Early challenges of CECL i mplementation Hosted by: Todd Pleune, - - PowerPoint PPT Presentation

Information Classification: General

Early challenges of CECL implementation

Hosted by: Todd Pleune, Protiviti Protiviti Presenters: Xiaojing Li, CoStar Group Matthew Murphy, State Street

Information Classification: General

Xiaojing Li Director, Quantitative Methods CoStar Group Matthew Murphy VP, Lean Strategy Consultant State Street

Meet the Webinar presenters…

Moderated by: Todd Pleune Managing Director, Model Risk Managaement Protiviti

Information Classification: General

• Implementation involves many bank areas; may necessitate organizational changes
• Timeline should support “no surprises” objective for this important change

– Full-year parallel run – Incorporate lessons learned in IFRS9 implementation

• Maintaining momentum flagged by many as a top challenge

– C-suite support, with dedicated Project Management Office and Steering Committee – Leverage both internal and external working groups

Senior Management and Board Support

Important to establish and maintain appropriate “tone at the top” CECL Planning and Budgeting ALM Risk Management Scenario Forecast Reporting Regulatory Oversight Governance Data Collection and Storage

Information Classification: General

• Consider existing frameworks to minimize process differences

– Harmonize with pre-existing forecasting processes - CCAR, DFAST, IFRS9, etc. – to the extent possible – Ultimate solution should be practical, operable, well-controlled and beneficial – Must account for model bias

• Weigh trade-offs – lower complexity vs. reserve accuracy, higher complexity vs.

auditability, up-front costs vs. ongoing maintenance, higher reserve vs. higher volatility, automation vs. judgment, etc. – Solutions that require material downstream data needs may prove difficult to implement – Allow for feeder model validation and regular re-calibration

• Fundamental decisions can have long-term implications

– Reasonable and supportable period – Mean reversion techniques – “Reasonably expected” Troubled Debt Restructurings – Timing of recoveries

Initial Modeling Decisions

What loss forecasting capabilities do you have in house today?

Information Classification: General

• The sooner gaps are addressed, the quicker an institution can begin compiling historical

data − Auditors understand that some historical data may not be obtainable, but they expect a plan to be in place to correct gaps over time

• Existing data streams must be subject to data quality assessment
• Existing systems may have limitations for storing of additional static data

− Anticipate delays and resistance to changes in data capture and storage

Data Gaps

Process complicated by merger of loan origination data with Accounting and Risk data sets

Information Classification: General

• Certain asset classes – e.g., long duration securities and loans – can be problematic

when trying to measure “life of asset” expected credit losses − Reserve process expanded to include assets beyond loans − Complete balance sheet review to ensure all assets are accounted for

• Must also decide on optional use of multiple scenarios

₋ Basic assumptions must be disclosed ₋ Reliance on market consensus insufficient, as consensus is historically unreliable in period leading up to economic contraction ₋ Must also establish a framework for weighting scenarios ₋ Requires regular updating, which can be expensive

• Despite increased sophistication of loss forecasting, must maintain qualitative factor

element to model

• Stress-testing and back-testing may help identify model characteristics that may produce

unintended consequences ₋ May also help in setting investment and lending strategies ₋ Different portfolio segments will demonstrate different levels of volatility

Other Model Challenges

One size does not fit all; must maintain flexibility in approach

Information Classification: General

• Must consider sufficiency of existing governance

− Some in-scope assets covered by different teams or committees − Model sophistication may necessitate a higher level of review and challenge − Multiple scenarios adds complexity

• New process will be subject to considerably more scrutiny than existing process

− Regulators − Internal and External Auditors − Model Validation

• Must also consider specific regulatory requirements

₋ How will you ensure SOX compliance? ₋ Policies and procedures should be clear on roles and responsibilities of all stakeholders

• Greater automation will allow for more frequent, and timely, loss estimations

₋ Monthly estimation will help minimize quarter-end surprises

Process Controls

Higher data requirements necessitate “production environment” approach

Information Classification: General

• Disclosures in advance of adoption (SAB74)
• Ongoing qualitative elements

− How loss estimates are developed − Factors that influence estimate - past events, current conditions, forecast(s), etc. − Risk characteristics of each portfolio segment − Changes in policies and impact of same − Asset purchase and sales

