1

2011 Emese Lazar, ICMA Centre

Market Risk
Office hours: Wednesday 2.30 4.30
Friday 1 3
2
2011 Emese Lazar, ICMA Centre
Topics
Market risk theory:
Lecture 1. The characteristics of markets and market risk &
capital requirements
Lecture 2. VaR models part 1
Lecture 3. VaR models part 2
Lecture 4. Advanced VaR models
Applied market risk:
Lecture 5. Equities and foreign exchange
Lecture 6. Fixed income products
Lecture 7. Equity and interest rate derivatives
Lecture 8. Fund management, banking & non-financial firms
3
2011 Emese Lazar, ICMA Centre
Reading list
Carol Alexander: Market Risk Analysis
Volume IV. Value at Risk Models
Carol Alexander & Elizabeth Sheedy: PRMIA Handbook
Volume III, Chapters 0 and A
Carol Alexander: Market Models
Kevin Dowd: Measuring Market Risk
Greg Gregorious: The Var Implementation Handbook
Steven Allen: Financial Risk Management
Philippe Jorion: Value at Risk
Nassim Taleb: Dynamic hedging
Glyn Holton: Value-at-Risk
4
2011 Emese Lazar, ICMA Centre
Assessment
30% Two open-book multiple choice tests on Blackboard
Test 1: 10 March 6 pm
Test 2: 27 April 6 pm
10% Assignments (5 in total); these are compulsory!
They are individual (zero points for similar submissions)
Mark is for the effort to solve the assignments, and
not for the correctness of the solutions!!!!!!!!!!!
60% Exam
5
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 1
A. The characteristics of markets and market risk
B. Capital requirements
6
2011 Emese Lazar, ICMA Centre
Contents
A. The characteristics of markets and market risk
1. An example
2. Market risk
3. Risk measures
4. Risk management
B. Capital requirements
5. Regulatory capital
5.1 Basel accords
5.2 Internal models
5.3 Standardized approach
6. Economic capital
6.1 Economic capital calculation
6.2 Capital allocation
6.3 Risk adjusted performance measures
7
2011 Emese Lazar, ICMA Centre
1. An example
You are given a portfolio of 100 mil to manage for
1 year; you are allowed to do anything with it.
What do you do?
What can go wrong? risk
8
2011 Emese Lazar, ICMA Centre
2. Risk
an estimate of loss due to unexpected changes in
risk factors a measure of uncertainty
We care only about negative effects because of non-linearity of utility functions
0
1
2
3
4
5
6
1 50 100 150 200 250

U
t
i
l
i
t
y
9
Financial risks
All decisions (taken by individuals or by financial companies)
involve some market risk
Identify the risk sources
Compute the risk exposures
Manage the risks involved (reduce risk & choose a risk profile that
suits your goals)
Some risks faced by companies can be managed by
shareholders, but some of these are better to deal with by the
company itself (for example, to reduce taxes, to reduce the volatility of
payoffs and to reduce the risk of bankruptcy)
Derivatives are zero-sum games (same amount is made as lost), so
its all about betting correctly, measuring & managing risks
2011 Emese Lazar, ICMA Centre
10
2011 Emese Lazar, ICMA Centre
Types of financial risk
Market risk: risk resulting from adverse moves of market price & rates
Interest rate risk
Equity risk
Exchange rate risk
Commodity price risk
Volatility risk
Credit spread risk (according to some definitions)
Credit risk: exposure to default (or downgrade) of a counterparty
Operational risk: exposure to failure of internal system
They are interrelated!
11
Financial risks
How big are these risks?
Notional size of the derivatives market ~ $700 trillion (10 times of
yearly world GDP): bonds - $80 trillion, equity - $40 trillion, loans -
$20 million, credit derivatives - $30 trillion
The MRs taken by banks have been increasing, whilst the
amount of CRs taken have been decreasing recently
Banks provide their clients with new products to help manage MR
(IR swaps, index linked loans etc.)
the introduction of new products also increases market volatility
They want to profit from market price movements (arbitrage &
speculation) trading book
They transfer some credit risks to the capital markets (loans
transformed into bonds, mortgages are securitized, CDS), thus CRs
become MRs
2011 Emese Lazar, ICMA Centre
12
2011 Emese Lazar, ICMA Centre
Risk factors
PORTFOLIO
Risk
factor 1
Risk
factor 2
Risk
factor 3
Risk
factor 4
13
2011 Emese Lazar, ICMA Centre
Market risk factors
= a variable (whose value will be known) that will affect
your portfolios future value
Portfolios: Risk factors:
Equities - stock & index returns etc.
Fixed income products - yield curve
Foreign investments - FX rate
Options - underlying asset prices
& their volatilities
Risk factors can be unobservable & need to be estimated!
Question: What is more risky: fixed or floating rate loans?
14
2011 Emese Lazar, ICMA Centre
3. How to measure risk?
Risk measure: a set of operations (a process) that
quantifies risk
Risk metric: a simple operation that quantifies risk
- statistical (probabilistic) metrics of portfolios or factors
e.g. standard deviation, correlation, or probability of an event, e.g. of
having negative return, or VaR
- exposure to risk factors (sensitivities)
e.g. delta of an option
15
2011 Emese Lazar, ICMA Centre
Value at Risk (VaR)
It is a positive amount so that the probability of losing
more than VaR
%, h days
in h days is exactly %
P ( P&L
h days
VaR
%, h days
) = %
P&L in h days at t
D
e
n
s
i
t
y

