capital adequacy stress tests: pre-provision net revenue and scenario design

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Dial-in Details

Welcome to the CRISIL web conference on “Capital Adequacy Stress Tests: Pre-Provision Net Revenue and Scenario Design”. We will commence the session shortly.

Please join the audio conference by dialing the number relevant to your location.

1

Toll Free NumbersUSA 1 866 746 2133UK 0 808 101 1573Singapore 800 101 2045Hong Kong 800 964 448India 1 800 209 1221Australia 1 800 053 698Poland 00 800 112 4248Germany 800 142 43444UAE 800 017 5282Argentina 0 800 142 43444China - North 10 800 713 1853China - South 10 800 130 1814South Korea 800 1424 3444Canada 800 1424 3444Switzerland 800 564 911Netherlands 0 800 022 9808

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Gurpreet Chhatwal,

Global Head of Risk & Analytics, CRISIL Global Research & Analytics

PPNR Modeling & Scenario Development

Capital Adequacy Stress Tests

March 2014

Speakers:

Joshua Hancher,

Senior Consultant, Risk and Analytics, CRISIL Global Research & Analytics

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

CRISIL GR&A Summary

3

NASSCOM® Exemplary Talent Awards: 2012 (The Talent

Magnets), 2011(Skill Enhancement Initiatives)

Top Research Outsourcer for Financial Services Industry

Analytics, 2006, 2007 and 2009

'Top Green IT Enterprise Award 2012' from CIO Magazine

Roopa Kudva named one of the 20 most powerful women in Business

by Fortune India in 2011

'Best Leadership Training Programme' award hosted by World HRD Congress, 2012

Awards & RecognitionsAwards & Recognitions

Roopa Kudva named one of the 30 most powerful women in

Indian business by Business Today magazine, for the third

time 2009-2011

Mumbai Corporate office - LEED certified “GREEN BUILDING" 2010

Global Leader inResearch & Analytics Space

Global Leader inResearch & Analytics Space

Regulated Entity with Strong Compliance Culture

Regulated Entity with Strong Compliance Culture

High QualitySkill SetsHigh QualitySkill Sets

Rigorous &ScalableProcesses

Rigorous &ScalableProcesses

Unmatched Financial Business Strength

Unmatched Financial Business Strength

Partnerships with Expert Networks

Partnerships with Expert Networks

25 year track record Pioneered Global Research &

Analytics Team of more than 2,300 analysts

25 year track record Pioneered Global Research &

Analytics Team of more than 2,300 analysts

Serve 12 of the top 15 global investment banks

Serve 2 of the top 10 global consulting groups Serve 3 of the top 15 global insurance

companies

Serve 12 of the top 15 global investment banks

Serve 2 of the top 10 global consulting groups Serve 3 of the top 15 global insurance

companies

Regulated by SEBI, an SEC-equivalent

The only regulated entity in this space

Regulated by SEBI, an SEC-equivalent

The only regulated entity in this space

Highest standards pertaining to regulatory requirements, compliance, and confidentiality

Highest standards pertaining to regulatory requirements, compliance, and confidentiality

Strong project management capabilities

Quality Assurance and Governance Information Security

Strong project management capabilities

Quality Assurance and Governance Information Security

Robust training programmes Business Continuity Inbuilt processes and tools to enhance

productivity

Robust training programmes Business Continuity Inbuilt processes and tools to enhance

productivity

Well-diversified business Strong balance sheet and zero debt Market reputation and global reach

Well-diversified business Strong balance sheet and zero debt Market reputation and global reach

30% CAGR over the past 10 years Consistent growth across business cycles

that provides stability to deliver high quality services to our clients

30% CAGR over the past 10 years Consistent growth across business cycles

that provides stability to deliver high quality services to our clients

Recognized for high quality service – Our tagline “The best people to work with” is a verbatim reproduction of what a client said in a survey

Highly experienced and stable management team to ensure high-touch engagement Large talent pool comprising MBAs/CAs/CFA/FRMs/PhDs and other university graduates

Recognized for high quality service – Our tagline “The best people to work with” is a verbatim reproduction of what a client said in a survey

Highly experienced and stable management team to ensure high-touch engagement Large talent pool comprising MBAs/CAs/CFA/FRMs/PhDs and other university graduates

CRISIL GR&A joined the FINCAD Alliance Program in order to leverage various FINCAD Analytics tool that meets the derivatives and fixed income needs of clients. FINCAD is the market leader for innovative derivatives solutions.

