank’s Credit Risk Analytics team with over 10 years of Financial Services experience. Juan is an expert applied econometrician and leads teams responsible for quantitative modeling and innovative ideas underlying macroeconomic forecast and stress testing to address the heightened quantitative and qualitative expectations around CCAR/DFAST/CECL programs. Prior to joining MUFG Union Bank, Juan worked for KPMG and Moody’s Analytics, Economy.com division, on several CCAR/Basel engagements as well as on private ABS/RBMS valuation projects. He holds a PhD from Columbia University.

Stevan is Senior Vice President and head of Quantitative Risk Analytics at Regions Bank, where his current responsibilities focus on quantitative aspects of forecasting and stress testing, risk rating, valuation, economic capital, credit strategy, reserve methodologies and credit portfolio management. Steve has 20 years of industry experience in quantitative modeling and risk management and has prior experience building portfolio management and analytics infrastructure at Merrill Lynch, Bank of Montreal and ABN AMRO. Steve has a Ph.D. in applied physics from Northwestern University, a B.S. in physics from University of Colorado in Boulder, and has held Series 7 and Series 63 certifications.

Juan Calcagno

Managing Director at MUFG Union Bank

Stevan Maglic

Senior Vice President and Head of Quantitative Risk Analytics at Regions Bank


Global Credit Data

Agenda

4

INTRODUCTION

PRACTIONERS EXPERIENCE

EXPERIENCE FROM SIMILAR STUDIES

US CECL INDUSTRY BENCHMARK STUDY METHODOLOGY

CONTACTS


Global Credit Data

Background

5

26 Banks Participated in Step 1 – CECL Methodology Survey:

CECL will have the most significant impact on American banking since the Dodd-Frank Act was adopted. Models currently being developed indicate that bank profitability—and bank department profitability—will be affected as institutions charge for credit loss provisions on new loans and credit downgrades on existing loans.

GCD IS WORKING WITH ACCENTURE AND THE IIF TO HELP U.S. FINANCIAL INSTITUTIONS BENCHMARK THEIR CECL MODELS AS THEY ARE DEVELOPED.

STEP 1 was a detailed CECL Methodology survey sponsored by Global Credit Data, the IIF and Accenture. The survey provided comparative information on bank’s planned methodologies for implementation of CECL. The questions focused on credit data and modeling for the C&I CRE and Consumer portfolios.

Any financial services companies who are implementing CECL, including regional banks, community banks, supranational

banks, global banks, insurance companies, specialty financial companies, etc.

What drives the difference in those numbers?

STEP 2 is a benchmarking study, similar to the highly successful benchmarking study conducted for IFRS 9. The CECL study will be launched in Q1/2019.

KEY QUESTIONS ADDRESSEDWHO SHOULD PARTICIPATE?How do your final numbers compare to

your peers?


https://www.globalcreditdata.org/services/ifrs-9-hypothetical-portfolio-study

Global Credit Data

Practitioners Perspective

6

Juan Calcagno

Managing Director at MUFG Union Bank

Stevan Maglic

Senior Vice President and Head of Quantitative Risk Analytics at Regions Bank

Principles-based Approach Is Not Prescriptive

CECL is vastly different than ACL and IFRS9 with a number of key decisions being left up to practitioner’s perspective

There Is A Need For Industry-Wide Impact Assessment

Overall, there is a lack of Quantitative Impact and benchmark studies in the marketplace

Scope From Practitioners Perspective

A study such as this can inform design decisions and can be a starting point to address issues such as pro-cyclicality and capital impact


Global Credit Data

Existing Studies

7

GCD’s IFRS 9 Benchmarking Study Study run in 2017 and 2018 on variability of IFRS 9 ECL

estimates between banks

26 banks participated, globally

Main outcome: • Principle-based approach

of IFRS 9 results in a widevariability of theECL provision charge between banks even under a common scenario and a common, hypothetical portfolio

