Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 49

1

Financial System Stress Tests





Seminar on Financial Stability
and Development
2
Overview
I. Stress tests in FSAPs

II. Stress tests in central banks & banking supervisory
agencies

III. Selected methodological issues

IV. Recent developments in stress test methodology
3
Stress Tests in FSAPs
Part I
4
Defining Stress Tests





Stress tests = a set of statistical techniques to help assess the
vulnerability of financial institutions & financial systems to
exceptional but plausible events

Defining exceptional but plausible

Against a specific historical scenario
Hypothetical scenario based on the analysis of past
volatility and correlations
Other methods
5
Types of Stress Tests
By aggregation

Individual exposures

Individual institutions

System-wide

on bank by bank* data
(bottom up)

on aggregate data
(top down)
By methodology

Sensitivity analysis

Scenario analysis

Contagion analysis
* Most system-wide stress tests so far have been on banks. Thus, focus on
banks and the banking sector here. Nonetheless, the issue of Stress tests for
nonbank financial institutions will also be discussed.
6
Recent Experience with Stress Tests





Financial crisis of late 1990s

More focus on financial sector risks
Balance sheet approach

Increasing interest in stress tests as a result

Financial institutions (BIS surveys in 2000 & 2004)
Supervisory agencies & central banks
International institutions: IMF, World Bank
7
Stress Tests in the FSAP





Stress tests a crucial tool in the quantitative assessments in
FSAPs

Stress tests tailored to country-specific circumstances
(complexity of the financial system, data availability)
8
Stress Tests as a Multi-Stage Process





1. Identify macroeconomic & market risks

2. Identify major exposures

3. Define coverage

4. Identify data required

5. Calibrate shocks (or scenarios)

6. Select & implement methodology

7. Interpret results
9
Stress Tests vs. Other Analytical Tools





Advantages of stress tests

Forward-looking assessment of risks
More precise than analyzing aggregated indicators
(NPLs, open positions)

Disadvantages of stress tests

Robustness with respect to assumptions?
Banks viewed as static portfolios rather than as dynamic units
Role of non-quantitative factors?
10
Other Analytical Methods
Complement Stress Tests





Macroprudential analysis

Financial Soundness Indicators (FSIs)
Links between FSIs and other factors, especially the macroeconomic
framework

Analysis of qualitative information

Legal, regulatory, accounting, tax conditions
Corporate governance
11
Evolving Role of Stress Tests in FSAP





Most missions: single-factor sensitivity analysis based on
historical extremes

More recent FSAP missions

More focus on scenario analysis

Active involvement of the authorities

Industrial countries: bank internal stress test models, authorities
macroeconomic models, inclusion of interbank contagion risk

Inclusion of nonbank financial institutions in some cases

12
Stress Tests in Central Banks &
Banking Supervisory Agencies
Part II
13
Role of Stress Tests
Help focus supervisory processes on a risk basis

Support macroprudential analysis at the central banks
(including in financial stability reports)

Assess effects of prospective policy changes
14
Stress Tests Conducted by
Central Banks & Banking Supervisors
Bank-by-bank stress tests conducted by supervisors

Financial system stability reports include or refer to
aggregate stress tests

Range of approaches examples:

Hungary (focus on sensitivity calculations)
Norway (focus on sources of credit risk)
15
Implementation Issues



1. Coverage institutions

2. Coverage exposures

3. Methodology

4. Organization



16
Coverage - Institutions



From systemic stability perspective, important to cover a
significant part of the system, but not all institutions

But for supervisory purposes, it might be useful to increase the
coverage

Coverage of foreign banks (branches, subsidiaries, exposures
to foreign banks)

Consolidation (consistency across time, inclusion of non-bank
financial institutions)

17
Coverage - Exposures



Basic exposures to be included

Credit risk (incl. indirect FX & interest rate risks)
Direct interest rate risk
Direct FX risk

Other exposures to be considered

Interbank contagion risk
Equity and/or real estate price risk
Later: liquidity, concentration risks

Include on- & off-balance sheet exposures
18
Methodology



Both sensitivity and scenario analysis

Scenario analysis: a combination of historical and hypothetical
scenarios

For the hypothetical scenarios, justification using
macroeconomic models (see later)

A mix of top down & bottom up

Factors: computation difficulty, confidentiality
Most calculations typically bank-by-bank, but model estimates on
aggregate basis

19
Methodology



Expressing the impact: in terms of capital adequacy (pre-shock
vs. post-shock); in terms of liquidity for liquidity stress test

Impact measured over what period?