CRISIL GR&A joined the FINCAD Alliance Program in order to leverage various FINCAD Analytics tool that meets the derivatives and fixed income needs of clients. FINCAD is the market leader for innovative derivatives solutions.

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Executive Summary

PPNR projections have assumed critical importance since the initial Fed stress tests

One of the most common themes for PPNR modeling improvement is integration with balance sheet and credit projections

Bottom-up model approaches have become crucial to achieve integration and robust model specification

Shifts in regulatory focus relating to scenario development have brought into question many existing industry practices on variable coverage and the incorporation of bank-specific/idiosyncratic risk

4

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Agenda

Pre-Provision Net Revenue Modeling– Regulatory Landscape– What is PPNR?– Walkthrough PPNR components

• Balance Sheet Forecasting• Net Interest Income• Noninterest Income & Expense

– Qualitative Adjustments to Model Output– Documentation of Stress Test Results

Scenario Development– Regulatory Landscape– UK versus US Stress Tests– Expansion of Supervisory Scenarios– Internal Scenario Development

CRISIL GR&A Case Studies– C&I Loan Balance Forecast Model– Fair Value of Loans Held-for-Sale

5

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Pre-Provision Net Revenue – Regulatory Landscape

6

Linkage between noninterest income & expense, balance sheet projections, and credit forecasts

Emphasis on bottom-up forecast methodologies Divergence in budgeting / strategic planning and stress test forecast methods Effective documentation of methodologies and results

Linkage between noninterest income & expense, balance sheet projections, and credit forecasts

Emphasis on bottom-up forecast methodologies Divergence in budgeting / strategic planning and stress test forecast methods Effective documentation of methodologies and results

PPNR modeling has gained in relative importance, but remains the most difficult modeled component of enterprise stress tests

CRISIL GR&A believes regulators are looking for:

1. Before Enterprise Stress Tests Budgeting and strategic planning processes Reliance on expert judgment

2. Initial Stress Tests (SCAP 2009) Banks must project capital based on

adverse economic scenarios Primary focus is credit quality

3. CCAR 2011-2014/CapPR 2011-2013 PPNR gains in relative importance

− Credit quality stabilizes− Investor emphasis on approval of

Planned Capital Actions

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

What is Pre-Provision Net Revenue?

7

In short, it is a lot of things…Ea

rnin

g As

set Y

ield

Earn

ing

Asse

t Yie

ldLi

abilit

y R

ates

Liab

ility

Rat

es

Net

Inte

rest

Mar

gin

Net

Inte

rest

Mar

gin

Other Consumer

Commercial & Industrial

Commercial Real Estate

International

Residential Mortgage

Auto Loans

Student Loans

Loans & LeasesLoans & Leases

Credit Cards

Deposits & Interest Bearing Liabilities

Deposits & Interest Bearing Liabilities

Customer Deposits

Short-term Debt and Borrowings

Investment SecuritiesInvestment Securities

(+)

=

Total Earning AssetsTotal Earning Assets

Loan Fees

Syndication Fees

Capital Markets Fees

Fair Value of HFS Assets

Credit Card Fees

Fiduciary Income

COLI

Balance Sheet-Related Fee IncomeBalance Sheet-Related Fee Income

VentureCapital

Additional Noninterest IncomeAdditional Noninterest Income

OREO & Collections

Environmental ExpenseEnvironmental Expense

FDIC Assessment

Specific Operational Risk

Goodwill Impairment

Salaries

Personnel Related ExpensePersonnel Related Expense

Employee Benefits

Relocation & Recruiting

Meetings & Training

Software & Equipment

Depreciation OperatingLease Expense

Occupancy Expense

Other ExpenseOther Expense

Net Interest IncomeNet Interest Income Noninterest IncomeNoninterest Income Noninterest ExpenseNoninterest Expense(-)