• Reasons: banks use different models, parameters and assumptions

High recognition by Regulators and Accountants

CECL Studies So Far Existing studies so far approached the impact of CECL

from a narrow position; no true benchmarking capability so far

No quantitative impact study yet, but there are few Interesting papers on the impacts of CECL and onprocyclicality

For example, Accenture studyconfirms high CECL reserve sensitivity to key assumptions

Intensive industry discussion currently going onregarding Day1-implemenation



https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3183416



Global Credit Data

Overview Methodology & Approach

8

Hypothetical Portfolio

CECL Parameters

CECL ScenariosCECL Model

Suite of Individual

Bank

Lifetime PD

Lifetime LGD

Lifetime EAD

ECL

Collect, anonymize, and aggregate data

Return anonymized data & benchmarking reports back to banks

Data Analysis

Result Interpretation

Socialize, Fine-tune, and Publish Study Report

PARTICIPATING BANKS PROJECT TEAMINDUSTRY EXPERT GROUP

Designs Template, Scenarios, Approach, FAQ, …

Ensures Template fits to US market and majority of CECL models

Run their individual models based on the hypothetical portfolio, scenarios and parameters provided

Submit result to GCD

Template Results submitted bybanks to GCD

End-to-End project management (from “template” to “data collection” to “report” to “industry advocacy”)

Industry Advocacy


Global Credit Data

Template: Hypothetical PortfolioThe hypothetical portfolio is composed of a set of loans chosen to provide coverage of various types of asset classes/borrowers, exposures and facilities, fit to the US market.

9

ASSET CLASS / BORROWER TYPE

Residential MortgageHelocAuto

Personal UnsecuredCredit Cards

Large CorporatesMid Market

Business Banking CRE – IP, Construction

OBLIGOR

Credit QualityIndustry

Geography

EXPOSURE

EAD RangeAvg. Usage

O/SLimit

Revolver / TermReference Rate

Maturity

FACILITY

LGD RangeFacility Type

Secured / Unsecured

PERMUTATIONS BY ASSET CLASS / BORROWER TYPE

Residential Mortgage 64Heloc 16Auto 32

Personal Unsecured 18Credit Cards 64

Large Corporates 12Mid Market 12

Business Banking 12CRE – Income Producing, Construction 4

Banks Participating in the Working Group of Hypothetical Portfolio Development

*The study will open for public commentary and feedback regarding the construction of the hypothetical portfolio until February 14th, 2019

ILLUSTRATIVE


Global Credit Data

Template: Scenarios & ParametersThe industry expert group has also developed CECL scenarios and parameters for banks to test their own models on the hypothetical portfolio.

10

CECL PARAMETERS

Banks are asked to run the CECL models various times with different parameters:

• 4 key parameters are included – Reasonable & Supportable (R&S) Period, Mean Reversion Period, Mean Reversion Method, and Historical Loss Rate* beyond R&S Period

• This results into 5 different runs per scenario

CECL SCENARIOS

This first run of the benchmark study will be based on FRB’s CCAR scenarios published in February 2019 (later iterations might include further scenarios)

• Using CCAR scenarios for this study helps reduce operational burden

• Banks could suggest additional scenarios to be included in the future iterations of the benchmark study


Global Credit Data

Template: Hypothetical Portfolio – Example (1/2)Example: Residential Mortgages (*)

11

* Preliminary Version: The study will open for public commentary and feedback regarding the construction of the hypothetical portfolio till February 14th, 2019