• Roll-forward of allowance
• Vintage (presenting amortized cost by year of origination)
• Past dues and non-accruals
• Collateral Dependent Financial Assets
• Difficult to assess market standard for disclosure prior to adoption

₋ Markets may demand more than what you have prepared for disclosure ₋ Should consider timing of when disclosures must be available

Disclosure Requirements

Not a last priority

Information Classification: General

The ASU requires enhanced disclosures to help investors and other financial statement users better understand significant estimates and judgments used in estimating credit losses, as well as the credit quality and underwriting standards of an organization’s portfolio.

CoStar Risk Analytics

Early Challenges of CECL Implementation

• gather relevant historical data
• Identify data gap in historical

information

• Identify loss drivers based on

historical experience

• identify data gap in current

information

• build an efficient data collecting and

updating system

• select macro-variables
• select local market variables
• build a forecast model for loss drivers
• build a forecast model for losses
• full transparency
• automatic report
• Interactive review

The ASU requires an organization to measure all expected credit losses for financial assets held at the reporting date based on historical experience, current conditions, and reasonable and supportable forecasts. Financial institutions and other organizations will now use forward-looking information to better inform their credit loss estimates.

Source: http://www.fasb.org/jsp/FASB/FASBContent_C /NewsPage&cid=1176168232900

Historical Experience Current Conditions Reasonable and Supportable Forecasts Enhanced Disclosure

Information Classification: General

CoStar Risk Analytics

CECL Road Map – Data is the Key

Current Expected Credit Loss Loss Driver Forecast Macroeconomic Forecast

Historical Loan Data Historical Loss Data Credit Risk Model Qualitative Adjustmen t Data System Documentatio n Model Governance Reporting & Communicating Current Loan Data Economi c Data Market & Segment Data Econometri c Model Scenari

• Analysis

Overcome the challenges down the road

Information Classification: General

CoStar Risk Analytics

Historical Experience

q Historical Loan & Collateral Data

• Loan current balance
• Origination date
• Origination term
• Remaining term
• Remaining Amortization term
• Remaining IO term
• Coupon rate
• Rate reset
• DSCR (origination & contemporaneous)
• LTV (origination & contemporaneous)
• Prepayment penalty structure
• Re-underwriting parameters
• Credit enhancement
• Property address
• Property type
• Multi-property or cross-collateralized
• Property value
• Net operating income
• Cap rate
• Tenants
• Leases
• ……
• Delinquency history
• Timing of default
• Default type
• Outstanding balance at default
• Workout or modification
• Reinstate Status
• Timing of loss
• Loss type (FC/REO/DPO/Note sale…)
• Outstanding balance at loss occurrence
• Liquidation expense
• Liquidation proceeds
• Prepay activity

q Historical Loan Default & Loss Data

Historical Experience Gather relevant historical data and identify data gaps in historical information

Information Classification: General

CoStar Risk Analytics

Historical Experience

q Variable Selection Process

§ Data availability § Significant relationship with credit risk and losses § Improve model performance (accuracy and discrimination power)

65% 67% 69% 71% 73% 75% 77% 79% 81% 83% 85% Apartment Office Retail Warehouse Hotel Incremental from Vintage Incremental from Loan Size Incremental from Region Incremental from IO Type Incremental from Seasoning ROC Power from LTV & DSCR

ROC (Receiver Operating Characteristics) Attribution

Developing A Sound Model with Continuous Performance Tracking

Source: CoStar Risk Analytics (Asof 2017Q4)

Identify loss drivers based on historical experience

Information Classification: General

CoStar Risk Analytics

Current Conditions

Time-varying Information: Loan, Collateral and Market

Actual Loan officer document

Estimated based on Current pool average

Calculated based on Other Info of Same Loan

Modeled based on Historical Experience

Undocumented Assumptions

Improving Data Updating Process Identify data gaps in current conditions

Information Classification: General

CoStar Risk Analytics

Current Conditions

Improving Data Updating Process

q A system to fill missing information and check available current information

§ Embedded calculation to derive missing values from available values § Embedded historical market/segment information § Embedded forecasted market/segment information § Flexible options for expertise’s