MSRValuation

Deposit Service Charges

(+)

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Balance Sheet Projections

8


Net InterestIncome

Noninterest Income / Expense

Documentation of Results

Qualitative Adjustments

Modeling Approach

Economic TheoryEconomic Theory

Banks serve as credit intermediaries to businesses and households− Commercial credit used primarily to finance investment− Consumer credit used primarily to finance consumption and residential investment

Banks serve as credit intermediaries to businesses and households− Commercial credit used primarily to finance investment− Consumer credit used primarily to finance consumption and residential investment

Variable SelectionVariable Selection

Common economic variables provided in regulatory scenarios Additional variable needs may arise to capture aggregate demand for loanable funds and supply side

effects of credit availability in stress scenarios

Common economic variables provided in regulatory scenarios Additional variable needs may arise to capture aggregate demand for loanable funds and supply side

effects of credit availability in stress scenarios

Modeling TechniquesModeling Techniques

Panel Regression: Easily understood and straight-forward scenario results Vector Auto Regression: Stands up well in both benign and stress periods ARIMA: Enables modeling of nominal terms instead of growth rates by addressing non-stationarity

Panel Regression: Easily understood and straight-forward scenario results Vector Auto Regression: Stands up well in both benign and stress periods ARIMA: Enables modeling of nominal terms instead of growth rates by addressing non-stationarity

Modeling new originations versus loan balances Linkage with other stress test processes Liquidity stress events Stress testing versus budgeting process

Modeling new originations versus loan balances Linkage with other stress test processes Liquidity stress events Stress testing versus budgeting process

CRISIL GR&A sees the key challenges as:CRISIL GR&A sees the key challenges as:

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Granular pricing assumptions commensurate with credit risk More defendable loan balance projections Explicit volume assumptions allow for direct modeling of related noninterest income items

Granular pricing assumptions commensurate with credit risk More defendable loan balance projections Explicit volume assumptions allow for direct modeling of related noninterest income items


9

Most banks lack historical new origination data Existing cash flow systems geared toward “balance targeting” Budgeting and strategic planning groups more accustomed to ending loan balances

Most banks lack historical new origination data Existing cash flow systems geared toward “balance targeting” Budgeting and strategic planning groups more accustomed to ending loan balances

Modeled ApproachModeled Approach

New Originations

Distinct Pricing (Spread to Index)

Obligor Risk Rating

Interest Income

Expected Loss, NPLs

Credit Projections

Pre-Provision Net Revenue

Commensurate with


Net InterestIncome




Modeling of new originations preferred…Modeling of new originations preferred…

…but remains a long-term solution…but remains a long-term solution

New LoanOriginations Balance

Beginning Balance t0 Ending Balance t‐1 Ending Balance t‐1plus: New Originations t0 Modeled Calculated

less: Contractual Payments t0 Modeled Modeled

less: Prepayments t0 Modeled Modeled

less: Credit Losses t0 Modeled Modeled

plus: Change in Utilization t0 Modeled Modeled

Ending Balance t0 Calculated Modeled

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Net Interest Income

10

Justifying ability to maintain a given loan spread in stress scenarios Granularity of portfolio segmentation Consistency of spread to index with assumed credit quality for new originations

Justifying ability to maintain a given loan spread in stress scenarios Granularity of portfolio segmentation Consistency of spread to index with assumed credit quality for new originations

Net Interest Income =

∑ (Balance Earning Asset i ) * (Index + Spread) * (days / interest day count) –

∑ (Balance Interest Bearing Liability i ) * (Index + Spread) * (days / interest day count)

Modeling needs are two-fold:1. Interpolation of yield curves and credit spreads provided in supervisory scenarios2. Modeling of spread to index and deposit pricing beta models