f
u
n
c
t
i
o
n
%
- VaR%,h, t
16
2011 Emese Lazar, ICMA Centre
Interpretation of VaR
It is the maximumamount of money we are 1 %
sure to lose over some period of time h days
In other words, out of 100 portfolios 1 % will lose less
than VaR
%. h
, and % will lose more than VaR
%, h
in h days
It is one number that quantifies possible losses
It is a simple metric, but it led to different estimation
procedures and methodologies and it introduced
different risk management approaches
In case of illiquid markets, adjustments need to be
made
17
2011 Emese Lazar, ICMA Centre
4. Who manages risk?
1. Portfolio managers
Only un-diversifiable risk is of importance, expressed by:
Beta of a stock: from CAPM
Tracking error: the error in mimicking the market index
2. Option traders (front office)
Only the volatility of assets is of importance, expressed by:
Implied volatility that will help compute hedge ratios
3. Risk management units (middle office)
- These are separate from the risk-taking units
- They recommend hedging strategies, set limits & implement stress tests
They identify, assess, report and control risk based on:
Statistical and implied volatility and correlation
18
2011 Emese Lazar, ICMA Centre
The risk management process - steps
1. Identification
- special care to hidden risks and risk interaction
2. Assessment
- based on a model broad choice
3. Monitoring
- updating & reporting of exposures
4. Control / Mitigation
- find the optimal balance for risk & return
- manage risk according to regulations
Very important!
19
2011 Emese Lazar, ICMA Centre
The importance of risk management
1. To decrease the probability of bankruptcy
2. To understand the sources of risk & measure it
3. To minimize everyday losses
4. To minimize losses in case of market crashes
Reasons that lead to the importance of managing risk:
High volatility (leading to an increased probability of bankruptcy)
Increased trading volumes (especially derivatives)
Advances in technology
New regulations
20
2011 Emese Lazar, ICMA Centre
B. Capital
Capital = Assets Liabilities
C = A D
Roles : - to cover large unexpected losses
- protect depositors
- provide confidence for investors & rating agencies
21
2011 Emese Lazar, ICMA Centre
Capital
Capital is a tool for:
1. Risk measurement & aggregation VaR
2. Performance measurement RAPM
3. Asset and business allocation
Another interpretation:
it is a call option on the value of the firms
assets with strike equal to the value of the debt
22
2011 Emese Lazar, ICMA Centre
Types of capital
Regulatory capital (RC): required by regulators (accounting measure)
Applies to all financial institutions; little freedom (a strict internal model
or the standardized approach can be used), bottom-up approach
Economic capital (EC): needed to cover all expenses and
provide protection against market, credit and operational risks
= desirable level of capital to hold as insurance against risks
= net asset value that guarantees solvency (risk memtic)
Institution specific; freedom to compute (any internal model can be
used), top-down approach
Book capital: the actual capital held
It can be computed separately for different businesses and
activities, and across different types of risk (market, credit, operational)
23
2011 Emese Lazar, ICMA Centre
5. Regulatory capital (RC) for banks
Roles of capital from the regulators perspective:
- reduce systemic risk (governments are guarantors)
- provides the game rules for banks competition
Limitations of regulation:
- it is constantly changing
- it is too general
- it led to the development of regulatory capital arbitrage industry
How objective/subjective should regulation be?
24
2011 Emese Lazar, ICMA Centre
5.1 Basel accords
Regulations:
1. Basel I Accord (1988) + 2 amendments (1995, 96)
mainly credit risk; market risk since 1996
principles for supervision and minimum capital standards
objectives: safety and market stability
fair value accounting (mark to market) as opposed to historical accounting