Typically one year (ideally, NPV of future impact)

Treatment of profits

A prudent assumption based on the past
An autonomous shock to non-interest income or net interest income (on
top of other shocks)
20
Organization
How frequently?

Standard set of tests on quarterly basis, market risk/sensitivity
analysis more frequently, elaborate analysis (e.g. contagion) less
frequently

Run by whom?

Which software?

Many start with Excel and E-Views, then integrate with
supervisory information systems

Presentation, dissemination of the results

By bank (supervision; links with EWS)
By peer groups (macroprudential surveillance)

21
Selected Methodological Issues
Part III
22
Selected Methodological Issues



1. Selecting macroeconomic scenarios

2. Foreign exchange (FX) risk

3. Interest rate risk

4. Credit risk

5. Interbank contagion risk

6. Liquidity risk

7. Equity price & real estate price risk
23
Macro Scenarios for Stress Tests
Historical scenarios

e.g. the 1997 turbulence and subsequent slowdown in East Asia

Hypothetical scenarios

Recognizing the limitations of macro models, especially for large
shocks, would it be possible to use the central banks existing macro
model?

Stochastic simulations based on the model?

Scenario design: relative sizes of shocks to the risk factors
Assessing likelihood of the scenarios

24
Worst Case vs. Threshold Approach






Shock to risk
factor 2
Shock to risk
factor 1
CAR=0%
CAR=8%
CAR=10%
p=1%
p=2% p=5%
A
B
C
25
Controversy on Probability





Stress test scenarios should not be attached any probability
measure, ST is different from the standard risk-modeling
toolkit (e.g., Kupiec, 2001)

Stress test scenarios need to be attached an explicit
probability measure; otherwise they are useless (e.g.,
Berkowitz, 1999)
26
Controversy on Probability





Practical solutions

Base stress test on extreme historical scenario

Relate stress test to extreme historical scenario

Use threshold approach with extreme thresholds
(useful when exposures negligible: proof by contradiction)

Combine historical scenario with hypothetical assumptions
(robustness analysis)
27
Foreign Exchange Risk
The standard sensitivity analysis



Note: C=capital, A
RW
=risk-weighted assets, F=net open position, e=exchange rate

The impact on capital adequacy roughly equals the shock
times the open position
|
|
.
|

\
|
A
A
~
A
A

~
A
A
RW
RW
RW RW
RW
RW
RW
A
C
C
A
A
C
C
F
e
A
e
F
C
A
C A
e
F
e
e A e C
1
1 )] ( / ) ( [
2
28
Foreign Exchange Risk
... but stress test must reflect non-linearity arising from
FX options

Off-balance sheet (OBS) positions, which include
options, are not negligible in many countries.
Example
Net FX position/capital (%)

1998

1999

2000

2001

2002

Jan-03

... w/o OBS items

34.2

58.8

55.6

77.5

25.9

14.7

... with OBS items

40.0

18.9

25.2

13.1

-49.8

-26.8

29
Indirect FX Risk
Nonperforming loans vs. exchange rate


Usually much more significant than the direct FX risk

The analysis requires

Regression of leverage vs. NPLs
Inclusion of stock and flow exposures in FX

|
|
.
|

\
|
A
A

A
~ A ~ A
c
c
c
c
c
c
c c
E
D
E
D
a
E
F
e
e
e E e D a TL NPL )] ( / ) ( [ ) / (
30
Interest Rate Risk
Duration is the key indicator, because