Net InterestIncome




Net Interest Income Calculation Interest rates and credit spreads applied to earning asset and interest-bearing liabilities to

derive net interest income


© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Noninterest Income & Expense

11


Net InterestIncome




Lack of correlation with macroeconomic variables Econometric analysis not suitable for all line items Linkage between balance sheet and credit forecasts Varying data availability for non-GAAP/IFRS information

Lack of correlation with macroeconomic variables Econometric analysis not suitable for all line items Linkage between balance sheet and credit forecasts Varying data availability for non-GAAP/IFRS information


Modeling Approach

Economic TheoryEconomic Theory

Banks generate higher levels of net revenues during periods of strong economic growth through greater lending activities, fiduciary income and lower environmental expense– Fiduciary & capital markets fees correlated with business investment and market indices– Fewer FTEs required to support lending activities during periods of weak economic growth

Banks generate higher levels of net revenues during periods of strong economic growth through greater lending activities, fiduciary income and lower environmental expense– Fiduciary & capital markets fees correlated with business investment and market indices– Fewer FTEs required to support lending activities during periods of weak economic growth

Variable SelectionVariable Selection

Common economic variables provided in regulatory scenarios Additional variables needed to capture the relationship between banking activities and economic cycles Some components are largely balance sheet and credit driven

Common economic variables provided in regulatory scenarios Additional variables needed to capture the relationship between banking activities and economic cycles Some components are largely balance sheet and credit driven

Modeling TechniquesModeling Techniques

Quantile Regression – Well-suited for capturing regime change during economic stress Generalized Linear Models – Useful for noninterest expense items that display bimodal distributions ARIMA – Enables modeling of nominal terms instead of growth rates by addressing nonstationarity Linear Regression – Can be ideal for generating Betas as a starting point for expert judgment

Quantile Regression – Well-suited for capturing regime change during economic stress Generalized Linear Models – Useful for noninterest expense items that display bimodal distributions ARIMA – Enables modeling of nominal terms instead of growth rates by addressing nonstationarity Linear Regression – Can be ideal for generating Betas as a starting point for expert judgment

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

12

Qualitative Adjustments to Model Output


Net InterestIncome




Expert judgment should be used to incorporate forward-looking business intelligence and address model weaknesses and limitations.

Strong PracticeForecasts driven by quantitative models

with expert judgment when necessary

Weak Practice:Sole reliance on

model output

Weak Practice:Sole reliance onexpert judgment

Some acceptable ways to size qualitative adjustments include– Out of sample backtesting and challenger models– Confidence intervals– Quantification of enterprise-wide sensitivity to model assumptions

Use of qualitative adjustments and associated governance framework must be abundantly documented

Some acceptable ways to size qualitative adjustments include– Out of sample backtesting and challenger models– Confidence intervals– Quantification of enterprise-wide sensitivity to model assumptions

Use of qualitative adjustments and associated governance framework must be abundantly documented

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.


13

Inadequate documentation was cited by the Fed as one of the most prevalent problems in PPNR submissions across all banks

Documentation should be concise and provide relevant information for credible challenge and review

– Clear delineation between model output and qualitative adjustments

– Logic and economic rationale for variable selection clearly explained

– Inundating regulators with information is never a good strategy

Results versus Model Development Documentation – there is a big difference.

FR Y-14A reporting templates versus modeled account line structure – must reconcile the two.

Inadequate documentation was cited by the Fed as one of the most prevalent problems in PPNR submissions across all banks

Documentation should be concise and provide relevant information for credible challenge and review

– Clear delineation between model output and qualitative adjustments

– Logic and economic rationale for variable selection clearly explained

– Inundating regulators with information is never a good strategy

Results versus Model Development Documentation – there is a big difference.

FR Y-14A reporting templates versus modeled account line structure – must reconcile the two.