2. Basel II Accord (2005, implementation in 2006-7)
better risk management, international harmonization
operational risk first mentioned in the proposal (1999)
driven by regulators
In 2008: Incremental Risk Charge (applied to default risk)
implementation in January 2010 (inspired by the credit crunch)
www.bis.org
25
2011 Emese Lazar, ICMA Centre
Basel I.
Total capital requirements:
RC
total
= RC
MR
+ RC
CR
Market risk capital:
RC
MR
= [ m
MR
*VaR
1%, 10 days
+ m
SR
*VaR
1%, 10 days, Specific
] * Trigger/8
where m
MR
, m
SR
are multipliers between 3 and 4, and between 4 and 5
VaR are for systemic and specific risks
Trigger is related to the quality of control in banks, between 8 and 25
Credit risk capital
RC
CR
= [ RWA
k
] * 8%
where RC = eligible capital
RWA = risk-weighted assets, after adjusting for credit risk
8% is called the capital adequacy ratio or Cook ratio or minimum solvency ratio
26
2011 Emese Lazar, ICMA Centre
Basel I.
Eligible capital is decomposed into:
Tier 1 (core capital): shareholder funds + retained earnings
Tier 2 (supplementary capital): long-term (>5 years) subordinated debt +
reserves + provisions + long-term hybrids
Tier 3 (sub-supplementary capital): short-term (2-5 years) subordinated debt
Regulation:
1.) Tier 1 capital > Tier 2 capital
2.) Subordinated debt < 50% of Tier 1 capital
Capital ratio for Tier x = Tier x capital / RWA
3.) % = 1%; h = 2 weeks; Capital at least 3 * average VaR
27
2011 Emese Lazar, ICMA Centre
Basel II.
Regulation has 3 parts:
Pillar 1: Minimum capital requirements
- min capital / RWA is still 8%; but the definition of RWA changed
- banks have choices relative to the risk calculation
- changes to the treatment of credit risk + introduction of op. risk
Pillar 2: Supervisory review
- banks have to assess their capital adequacy relative to the overall risk
- supervisors have to review these and take action if needed
- capital is to cover losses AND to manage the business
- includes new techniques such as stress testing for collaterals etc.
Pillar 3: Market discipline and Public disclosure
- effective market disclosure of capital levels and risk exposure
28
Basel II.
Cooke ratio must be at least 8%
Eligible capital = Tier 1 + Tier 2 + Tier 3 (and must be hedged)
Tier 2 has to be less than Tier 1
Total capital = RWA = RC
total
= RC
MR
+ RC
CR
+ RC
OpR
Eligible Capital
=Cooke ratio
Total Capital
29
Basel II.
Market risk capital charge (RC
MR
) applies to all on and off
balance sheet positions in a trading book.
RC
MR
= GRC + {SRC or IRC}
GRC = general risk charge
SRC = specific risk charge IRC = incremental risk charge
SRC or IRC applied to equity & IR exposure (due to changes
in credit spread), but not to commodities and currencies
For IR: report per currency and per maturity band
For currencies: report exposures in each currency
For commodities: report per maturity band
2011 Emese Lazar, ICMA Centre
30
Basel II.
Banks can use an internal model to estimate RC
MR
if:
Independence between risk taking & control
Regular backtesting
Involvement of management in risk control
Use internal models for trading & exposure limits
Rigurous stress testing
Documentation for the risk moel
Regular auditing
Recommendation:
1% 10-day VaR estimated daily
Historical sample at least 1 year (except when volatility increases),
updated every 3 months
Models: internal models or standardized rules
Models must capture risks of options portfolios using 10-day changes
and capture non-linearities and volatility
31
5.2 Internal models
1.GRC is the maximum of k times the average VaR over the last 60
days, and yesterdays VaR
The k multiplier is between 3 and 4, depending on the models
backtesting results
This gives a measure higher than the 0.000000001% 10-day VaR
Possible problem: VaR doesnt increase enough when risk
increases this is why backtesting is required
2. In some cases, GRC might be computed using scenario analysis
60
1 10 1 10 1
1 60
t %, days ,t i %, days ,t
i
k
GRC max VaR ,VaR

