The views expressed in these papers are those of the authors only, and the presence of them, or of links

to them, on the IMF website does not

imply that the IMF, its Executive Board, or its management endorses or shares the views expressed in the papers.

Stress Testing Market Risks and Derivatives Portfolios

Presentation at the Conference

Macroprudential Supervision: Challenges for Financial Supervisors
Seoul, November 8, 2006

Gerald Krenn
Austrian Nationalbank, Financial Markets Analysis and Surveillance Division
[email protected]

Views expressed herein are those of the presenter and not necessarily those of Oesterreichische
Nationalbank.
www.oenb.at
 Oest er r eichische Nat ional bank

Agenda

I. Stress tests for market risk: basic concepts

II. Maximum Loss: a stress testing method uncovering portfolio-specific

worst-case scenarios

III. Stress testing: integrating market and credit risk (the methodology of
the “Systemic Risk Monitor”)

2 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

An Example

• Swap entered by SK Securities in Jan. 97 (see Gay et al. (1999))

• Payout after 1 year depended on FX rates of THB, IDR, JPY vis-à-vis USD:
⎡ THB0 3 ⋅ IDR0 − IDR1 − IDR2 JPY0 ⎤
Payout = USD 53m ⋅ ⎢5 ⋅ ( − 1) + Max (0 , ) + Max (0 , 1 − ) − 0.97 ⎥
⎣ THB2 IDR2 JPY2 ⎦
If positive: a profit; if negative: a loss

• Decision based on historical What really happened

volatilities THB IDR JPY VaR THB IDR JPY Loss

Volatility p.a. 1.23% 2.20% 6.88% USD 16 m Depreciation (1y) 51.8% 77.9% 2.9% USD 189 m

• How a stress test could have looked like

THB IDR JPY Loss

Scenario 1: minor crisis -15% -15% 0% USD 58 m

Scenario 2: midsize crisis -30% -30% 0% USD 116 m

Scenario 3: major crisis -50% -50% 0% USD 184 m

I. Basic concepts 3 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Stress Testing Market Risk

Ingredients for stress testing

• Portfolio: the trading book (subject to market risk)
Balance sheet Derivatives
positions - Interest rate derivatives (swaps, bond-options, caps,
- Bonds floors, …)
- Equity - Equity derivatives (equity options, index futures, …)
• - FX
Scenarios: possible future states of options
the market
r = (r1,...,rn) vector of risk factor values
ri are: interest rates, exchange rates, equity indices etc.
• Portfolio valuation function P as a function of r: P = P (r)
• Current state of the market: rCM
• Hence, current portfolio value: P (rCM)

Performing stress tests

1. Select scenarios rstress1, rstress2,... (according to some selection principle)
2. Calculate portfolio values P (rstress1), P (rstress2),…
3. Derive some measure of riskiness of the scenarios

I. Basic concepts 4 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

How to Select Scenarios

• Standard scenarios
• E.g. 200 bp interest rate shift

• Historical scenarios
• Replay a historical crisis
• Historically observed risk factor changes

• Subjective worst-case scenarios

• Initial shock is translated into risk factor changes
• Involvement of a wide range of staff, including senior management

I. Basic concepts 5 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Example: Interest Rate Risk in the Banking Book

• Standardized framework according to the Basel document on the principles for the
management and supervision of interest rate risk
• Part of Basel II - Pillar 2
• Coverage: interest rate sensitive positions of the banking book (on- and off-balance
sheet)
• Scenario: 200 basispoint shift of yield curves in all currencies
– Per currency: take the worst case depending on the distribution of assets and
liabilities in a re-pricing scheme
• Compare resulting decline in economic value to the sum of Tier 1 and 2 capital
– Above a 20% threshold: bank considered as outlier

I. Basic concepts 6 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Dangers of Scenario Selection

• A stress scenario for one portfolio might be a lucky strike for another portfolio
• Standard and historical scenarios may nourish a false illusion of safety
• Subjective worst-case scenarios might be too implausible to trigger management
action

Requirements for “objective worst-case scenarios”:

– Scenarios should be portfolio-specific
– There should be some “objective” measure of plausibility
– Consider only scenarios with minimal level of plausibility
– Within plausible scenarios, look for the most harmful one

Maximum Loss: Framework for selecting objective worst-case scenarios

I. Basic concepts 7 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Maximum Loss

• Good overview on Maximum Loss in doctoral thesis by Studer (1997)

• Chose trust region TR: Set of scenarios above a certain minimal plausibility threshold
• MaxLossTR ( P ) := sup {P(rCM ) − P(r )}
Maximum Loss defined as:
r∈TR

• “Above the plausibility threshold no loss worse than Maximum Loss can happen”

Choice of trust region

• By means of the multivariate risk factor distribution
• Trust region shall have some predefined probability (p) and contain only scenarios
with “highest density”
• In case risk factors have an elliptic distribution (e.g. multivariate normal, Student-t):
Trust region is an ellipsoid of scenarios with Mahalanobis distance to rCM below some
threshold kp:
{ }
TR = r : (r − rCM )' Σ −1 (r − rCM ) ≤ k p
2

Σ
( is the co-variance-matrix)

II. Maximum Loss 8 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Trust Region: Area of Highest Density

II. Maximum Loss 9 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank
Within Trust Region: Find Scenario with Smallest
Portfolio Value (= Maximum Loss)

II. Maximum Loss 10 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Benefits of Maximum Loss

• Maximum Loss not only quantifies risks but also identifies a worst case-scenario
• Searching for worst-case scenarios yields more harmful and more plausible
scenarios than other ways of identifying stress scenarios
• Sample portfolio consisting of options on different international stock indices
– Stress scenarios are identified in different ways
• Worst-case according to the recommendations of the Derivatives Policy Group
• Recurrence of Black Friday in October 1987
• Worst-case scenario implied by Maximum Loss

Relative Loss Plausibility

Worst DPG - 183% once in 10 yrs
Black Friday - 154% once in 19 yrs
Worst Case (ML) - 279% once in 8 yrs

II. Maximum Loss 11 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

Benefits of Maximum Loss

Identifying key risk factors of the worst case scenario = Locating the vulnerable spots
of a portfolio

Example: Again option portfolio

Risk Rel. Loss Explanatory Power

Fact Chan
Report 1 FTSE100
ors -13%
ges 206 74%
%
FTSE100 -13% 264
Report 2 94%
DJI -8% %

Loss (rreport )
Explanatory Power =
Loss (rworst case )

II. Maximum Loss 12 / 20 Stress Testing Market Risks and Derivatives

Portfolios
 Oest er r eichische Nat ional bank

The Problem of Dimensional Dependence

• n : number of risk factors

• Consider elliptic risk factor distributions; then trust regions are ellipsoids
• Trust region shall have probability p
• k : radius of ellipsoid
• n, p, and k depend on each other: e.g. p depends on k and n
In case of the normal distribution:
k2 n s
1 −1 −

2 Γ(n / 2) ∫0
p(k , n) = 1 − Fχ 2 (k ) = 1 − n / 2
2
s e ds
2 2
n

• To keep p fixed: k has to increase as n increases

• If we add an “empty risk factor” (i.e. a factor on which the portfolio value does not
depend),
k has to increase in order to hold p fixed
• We therefore search for MaxLoss within a larger trust region when we add an empty
risk factor
• Also MaxLoss is likely to be larger once having added an empty risk factor
==> Makes
II. Maximum Loss it hard to compare Maximum
13 / 20 Loss across portfolios
Stress Testing Market Risks and Derivatives
Portfolios
 Oest er r eichische Nat ional bank

Systemic Risk Monitor (SRM) – Basic Structure

III. Integrating Market and Credit 14 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Stress Testing in SRM

• 26 market risk factors + 8 credit risk factors = 34 risk factors

• The time horizon in SRM is 3 months
• These factors are modeled statistically
– Allows for a Monte Carlo-simulation for analyzing the actual situation
(sampling from the un-conditional distribution)
– Allows for a Monte Carlo-simulation for stress testing (sampling from the
conditional distribution)
• For stress testing, a set of risk factors is set to some predefined values
• Remaining factors are sampled from the conditional distribution