This allows to express changes in capital adequacy ratio as



where

) 1 ( ) (
) (
,
) 1 ( ) (
) (
L
L L
L
L
A
A A
A
A
r
r D
r L
r L
r
r D
r A
r A
+
A
~
A
+
A
~
A
D
A
RW
RW
RW
A
L
L
A
L A
A
RW
A
A RW L A
GAP
r
A L
C
C
A
A
C
C
A
A
r
r
r
r
D D
r
A L
r
r A r r C
+
~
A
A

A
A

|
|
.
|

\
|
A
A
+
+

+
~
A
A
1
) / (
1
1
1
1
1
) / ( )] ( / ) , ( [
A
L
L
A
L A D
r
r
r
r
D D GAP
A
A
+
+
=
1
1
31
Interest Rate RiskIssues
Adequacy of the available data, including

Do banks report residual maturity properly?
Does the indicator capture the whole balance sheet?
Are off-balance sheet contracts included?

Simplified method: residual maturity plus weigths proposed by
Basel Committee

Nonlinearity
(duration changes with large changes in interest rates)

NPV may differ from the regulatory capital

Correlation between risk-weighted assets and assets

Indirect interest rate risk (see under credit risk)
32
Credit Risk Modeling
The most significant source of risk

Also, the most in need of strengthening

1. Mechanical approaches

2. Approaches based on corporate sector data
(leverage, interest coverage) & possibly household sector data

3. Approaches based on loan performance data
(including the VAR model already estimated)

33
1. Mechanical Approaches
Assume an inflow of new NPLs

Function of existing NPLs, performing loans, or a weighted sum of
the two

Assume provisions on existing NPLs

Increase in provisioning rates
Credit migration within NPLs (transition matrix)

Credit expansion model: inflow of new loans, followed by
credit migration to and within NPLs

Do the above by sectors (e.g. corporate & household)
34
2. Data on Borrowers
Leverage vs. NPLs (a possible model) *




Top-down calculations

* Notes: Based on an actual model used by IMF staff for cross-country panel data
estimates. Npls ratio of non-performing loans to total loans, lev leverage ratio, rcc
real cost of capital, reer real effective exchange rate, y-hat real GDP growth rate, p-
hat inflation rate, m-hat growth rate of M1, d-hat growth rate of domestic credit,
roe corporate sector return on equity


rcc m p y reer lev npls
5 4 4 3 2 1 0
o o o o o o o + + + + + + =
reer roe d p y rcc npl lev
7 6 5 4 3 2 1 0

| | | | | | | | + + + + + + + =
m y lev rcc
3 2 1 0
+ + + =
35
2. Data on Borrowers
Logit model predicting individual bankruptcy probabilities
as a function of age, size, industry characteristics &
corporate soundness indicators (earnings, liquidity,
financial strength)

Include interest and exchange rates on the right hand side
(to capture the indirect risk)

Link to individual banks through their exposures to the
various groups of companies

Predict bank potential losses
(also taking into account collateral)
36
2. Data on Borrowers
A simpler approach: exposure variables

Net open FX position & ratio of FX income to FX costs (for
indirect FX risk)
Interest coverage (for indirect interest risk)

If exposure variable exceeds an estimated (assumed)
threshold, default rate rises

Similarly to previous approach, translate to bank losses
(after collateral)
37
3. Loan Performance Data
Advantages

Also available for household sector
(with rapid lending growth in many countries)

Should be more readily available than leverage

Disadvantage

Lagging indicators of asset quality


38
Introducing Contagion Risk
Need to compile data for the following matrix

Bank 1 Bank 2 Bank n
Bank 1 -- -- Exposure of
bank 1 to
bank 2
Exposure of
bank 1 to
bank n
Bank 2 Exposure of
bank 2 to
bank 1
-- -- Exposure of
bank 2 to
bank n









-- --


Bank n Exposure of
bank n to
bank 1
Exposure of
bank n to
bank 2
-- --
39
Introducing Contagion Risk
Exposure = all uncollateralized lending
(including both on- & off-balance sheet exposures)

Currently, only data on total exposure of a bank to
interbank market are available

Two types of the contagion stress test

Pure contagion test: A fraud in a bank; impact on other
banks through interbank exposures