Hypothetical portfolio

contains 64 different

Residential mortgage

types


ATTRIBUTES NOTES PERMUTATIONSGeography U.S. 1Property Type Single Family Condo 2Credit Quality Prime (FICO=760) Subprime (FICO=660) FICO at Mortgage Origination 2Mortgage Type 30 Year Fixed 3/1 ARM 2Loan Size Jumbo (>= $427K) Conforming (< $427K) 2Loan Purpose Ow ner Occupied 1Origination LTV 80 1Origination Year February 1, 2019 February 1, 2003 3/1 ARM Originated in 2003 is already resetDelinquency Status Current 30 DPD 60 DPD 2019 Originated Loans are all "Current"Exposure at Origination 1M 250K Determined by Loan SizeInterest Rate 4% 6% Determiend by Credit QualityHistorical Loss Rate 1% 3% Determined by Credit QualityReference Curve 10 Year Treasury For 3/1 ARM OnlyRefinance/Reset Interest Rate 5% For 3/1 ARM OnlyPrepayment Rate Bank ow n model determinesCurrent LTV Bank ow n model determines

TOTAL PERMUTATIONS 64

Banks' inputs / intermediate outputs

LEGENDSMajor Attributes for segmentation

4

Attributes in dependency of other attributes

PORTFOLIO ATTRIBUTES OF RESIDENTIAL MORTGAGE

DETAILS

Global Credit Data

Template: Hypothetical Portfolio – Example (2/2)


Obs# Geography Property Credit Mortgage Loan Purpose Origination Origination Delinquency Interest Historical Origination Reset Reference Prepayment CurrentType Quality Type Size LTV Year Status Rate Loss Exposure Rate Curve Rate LTV

1 U.S. Single Family Prime (FICO=760) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 1,000,000$ 2 U.S. Condo Prime (FICO=760) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 1,000,000$ 3 U.S. Single Family Subprime (FICO=660) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 1,000,000$ 4 U.S. Condo Subprime (FICO=660) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 1,000,000$ 5 U.S. Single Family Prime (FICO=760) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 1,000,000$ 5 10YR_Treasury6 U.S. Condo Prime (FICO=760) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 1,000,000$ 5 10YR_Treasury7 U.S. Single Family Subprime (FICO=660) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 1,000,000$ 5 10YR_Treasury8 U.S. Condo Subprime (FICO=660) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 1,000,000$ 5 10YR_Treasury9 U.S. Single Family Prime (FICO=760) 30 YR Fixed Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 250,000$ 10 U.S. Condo Prime (FICO=760) 30 YR Fixed Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 250,000$ 11 U.S. Single Family Subprime (FICO=660) 30 YR Fixed Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 250,000$ 12 U.S. Condo Subprime (FICO=660) 30 YR Fixed Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 250,000$ 13 U.S. Single Family Prime (FICO=760) 3/1 ARM Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 250,000$ 5 10YR_Treasury14 U.S. Condo Prime (FICO=760) 3/1 ARM Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 4 1 250,000$ 5 10YR_Treasury15 U.S. Single Family Subprime (FICO=660) 3/1 ARM Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 250,000$ 5 10YR_Treasury16 U.S. Condo Subprime (FICO=660) 3/1 ARM Conforming (<$427K) Ow ner Occupied 80 February 1, 2019 Current 6 3 250,000$ 5 10YR_Treasury17 U.S. Single Family Prime (FICO=760) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 4 1 1,000,000$ 18 U.S. Condo Prime (FICO=760) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 4 1 1,000,000$ 19 U.S. Single Family Subprime (FICO=660) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 6 3 1,000,000$ 20 U.S. Condo Subprime (FICO=660) 30 YR Fixed Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 6 3 1,000,000$ 21 U.S. Single Family Prime (FICO=760) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 4 1 1,000,000$ 5 10YR_Treasury22 U.S. Condo Prime (FICO=760) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 4 1 1,000,000$ 5 10YR_Treasury23 U.S. Single Family Subprime (FICO=660) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 6 3 1,000,000$ 5 10YR_Treasury24 U.S. Condo Subprime (FICO=660) 3/1 ARM Jumbo (>$427K) Ow ner Occupied 80 February 1, 2003 Current 6 3 1,000,000$ 5 10YR_Treasury