=

= `
)
32
2011 Emese Lazar, ICMA Centre
5.3 Standardized approach
An alternative to internal models
Total capital requirements:
RC
total
= RC
MR
+ RC
CR
Market risk capital:
RC
MR
= RC
MR, equity
+ RC
MR, currency
+ RC
MR, commodity
+ RC
MR, IR
RC
MR, equity
= GRC + SRC
where GRC is 8% of the exposure (netting long and short exposures)
SRC is 8% of the exposure (without netting long and short
exposures) OR
4% of the exposure for liquid & diversified portfolios
RC
MR, IR
is more complex, based on maturity bands
The same SRC is applied for internal models as well, unless the
model incorporates specific risk
33
2011 Emese Lazar, ICMA Centre
6. Economic Capital (EC)
Top-down approach: starts with the overall performance; used for risk
budgeting (for EC allocation)
Earnings volatility approach based on earnings:
assumes that the capital is equal to the value of a perpetual stream of
expected earnings
EC
%, h
= EaR
%, h
/ k k = required rate of return
EaR
%, h
= difference between earnings &
earnings under % worst-case scenario
Option-theoretic approach based on stock prices:
assumes that the capital is a call option on the firms assets
(strike = debt level that ensures no bankruptcy at % level)
Bottom-up approach: starts with modelling individual transactions and
businesses and then aggregating the risks (for EC computation)
It computes market, credit & operational VaR & adds them up
34
2011 Emese Lazar, ICMA Centre
EC aggregation
For each type of risk:
EC
i, total
= p * EC
i, VaR
+ (1 p) * EC
i, stress
Stress EC: Capital required to cover losses under several extreme scenarios
Aggregation across different types:
1.) assuming perfect correlation
EC
total
= EC
i, total
2.) assuming non-perfect correlation:
2 2
1 2 12 1 2
2
total
EC EC EC EC EC = + +
35
6.1 Economic capital calculation
It is the capital level that guarantees with a high probability that the asset value
remains positive (the firm stays solvent) over a given period
Notation:
A
0
= assets r
A, h
= return on assets A
h
= A
0
(1 + r
A, h
)
D
0
= liabilities r
D, h
= return on liabilities D
h
= D
0
(1 + r
D, h
)
(cost of debt)
Asset returns are uncertain; r
A, h
will follow a distribution
What is the maximum level of debt that guarantees solvency at level?
Find D
h
*
= so that P(A
h
< D
h
*
) =
In this point D
h
*
= A
h,
= A
0
(1 + r
A, h,
)
We also have that D
h
*
= D
0
*
(1 + r
D, h
)
Thus D
0
*
= A
0
(1 + r
A, h,
) (1 + r
D, h
)
1
The economic capital is EC = A
0
D
0
*
We also notice that E (A
h
) = A
0
(1 + E (r
A, h
))
Thus EC = PV (E (A
h
), using E (r
A, h
) ) PV (A
h,
, using r
D, h
)
2011 Emese Lazar, ICMA Centre
36
2011 Emese Lazar, ICMA Centre
6.2 Capital allocation
EC allocation is done in 2 ways:
Allocate a priori to maximise risk-adjusted returns
Allocate a posteriori to different business units important!
Uses of EC allocation:
Management decision support
Performance measurement & compensation
Pricing, profitability assessments & setting limits
Building optimal risk-return portfolios
37
2011 Emese Lazar, ICMA Centre
Classification of allocations
- Stand-alone EC: the amount the unit would consume if it were
an independent firm
EC
i
> EC
firm
- Incremental EC: the amount the unit contributes to the firms EC
or the amount that would be released if the unit was sold
EC
i
= EC
firm
EC
firm i
EC
i
< EC
firm
- Marginal EC: the contribution of the unit viewed as part of the firm
(considering diversification benefits)
EC
i
= EC
firm
38
2011 Emese Lazar, ICMA Centre
Marginal EC
The risk contribution of the i
th
unit to the total EC:
rc
i
= EC
i
/ EC where EC = EC
i
It can be shown that:
EC
i
= x
i
* [ EC (x
i
) / x
i
]
where x
i
is the size of the i
th
unit
39
2011 Emese Lazar, ICMA Centre
Example of marginal EC
Stand-alone capital Incremental capital Marginal capital
(mil) % (mil) % (mil) %
Business 1 50 50% 26 70% 41 66%
Business 2 30 30% 8 22% 15 24%
Business 3 20 20% 3 8% 6 10%
Total 100 100% 37 100% 62 100%
% of EC 161% 60% 100%
Note: it is assumed that the correlation between the capital of different units is zero
40
2011 Emese Lazar, ICMA Centre
6.3 Risk adjusted performance measures (RAPM)
a consistent metric that spans all asset and risk classes
providing a benchmark for evaluating the performance of
alternative business units
It is an alternative for ROA (which doesnt take account for leverage effect and different
risk levels) and ROE (which doesnt take account for different risk levels)
It is a tool for capital allocation purposes (derived from ranking of RAPM)
It is ROC with risk adjustments where C is the capital contribution of the asset
to the overall portfolio (hence the link with capital allocation)
Types:
RAROC: adjusts the P&L example: RAROC = [E(P&L) k * EC] / C
RORAC: adjusts the capital example: RORAC = E(P&L) / EC
RARORAC: adjusts both
41
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 2
VaR models part I.
42
2011 Emese Lazar, ICMA Centre
Contents
1. Introduction & preliminaries
2. Historical VaR
3. Normal Linear VaR
4. Monte Carlo VaR
5. Methodology comparison
6. Advantages and Disadvantages of VaR models
7. Expected shortfall
8. Marginal VaR and Incremental VaR
43
2011 Emese Lazar, ICMA Centre
1. Introduction
Many types of traditional limit setting methods existed
before the introduction of VaR
e.g. the size of net open positions, the greeks, the maturity mismatch
several problems and shortcomings
e.g. risk exposures move quickly, good and bad risks are not differentiated
It became clear that what was important is the
probability distribution of potential losses
VaR
44
2011 Emese Lazar, ICMA Centre
VaR
VaR is a positive amount so that the probability of having a loss
bigger than VaR
%, h, t
in h days at time t is exactly %
P ( P&L
h, t
VaR
%, h, t
) = %
Profit and Loss in h days at t
D
e
n
s
i
t
y