Macro contagion test: Macro shocks are grossed-up to trigger
failure of weakest bank; followed by interbank contagion
40
Introducing Contagion Risk
Implementation (example for 4 banks)

Capital adequacy
ratio P
i

CAR>=10 0.02
9=<CAR<10 0.05
8=<CAR<9 2
6=<CAR<8 25
4=<CAR<6 50
CAR<4 100

Si = (C
i
-E
1i
)/(A
i
-E
i1
),
where i=2, 3, and 4.
S
2
=(C
2
-E
12
-P
3
*E
32
-P
4
*E
42
)/(A
2
-E
21
-P
3
*E
23
-P
4
*E
24
)
S
3
=(C
3
-E
13
-P
2
*E
23
-P
4
*E
43
)/(A
3
-E
31
-P
2
*E
32
-P
4
*E
34
)
S
4
=(C
4
-E
14
-P
2
*E
24
-P
3
*E
34
)/(A
4
-E
41
-P
2
*E
42
-P
3
*E
43
)

... estimate as a part
of the EWS model
E
11
E
12
E
13
E
14

E
21
E
22
E
23
E
24

E
31
E
32
E
33
E
34

E
41
E
42
E
43
E
44


41
Aggregate stress test vs. interbank contagion stress test

Second round
bank failures
triggered by
contagion
Aggregate
impacts for
stress test
output
Matrix of
interbank
exposures
Failure of
individual
banks
Impact on
each
banks
capital
ratio
Aggregate
stress test
shock
Second round
bank failures
triggered by
contagion
Aggregate
impacts for
stress test
output
Matrix of
interbank
exposures
Failure of
individual
banks
Impact on
each
banks
capital
ratio
Aggregate
stress test
shock
Introducing Contagion Risk
42
Equity & Real Estate Price Risk
Equity price risksimilar to FX risk

Net open positions in equities
Need to include off-balance sheet exposures

Banks exposure to real estate price risk

Direct exposure (investment in real estate)
Credit exposure (developers etc.)
Degree of real estate collateralization
loan to value ratio
default probability (from credit risk stress test)
43
Concentration Risks (Credit)
Simple example: sensitivity analysis for large exposures

More sophisticated example

Run regressions for default probability on corporate data
(company-by-company), with dummy variables for the
sectors/regions

Ways to define default probability (actual defaultrun a logit
regression; or set a threshold for interest coverage ratio)

For a set of a banks exposures to sectors/regions, calculate
implied default probability
44
Liquidity Risk
Focus on bank liquidity stress tests

Results reported off-site, validate during on-site visits

Off-site cross-check (sensitivity analysis)

Overall risk: assume a % of deposits withdrawn
(percentages determined based on past bank runs, vary for different
maturities)

Concentration risk in deposits
(same as above, but for a percentage of the largest deposits)
45
Recent Developments in
Stress Test Methodology
Part IV
46
Bank Internal Stress Test Models
Two surveys of stress test practices in commercial banks
(2000 & 2004)

More attention to bank internal stress tests in on-site visits

Consider issuing guidelines on stress tests in commercial
banks?

Cross-check results of bank & supervisor stress tests
47
Stress Tests vs. Early Warning
Systems
Consider designing an EWS system in the form of a
statistical model of detection of bank failure/stress

Could be back tested against the ratings

BIS working paper on EWS for banking supervision
48
Cross-Market Contagion
Coverage of non-bank financial institutions in the
framework for consolidated supervision

Contagion between banks and non-bank financial
institutionse.g. insurance companies

Credit derivatives


49
Further Reading
Blaschke et al., 2001, ST of Financial Systems: An
Overview of Issues, Methodologies, and FSAP
Experiences, IMF WP 01/88
www.imf.org/external/pubs/cat/longres.cfm?sk=15166.
0

IMF & WB, 2003, Analytical Tools of the FSAP
www.imf.org/external/np/fsap/2003/022403a.pdf

ihk, 2004, ST: A Review of Key Concepts, Czech
National Bank technical note 2/2004
http://www.cnb.cz/en/pdf/IRPN_2_2004.pdf

You might also like