S S ( CO ) C f ( $ ) O O C $

Inputs Provided by Industry Working Group Banks' Own Inputs

INPUT FOR BANKS

Example: Residential Mortgages (*)

Global Credit Data

Benchmark Creation (1/2)Each bank will run the hypothetical portfolio with each CECL parameter and scenario setup, then submit intermediate and final outputs to the project team for analysis

13

INTERMEDIATE STEPS & OUTPUT

Hypothetical Portfolio

CECL Parameters

CECL ScenariosCECL Model

Suite of Individual Bank

Lifetime PD

Lifetime LGD

Lifetime EAD

ECL

Default Rates

Loss Rates

Collateral Values

Exposure Development

Maturity Time

…

FINAL RESULTBANKS’ OWN RUNS

PARTICIPATING BANKS PROJECT TEAMINDUSTRY

EXPERT GROUP

Data submissionby banks


Global Credit Data

Benchmark Creation (2/2)Banks provide their modeling outcomes for each loan in the template – 2ND Run Example


Lifetime Mean Reversion Period R&S Period Period Beyond R&S Cum. PD Cum. PD Cum. PD Cum. PD Cum. PD LGD LGD LGD LGD LGD EAD EAD EAD EAD EAD Expected Life R&S PeriodECL ECL ECL ECL Year 1 Year 2 Year 3 Year 4 Year 5 Year 1 Year 2 Year 3 Year 4 Year 5 Year 1 Year 2 Year 3 Year 4 Year 5 (in Years) (in Quarters)

OUTPUT FROM BANKS

CCAR Base ScenarioCECL PARAMETERS - 2ND RUN

Lifetime Mean Reversion Period R&S Period Period Beyond R&S Cum. PD Cum. PD Cum. PD Cum. PD Cum. PD LGD LGD LGD LGD LGD EAD EAD EAD EAD EAD Expected Life R&S PeriodECL ECL ECL ECL Year 1 Year 2 Year 3 Year 4 Year 5 Year 1 Year 2 Year 3 Year 4 Year 5 Year 1 Year 2 Year 3 Year 4 Year 5 (in Years) (in Quarters)

OUTPUT FROM BANKS

CCAR Severely Adverse ScenarioCECL PARAMETERS - 2ND RUN

ECL outputs are collected by Mean Reversion period, Reasonable and Supportable period, Beyond R&S period, and lifetime. PD, LGD, and EAD are collected for 5 years at year end. Banks’ own CECL parameter are also collected when used. In this 2nd Run example, it’s the R&S Period (last column).

2nd Run – CCAR Base Scenario

2nd Run – CCAR Severely Adverse Scenario

Global Credit Data

Secure Data Submission ProcessGCD has a recognized track record of data pooling while ensuring data quality, security, and confidentiality

15

Cooperation Results will be discussed with banks in detail

based on sound and profound audit and peer comparison reports.

Data Quality GCD performs an independent audit by comparing

data from participating banks to point out likely data collection errors, so that participating banks can fix these quickly.

GCD uses automated validations of the data fields and values in their input portal as well as in-cycle audit where a bank’s data is manually audited before aggregation, looking for biases, bad data, etc.

Data Security Data will be transmitted only by

encrypted and secured protocols and processed by a mature data portal. Our data transmission hosting platform is tailored to the security requirements of our financial services contributors.

Data Confidentiality GCD’s data collections are

based on detailed data pool regulations with strict rules on confidentiality.

GCD is in collaboration with an independent data agent bound to the same strict rules.

PRINCIPLES Banks submit their data to GCD’s

data portal https://www.globalcreditdata.net

Data portal is secured and only accessible by whitelisted IP addresses

Mature data pooling infrastructure operated and hosted by CapGemini

SUBMISSION PROCESS


Global Credit Data

Analysis & Socialization

16

Each participating bank will receive a benchmarking report comparing the different outcomes between banks. Unique to each report will be an indicator of where that particular bank (orange line in the graph) compares to the other banks for each individual hypothetical borrower.