f
u
n
c
t
i
o
n
%
-VaR%,h, t
45
2011 Emese Lazar, ICMA Centre
Discussion
VaR is an estimate of the maximum loss that you are 1 %
certain to have (it occurs in only % of the cases)
not unique; it depends on the assumptions and the methods used
VaR assumes that the trading positions are fixed during the
period under study
is this a realistic assumption?
VaR does not address the distribution of potential losses that
occur when the VaR estimates are exceeded (% of all cases)
this is an important question as well
Two subjective parameters: %, holding period + method
regulations try to standardise these
46
2011 Emese Lazar, ICMA Centre
Discussion
Interpretations:
It is the maximum amount of loss that 100 * (1 %) portfolios would
have out of 100 (100 * % portfolios would lose more than VaR)
It is the maximum amount of loss a portfolio would have in 100 * (1
%) periods of length h out of 100 ( 100 * % times you lose more)
Parameter choice:
significance level % (confidence level 1 %): depends on objectives
and regulation; also for example on the desired credit rating that depends
on default probability
e.g. if you want to compute the capital that can cover losses that happen 99 days out of
100, then use % = 1%.
% small for capital requirements, high for backtesting
Holding period / risk horizon: depends on the optimal hedging period
and on the liquidity of the assets (liquidity decreases in stress markets!)
e.g. h = 10 days for estimation of RC and 1 day for estimation of EC
short periods for both capital requirements and backtesting
47
2011 Emese Lazar, ICMA Centre
Properties of VaR
1. VaR increases with the holding period h
2. VaR increases when significance level % goes down
3. Scaling VaR: For linear portfolios (not for options) we can
compute h-day VaR from 1-day VaR (assuming iid normal
returns and constant weights OR iid normal risk factors and
constant risk factor sensitivities):
VaR
%, h, t
= h * VaR
%, 1 day, t
+
1 day
* (h h)
- if
1 day
= 0 (a reasonable assumption for short holding periods):
VaR
%, h, t
= h * VaR
%, 1 day, t
48
2011 Emese Lazar, ICMA Centre
Data transformation
Raw data: prices
To estimate VaR, the following transformation can be used:
Properties
Simple P&L simple
ignores initial portfolio value
Discounted P&L more precise when rates are high
and/or holding period is long
Relative returns not additive
Discounted relative returns more precise
still not additive; not used
Log returns best for short holding periods
Discounted log returns more precise
not used because often holding period is short
49
2011 Emese Lazar, ICMA Centre
P&L
Simple P&L: P&L
h,t
= P
t+h
P
t
Disadvantage: when the holding period is long and/or interest
rates are high, then the effect of interest rates becomes important
e.g. an expected P&L of 1 mil during 1 year is actually worth less when interest
rates are high than an expected P&L of 1 mil when interest rates are low
Discounted P&L: based on the discounted value of the portfolio
at the end of the holding period
P&L
d
h,t
= B
h,t
P
t+h
P
t
where B
h,t
= (1 + r
h,t
)
h/365
is the discount bond at t
Advantage: to get rid of the effect of the interest rates
50
2011 Emese Lazar, ICMA Centre
Returns
P&L
Disadvantage: the initial value of the portfolio is ignored
e.g. a loss of 50 when the portfolio is valued at 100 is different from a loss of
50 when the portfolio is valued 1000
Relative returns: r
h,t
= (P
t+h
P
t
) / P
t
Disadvantage: it is not additive
e.g. if the evolution of the portfolios value is 100, 200, 100 then the average
return would be 25% instead of 0%
Log returns: r
h,t
= ln (P
t+h
/ P
t
)
it is best to use log-returns when holding period is short!
51
2011 Emese Lazar, ICMA Centre
Returns
When using returns, first we compute VaR for the returns
(using any method & returns data) denoted by VaR
r
%, h, t
Then we multiply this by the portfolios current value to
obtain VaR in monetary terms:
VaR
%, h, t
= VaR
r
%, h, t
* P
t
For log returns the exact formula would be:
VaR
%, h, t
= P
t
* (exp [ VaR
r
%, h, t
] 1)
For log-returns this formula is more precise,
but it is ok to use the previous formula, as the two give
similar results
52
2011 Emese Lazar, ICMA Centre
Discounted returns
Discounted relative returns:
r
h,t
= (B
h,t
P
t+h
P
t
) / P
t
Disadvantage: not additive
Discounted log returns:
r
h,t
= ln (B
h,t
P
t+h
/ P
t
)
Not used in practice!
53
2011 Emese Lazar, ICMA Centre
Constant linear portfolios
portfolio value at time t:
number of assets i: n
it
price of asset i at time t: p
it
weight of asset i at time t:
Keep constant the number of assets: n
it
constant
this is the case if trading is not allowed
Keep constant the holdings value: n
it
p
it
constant
change n
it
when p
it
changes
Keep constant the weights: w
it
constant
rebalance entire portfolio when p
it
changes
( )
1 1