Stress Tests are like standardized tests:there is a math and written portion of the exam


Net InterestIncome




© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Scenario Development – Regulatory Landscape

14

Major learnings on scenario development1. Additional variables are needed to build robust models2. Sole reliance on macroeconomic scenarios does not adequately stress the risk

profiles of all bank holding companies

Major learnings on scenario development1. Additional variables are needed to build robust models2. Sole reliance on macroeconomic scenarios does not adequately stress the risk

profiles of all bank holding companies

These learnings are reflected in new PRA and EBA stress test designThese learnings are reflected in new PRA and EBA stress test design

Greater expectations for variable selection beyond core set provided in supervisory scenarios

Use of standard off-the-shelf scenarios not suitable for capturing bank-specific risks and vulnerabilities

Robust risk-identification processes to feed idiosyncratic scenario development

Greater expectations for variable selection beyond core set provided in supervisory scenarios

Use of standard off-the-shelf scenarios not suitable for capturing bank-specific risks and vulnerabilities

Robust risk-identification processes to feed idiosyncratic scenario development

CRISIL GR&A believes regulators are looking for:CRISIL GR&A believes regulators are looking for:

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Both Stress Testing Frameworks Require Two Sets of Scenarios

Scenario Development – UK versus US Stress Tests

15

The final shape of UK stress tests is yet to be finalized, but the proposed framework has more similarities than differences with US CCAR

requirements – particularly with respect to scenario design.

Risk Factors Included in Common and Internal Scenarios

Internally Generated Bespoke

ScenariosSupervisory

Baseline & Stress Scenarios

Bank-specific/ Idiosyncratic risk

Bank-specific/ Idiosyncratic risk

Systemic Risk

Systemic Risk

Total Risk Exposure

=Systemic

Risk+

Idiosyncratic Risk

Total Risk Exposure

=Systemic

Risk+

Idiosyncratic Risk

Supervisory Stress Scenarios� Applied across all bank participants� Additional variables often required

Supervisory Stress Scenarios� Applied across all bank participants� Additional variables often required

Internally generated scenarios� Designed individually by each bank� Typically result in greater capital

depletion than supervisory scenarios

Internally generated scenarios� Designed individually by each bank� Typically result in greater capital

depletion than supervisory scenarios

1 2

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Modeling the components of aggregate variables like GDP (provided in supervisory scenarios) commonly needed

Geographic and industry-specific variables may improve model performance and also incorporate idiosyncratic risk

Most variables can be forecast using structural and agnostic models

Modeling the components of aggregate variables like GDP (provided in supervisory scenarios) commonly needed

Geographic and industry-specific variables may improve model performance and also incorporate idiosyncratic risk

Most variables can be forecast using structural and agnostic models

Identification & Forecasting of Additional Variables Identification & Forecasting of Additional Variables

Addressing Scenario Development Issues

16

Univariate ModelsUnivariate Models

ARMA, ARIMA, ARCH/GARCH, Interpolation, Exponential Smoothing Models with/without seasonality Provide easily interpretable equations and useful for short term forecasting Fail to provide an explanation of the causal structure behind the evolution of a time series E.g. Volatility, stock price, interest rates, market illiquidity could be forecasted/filled-in using such methods

Multivariate ModelsMultivariate Models

Multiple Regression, Logit and Probit Model, VAR/VECM, Generalised Linear Model, ARIMAX, MultivariateGARCH, Copula, Factor Analysis, Dynamic Stochastic Models

Exploit the relationship between the dependent variable and several independent variables and their lags Framework provides a systematic way to capture rich dynamics in multiple time series

Machine Learning Models

Machine Learning Models

Artificial Neural Networks, Support Vector Machines, Random Forests, Decision Trees, Cluster Analysis Used for forecasting and classification purposes primarily Exploits the non-linearity and nonlinear relationships among the time series, requires huge amount of data

for training purposes

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

(15%)

(10%)

(5%)

0%

5%

10%

4Q07

1Q08

2Q08

3Q08

4Q08

1Q09

2Q09

3Q09

4Q09

1Q10

2Q10

3Q10

4Q10

1Q11

2Q11

3Q11

4Q11

1Q12

2Q12

3Q12

Quarterly % Change

C&I Loan ModelOut of sample backtest (2008‐2013)