• Stress is considered in two ways

1. Direct stress from the stressed risk factors
2. Indirect stress (“statistical feedback”) from the remaining risk factors that are
influenced by the stressed risk factors

III. Integrating Market and Credit 15 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Statistical Modeling of Risk Factors

• Multivariate distribution of risk factors is estimated in a 2-step procedure:

– Step 1: Modeling of marginal distribution of each risk factor by models which
are optimized with respect to their out-of sample density forecast
– Step 2: Modeling of dependencies between individual risk factors by a grouped
t-copula
• Our goal is to have enough flexibility in order to capture
– Marginal distributions of the various risk factors
– Patterns of dependence between risk factors
• Market risk factors and credit risk factors are treated in a common statistical model

III. Integrating Market and Credit 16 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Marginal Distributions: Model Selection

• No aggregation of higher frequency data, i.e. use quarterly data directly

Logarithmic changes of JPY 3 months zero prices

0,4%
• GARCH 0,3%

– Testing procedure favors consideration 0,2%

of GARCH effects 0,1%

0,0%
– Makes sense for analysis of -0,1%

current situation -0,2%

– Should be used with care for stress tests -0,3%

-0,4%

81/03
82/06
83/09
84/12
86/03
87/06
88/09
89/12
91/03
92/06
93/09
94/12
96/03
97/06
98/09
99/12
01/03
02/06
03/09
04/12
• Distribution of Residuals
– Extreme value distribution performs best in the test procedures
– Simulations show that extreme value distribution leads to too extreme movements
– SRM now uses t-distribution as marginals

III. Integrating Market and Credit 17 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Modeling Dependencies: Grouped t-Copula

• Copula models dependencies between risk factors

– Copula is the part of the multivariate distribution which is not contained in the
marginal distributions
• Concept of tail-dependence for assessing dependencies
λ between two variables is defined as:
– The coefficient of tail-dependence

– Is roughly speaking the probability that one variable is very large (small) given the
other variable is very large (small)
– In case λ > 0, “one variable can pull up (down) the other variable”
• λ
For the multivariate normal distribution we have = 0 (no tail-dependence)
– Real data show tail-dependence
• An alternative is given by the t-copula
– There is tail-dependence between risk factors λ ( > 0)
– Scenarios can be generated easily in a Monte Carlo-simulation
– Drawback: between all risk factors there is the same tail-dependence

III. Integrating Market and Credit 18 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Modeling Dependencies: Grouped t-Copula

• As an alternative to the t-copula the grouped t-copula was introduced by Daul et al.
(2003)
– Risk factors are arranged into groups
– Within each group risk factors have the same tail-dependence
– Each group is characterized by a parameter (degrees of freedom)
• Grouped t-copula was adopted for SRM
– Is suited equally well for MC-simulations as the plain t-copula
– In SRM risk factors were arranged into 4 groups (in parentheses: estimated
degrees of freedom)
• Credit risk factors (20)
• FX (14)
• Equity (5)
• Interest rates (11)

III. Integrating Market and Credit 19 / 20 Stress Testing Market Risks and Derivatives
Risk Portfolios
 Oest er r eichische Nat ional bank

Literature

Basel Committee on Banking Supervision (2004): “Principles for the Management and Supervision of Interest Rate
Risk”

Daul S., E. DeGeorgi, F. Lindskog, A. McNeil (2003): “The grouped t copula with an application to credit risk”, RISK Vol.
16, pp 73-76

Gay G. D., Kim J., Nam J. (1999): “The Case of the SK Securities and J.P. Morgan Swap: Lessons in VaR Frailty”,
Derivatives Quarterly, Spring 1999, pp. 13-26

Studer G. (1997): “Maximum Loss for Measurement of Market Risk”, Doctoral Thesis, Swiss Federal Institute of
Technology, Zürich

20 / 20 Stress Testing Market Risks and Derivatives

Portfolios