k k
t it it t it it
i i
P n p P n p
= =
= =
it it
it
t
n p
w
P
=
54
2011 Emese Lazar, ICMA Centre
Constant derivative portfolios
portfolio value:
risk factor j value at time t: f
jt
sensitivity to risk factor j: s
jt
these also depend on time and on risk factor value f
jt
!
Keep constant the number of assets: n
it
constant
this is the case if trading is not allowed
Keep constant the holdings value: n
it
p
it
constant
change n
it
when p
it
changes
Keep constant the weights: w
it
constant
rebalance entire portfolio when p
it
changes
Keep constant the sensitivities: s
jt
constant
rebalance portfolio when risk factor value f
jt
changes
( )
1 1 1

k l l
t it it jt jt t jt jt
i j j
P n p s f P s f
= = =
= = =
55
2011 Emese Lazar, ICMA Centre
Level of measurement
VaR can be measured at two levels:
Portfolio level: the risk factor is the portfolio itself; more
precisely the risk sources are not identified.
The VaR of the portfolio is computed from historical
data on the portfolio or from properties of the portfolio
Risk factor level: the risk of the portfolio is attributed to
different sources, called risk factors. The risk factors are
identified and the sensitivity of the portfolio to these risk
factors is measured. This is much better than portfolio level
measurement when trading strategies change!
The VaR of the portfolio is computed from historical
data on the risk factors or from properties of the risk factors,
using the sensitivities.
56
2011 Emese Lazar, ICMA Centre
Types
Historical VaR
Normal Linear VaR
Monte Carlo VaR
57
2011 Emese Lazar, ICMA Centre
2. Historical VaR (portfolio level)
Assumptions:
Only one risk factor, this is the portfolio itself
Future behaviour of this is the same as past behaviour
(it requires past data)
No distributional assumptions about risk factors
Computation:
Using data on a specified historical period, we construct a hypothetical
P&L (or returns) series for the current portfolio
This P&L (returns) series will be measured over a given time interval called
holding period (e. g. 1 day)
This will give a P&L (returns) distribution
The VaR
%,h,T
will be the lower % percentile of this density
58
2011 Emese Lazar, ICMA Centre
Historical VaR
Methodology: (for P&L, but similar for returns)
1. Create weights for each observation: weight = 1/N
(assuming N = 500)
2. Order the P&L observations in increasing order
3. Compute cumulative weights
4. VaR
%
will be the value that has cumulative weight %
(cumulative weight = 1%)
Order P&L Weight Cum. Weight
178 -359300 0.002 0.2000%
31 -297000 0.002 0.4000%
259 -286000 0.002 0.6000%
74 -256000 0.002 0.8000%
154 -228000 0.002 1.0000%
-VaR1%
Order P&L Weight
1 25480 =1/500
2 45897 =1/500
3 -10987 =1/500
59
2011 Emese Lazar, ICMA Centre
Historical VaR - histogram
0%
2%
4%
6%
8%
10%
12%
14%
16%
0 100 200 300
-VaR5%
60
2011 Emese Lazar, ICMA Centre
Weighted historical VaR
Motivation:
Previously, when computing VaR we used the same weight for each
historical observation
There are periods characterized by larger or smaller volatility (caused by