• verride

§ Exportable Configurations § Automated documentation

Build an efficient data collecting and updating system

Information Classification: General

CoStar Risk Analytics

Reasonable and Supportable Forecasts

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual History

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual History Modeled from GDP

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual History Modeled from GDP & UR

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual History Modeled from GDP & UR & BBB

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual Modeled from GDP & UR & BBB & HPI

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

40 60 80 100 120 140 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2 Actual Modeled by GDP & UR & BBB & HPI & CPI

• 40%
• 30%
• 20%
• 10%

0% 10% 20% 30% 2000Q1 2000Q4 2001Q3 2002Q2 2003Q1 2003Q4 2004Q3 2005Q2 2006Q1 2006Q4 2007Q3 2008Q2 2009Q1 2009Q4 2010Q3 2011Q2 2012Q1 2012Q4 2013Q3 2014Q2 2015Q1 2015Q4 2016Q3 2017Q2

CRE Price Index CRE Price Annual Growth

Building the Multivariate Regression Model

Source: CoStar Risk Analytics, FRB (Asof 2017Q4)

Select macro variables for loss driver forecast

Information Classification: General

Information Classification: General

CoStar Risk Analytics

Reasonable and Supportable Forecasts

q Simplified Model vs. Advanced Econometric Model System

§ Build a single equation forecast model with limited data and resources § Less data requirement § Less cost § Build a multi-equation forecast system with richer data and more resources § More transparency § More outputs CRE Price

Macroeconomic Factors Local Market Factors

Single Equation Forecast Model Construction Starts Supply of Space Demand for Space Vacancy Rent NOI Cap Rate CRE Price

Macroeconomic Factors Local Market Factors

Multi-equation Forecast System

Choosing the Appropriate Model Structure

Loss Driver

Build a sound model to project loss drivers

Information Classification: General

CoStar Risk Analytics

Reasonable and Supportable Forecasts

q Variable Selection Process

§ Data availability § Significant relationship with credit risk and/or losses § Improve model performance (accuracy and discrimination power) Loan Characteristics Collateral Characteristics Market Conditions Competing Risk Intangible Factors

q Model Selection Process

§ Logit / Probit regression model § Duration model/Survival analysis § Hazard rate model § Option-based model § Vintage model § Roll rate method § Transition matrix § Hybrid model

q Model Validation Process

§ In-sample fitting § Out-of-sample backtesting § By vintage § By calendar year § By asset type § ROC/KS/CAP

Build a sound model to project losses based on loss driver forecast

Information Classification: General

CoStar Risk Analytics

Enhanced Disclosure

q Data & Infrastructure

§ Asset credit quality § Underwriting standards § Data gathering process and Quality Assurance § Data warehouse system

q Model Methodology

§ Calibration dataset § Validation dataset § Model methodology & assumptions § Qualitative adjustments § Backtesting result

q Model risk management & governance control

§ Model limitations § Model performance monitor Improving Documentation and Reporting Process Create documentation with full transparency for effective communication

Information Classification: General

CoStar Risk Analytics

Enhanced Disclosure

Build automatic and dynamic report & review system

Information Classification: General

Live Q&A Please submit your questions via the question function on the right hand side.

Xiaojing Li Director, Quantitative Methods CoStar Group Moderated by: Todd Pleune Managing Director, Model Risk Management Protiviti Matthew Murphy VP, Lean Strategy Consultant State Street

Information Classification: General

Thank you for attending

You may also be interested in…

Americas largest risk and regulation convention of its kind… returns for 2018 bigger and better than ever. View full speaker line-up and agenda at www.risk- americas.com

Information Classification: General

Continue the discussion in our brand new Facebook Community… We have just started a Facebook group exclusively for thought-

• leadership. Our Facebook group will now be the first to receive the

newest articles and issues of our freely available magazine. Follow our page and help us grow… Have you got your complimentary copy of the latest Issue of Risk Insights Magazine? Issue Seven is a 60+ page financial risk and regulation publication. Written by the industry, highlighting the latest changes impacting the risk professional today. Get your free copy at www.cefpro.com/magazine