Recession Dates Actual Champion Challenger

Case Study 1Forecasting C&I Loan Balances

17

Champion Model

Independent Variables Coefficient T-stats

Constant 0.0170 1.50

SPREAD1 -0.0097 -2.25

US Real GDP (Lag 2) 0.0040 1.70

Dow Jones (Lag 5) 0.1467 2.44

Adjusted R^2 = 0.30 AIC = 3.93

Log-Likelihood = 102.50 Schwarz Criteria = 3.78

F-Statistics = 8.01 Durbin-Watson Stats = 1.91

Champion Model output closely tracked actual C&I loan growth in a sample backtest High model fit (Mean Average Error = 0.0244) Directionally correct (Hit ratio = 79%)

Model benchmarking shows that the champion model has been developed with adequate variable coverage to inform C&I model coefficients that result in reasonable projections

Champion Model output closely tracked actual C&I loan growth in a sample backtest High model fit (Mean Average Error = 0.0244) Directionally correct (Hit ratio = 79%)

Model benchmarking shows that the champion model has been developed with adequate variable coverage to inform C&I model coefficients that result in reasonable projections

Macroeconomic drivers from the Fed’s supervisory scenarios were used to model commercial loan balances as part of the PPNR modeling process

Vector Auto Regression was the chosen methodology for its ease of use and suitability for forecasting both benign and stress scenarios

Challenger Model was also developed using supplemental macroeconomic variables to help ensure the reasonableness of the primary model output

Macroeconomic drivers from the Fed’s supervisory scenarios were used to model commercial loan balances as part of the PPNR modeling process

Vector Auto Regression was the chosen methodology for its ease of use and suitability for forecasting both benign and stress scenarios

Challenger Model was also developed using supplemental macroeconomic variables to help ensure the reasonableness of the primary model output

CRISIL GR&A Approach Insights

© 2

014

CR

ISIL

Ltd

. All

right

s re

serv

ed.

Case Study 2Fair Value of Loans Held-for-Sale (HFS)

18

Macro to Micro Model

Significantly reduced market risk-related losses by separately taking into account underlying exposure and credit spread indices (i.e., bottoms-up approach)

Significantly reduced market risk-related losses by separately taking into account underlying exposure and credit spread indices (i.e., bottoms-up approach)

Bank had a substantial loan warehouse subject to fair value accounting

Top-down approach was previously used by modeling dollar losses on the loan warehouse

Bank had a substantial loan warehouse subject to fair value accounting

Top-down approach was previously used by modeling dollar losses on the loan warehouse

Macro-to-micro model was developed to forecast credit index spreads widely used by the market to price and hedge CMBS loans

Multiple Regression was used to model CMBS spreads using the Fed’s macroeconomic scenario variables

Noninterest income impact was then calculated using macro-to-micro model output to value warehouse loans in supervisory base, adverse, and severely adverse scenarios

Macro-to-micro model was developed to forecast credit index spreads widely used by the market to price and hedge CMBS loans

Multiple Regression was used to model CMBS spreads using the Fed’s macroeconomic scenario variables

Noninterest income impact was then calculated using macro-to-micro model output to value warehouse loans in supervisory base, adverse, and severely adverse scenarios

Fair Value Gain/Loss:Loans HFS

2.0 2.2 2.4 2.6 2.8 3.0 3.2

3Q13

A

4Q13

F

1Q14

F

2Q14

F

3Q14

F

4Q14

F

1Q15

F

2Q15

F

3Q15

F

4Q15

F

Cred

it Spreads

Base Adverse Severely Adverse

CMBS Credit SpreadsCoefficient T-Stats

Intercept .73 2.66Market Volatility Index (Qtrly % Change) .03 3.61Commercial Real Estate Price Index (Qtrly % Change) (2.94) (1.26)

Background

Client Impact

CRISIL GR&A Approach

capital adequacy stress tests: pre-provision net revenue and scenario design

Economy & Finance