some event) or that are of less importance (like older data)
When computing VaR, then more volatile data periods should be
normalized by their volatility; similarly, older data should have less
weight
Solution:
Age-weighted historical VaR (exponential weighting)
Volatility-weighted historical VaR
61
2011 Emese Lazar, ICMA Centre
Age-weighted historical VaR
Methodology: (for P&L, but similar for returns)
1. Order the observations in inverse time order (1 = most recent)
2. Create weights for each observation using exponential weighting
3. Order the observations in increasing order
4. Compute cumulative weights
5. VaR
%
will be the value that has cumulative weight %
(cumulative weight = 1%)
Order P&L Weight
1 25480 =(lambda^L4)/(1/(1-lambda)-1)
2 45897 =(lambda^L5)/(1/(1-lambda)-1)
3 -10987 =(lambda^L6)/(1/(1-lambda)-1)
Order P&L Weight Cum. Weight
178 -359300 0.0001% 0.0001%
31 -297000 0.9375% 0.9376%
259 -286000 0.0000% 0.9376%
74 -256000 0.0655% 1.0032%
-VaR1%
62
2011 Emese Lazar, ICMA Centre
Volatility-weighted historical VaR
Methodology: (using returns only!)
1. Estimate volatility using a time-varying volatility model where
t
is
the volatility process
2. Replace the returns by volatility-adjusted returns r
*
t
= r
t
/
t
3. Compute VaR
r*
%, h, T
for these adjusted returns
4. Use the following formula to compute the VaR of the portfolio
VaR
r
%, h, T
= VaR
r*
%, h, T
*
T
returns volatility adjusted returns
-0.0045 24% =K4/L4
0.0965 26% =K5/L5
0.0097 25% =K6/L6
63
2011 Emese Lazar, ICMA Centre
VaR Scaling
Often VaR
10-day
(or VaR
1 month
) is needed, but there might be not
enough data to compute these using 10-daily (or monthly) returns
Since returns are not assumed to be iid normally distributed, the
square-root of time scaling would be misleading
1. An alternative is to compute 1-day VaR and then transform this
into h-day VaR using scaling
2. A second alternative is to transform 1-day returns into h-day
returns (using scaling for each component of the portfolio), and
then compute h-day VaR
Scaling: finding a scale exponent
1
(xi) such as:
where X
%,h
is the % percentile for h-day returns
1/
%, %,1 h
X h X =
64
2011 Emese Lazar, ICMA Centre
VaR Scaling
Procedure to estimate
1
: for a fixed %, is the slope of the log-
log plot of holding period (vertical axis) versus the percentile ratio
(horizontal axis); the scale exponent will be
1
If
1
= 0.5 then the square-root scaling is correct; often
1
0.4
( )
( ) ( )
%, %,1
ln

ln ln
h
h
X X
=

0
0.5
1
1.5
2
2.5
3
3.5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
( ) ( )
%, %,1
ln ln
h
X X
( ) ln h
65
2011 Emese Lazar, ICMA Centre
Kernel fitting
Often historical VaR has to be estimated at extreme percentiles
(% is very small)
Obviously there is not enough data to estimate these precisely
One method is to smooth the histogram kernel fitting
This method gives an estimate for the percentiles
66
2011 Emese Lazar, ICMA Centre
Cornish-Fisher Approximation
Another method to estimate VaR is by using an estimation
based on the moments of the returns
VaR
%, h
= X
%,h
*
h