In addition, as per GCD’s principles and procedures, banks will receive the full, anonymized data set.

Orange line: your BankBox plots: show other participants estimates per each individual borrower

HYPOTHETICALBORROWER

YOUR BANK’S ECL OTHER BANKS’ ECL

1 5bp 10bp 15bp 45bp …

2 12bp 30bp 14bp … …

…. … … … … …

ILLUSTRATIVE. Results shown from GCD’s IFRS9 peer comparison report


Global Credit Data

TimelinesThe Project team aims at completing the Industry Benchmark Study by end of July 2019, in order to assist participant banks’ CECL implementation and the first regulatory submission.

17

(Apr 16th – May 30th)

PORTFOLIO, SCENARIO, & PARAMETER DESIGN

• Design hypothetical portfolio across different instruments, risk characteristics, and other major dimensions

• Determine diversification and coverage (number of observations) of each segment

• Define CECL parameters• Integrate FRB 2019 CCAR

scenarios

METHODOLOGY SURVEY STUDY

(Completed)

• Surveyed industry readiness for CECL

• Collected banks’ CECL methodology by products and segments

• Published survey results

RESULTS GENERATION & COLLECTION

(Feb 15th – Apr 15th)

• Each bank applies its own CECL models on the industry hypothetical portfolio

• Test parameters and scenarios developed by industry working group

• Submit testing results

• Anonymize and validate data submission from banks

• Perform data quality check and output validation

• Consolidate submission and share anonymized results and peer benchmarking report with participating banks

The hypothetical portfolio, scenarios, and parameters are open to participating banks for commentary and feedbacks from now to February 14th, 2019.

PEER BENCHMARK

(Awaiting Feedback from Participating Banks)

Participating banks will receive anonymized benchmark data by end of May 2019.

ANALYSIS & SOCIALIZATION

(Jun 1st – Jul 15th)

• Analyze results submitted by banks

• Socialize preliminary analytics results to participant banks

• Deep-dive significant viability in results and conduct root-cause analysis with banks

Study Report will be published in July 2019.


Global Credit Data

Why Should Your Bank Take Part? The CECL Benchmarking Study is the only one of its kind and provides a unique, and much needed, opportunity

to gain insight in how your expected loss models benchmark with peers - neutral to your bank’s portfolio and macro-economic forecast.

The study allows banks and financial services firms to benchmark their CECL models against peers prior to full implementation.

The IFRS9 study showed large variability in ECL outcomes across banks when using the same portfolio and scenarios. In once instance, for the US and Canadian banks in the study, the largest ECL estimate was 5 times larger than the lowest.

CECL is expected to have a large impact on reserves. ALLL reserves are around 1.5% of total loans for the whole industry*.

GCD has a proven track record conducting similar studies, having conducted an IFRS9 benchmarking study in 2018.

Similarly, this study will also be based on GCD’s highly secured and mature data pooling infrastructure and associated agreements.

18

* Based on 9/30/2018 Call Reports (add loan type) for Wells Fargo, Citibank, Bank of America, and JPMorgan


Global Credit Data

Further Information For further information, please consult https://www.globalcreditdata.org/services/benchmarking-cecl-models.

The draft data templates are stored on the website for download and feedback (registration required)

Further guidelines (e.g. mapping advice etc.) will be added in the upcoming weeks

A living FAQ section will support the submission process

Or contact ...

19

Soner Tunay

Head of Quantitative Analytics, Finance & Risk Services at Accenture Consulting

[email protected]

Daniela Thakkar

Methodology & Membership Executive, Global Credit Data

[email protected]

Martin Boer

Director of Regulatory Affairs, Institute of International Finance

[email protected]


https://www.globalcreditdata.org/services/benchmarking-cecl-models




us current expected credit losses (cecl) industry benchmark … · gcd is working with accenture...

Documents