h
where:
X
%
is the estimated percentile
Z
%
is the standard normal percentile

h
is the skewness for h-day returns

h
is the excess kurtosis for h-day returns
( ) ( ) ( )
2
2 2 2
%, % % % % % %

1 3 2 5
6 24 36
= + +
h h h
h
X Z Z Z Z Z Z
67
2011 Emese Lazar, ICMA Centre
Properties
Characteristics:
Does not assumes any distribution
Relies on that future will be the same as past
Advantages: most popular VaR methodology!
Intuitive & simple
Doesnt assume any particular distribution
Allows weighting of observations to different criteria
It can model extreme events
Disadvantages:
Future behaviour might be quite different from past behaviour
Assumes stable market conditions
Lot of data is needed + How to choose the length of historical data?
68
2011 Emese Lazar, ICMA Centre
3. Normal Linear VaR (portfolio level)
For P&L:
Assumption:
P&L is the only risk factor & it is iid normally distributed
P&L
h,t
~ N(
h
,
h
2
)
Computation:
VaR
l
%, h, t
= Z
%
*
h

h
VaR
l
%, h, t
is the lower percentile of the P&L distribution:
P(P&L
h,t
VaR
l
%, h, t
) = %
Z
%
is the lower percentile of the standard normal distribution:
P(z Z
%
) = %
69
2011 Emese Lazar, ICMA Centre
Normal Linear VaR
For (relative/arithmetic or log) returns:
Assumption:
The returns are the only risk factor and are normally distributed
r
h,t
~ N(
r
h
,
r
h
2
)
where
h
= P
t
*
r
h
and
h
= P
t
*
r
h
Computation:
VaR
l
%, h, t
= P
t
* VaR
l,r
%, h, t
(relative returns)
VaR
l
%, h, t
= P
t
* (exp [ VaR
l,r
%, h, t
] 1) (log returns)
VaR
l,r
%, h, t
= Z
%
*
r
h

r
h
is the VaR for returns:
P(r
h,t
VaR
l,r
%, h, t
) = %
70
2011 Emese Lazar, ICMA Centre
Normal Linear VaR
Standard normal
density function
Density function of
P&L
-Z% 0
-VaR%,h 0
Z
%
= - NORMSINV(%)
71
2011 Emese Lazar, ICMA Centre
Zero mean
P&L
h,t
~ N(0,
h
2
):
VaR
l
%, h, t
= Z
%
*
h
r
h,t
~ N(0,
r
h
2
)
VaR
l
%, h, t
= P
t
* VaR
l, r
%, h, t
for r
h,t
= (P
t+h
P
t
) / P
t
VaR
l
%, h, t
= P
t
* (exp[VaR
l, r
%, h, t
] 1) for r
h,t
= ln (P
t+h
/ P
t
)
where VaR
l, r
%, h, t
= Z
%
*
r
h
is the VaR for returns
In practice the effect of mean becomes significant when h > 1 month
72
2011 Emese Lazar, ICMA Centre
Properties
Characteristics:
Assumes normal distribution
Relies on parameters estimated from historical data
Advantages:
Simple, easy to use
Disadvantages: only for linear portfolios!
Assumes normality
Assumes stable market conditions
Not suitable when there are discontinuous payoffs in the portfolio
73
2011 Emese Lazar, ICMA Centre
Example 1
Compute the VaR for a portfolio for h = 1 day and 10 days
% = 1%;
1 day
= 5,000;
1 day
= 20,000
VaR
l
1%, 1 day
= Z
1%
*
1

1
= 2.32634 * 20,000 5,000
= 41, 527
VaR
l
1%, 10 days
= Z
1%
*
10

10
= 2.32634 * 20,000 * 10 5,000 * 10
= 97, 131
74
2011 Emese Lazar, ICMA Centre
Example 2
Compute VaR for a portfolio of P = 1 mil for h = 1 day and 10 days
% = 1%;
r
1 day
= 0.005;
r
1 day
= 0.02
VaR
l
1%, 1 day
= P * VaR
l,r
1%, 1 day
= 1 mil * [Z
1%
*
r
1

r
1
]
= 1 mil * [2.32634 * 0.02 0.005]
= 41, 527
VaR
l
1%, 10 days
= P * VaR
l,r
1%, 10 days
= 1 mil * [Z
1%
*
r
10

r
10
]
= 1 mil * [2.32634 * 0.02 * 10 0.005 * 10]
= 97, 131
75
2011 Emese Lazar, ICMA Centre
4. Monte Carlo VaR
It is a general framework
Assumptions:
What the risk factors are
The risk factors follow a certain (normal?) distribution with
some parameters
The risk factors affect the portfolio in a certain way
Computation:
1. We simulate a large number of possible outcomes for the rf
(multivariate normal using Cholesky decomposition),
2. Each of these leading to a possible P&L;
3. These will lead to a density for the h-day P&L
The VaR
MC
%, h, t
is the lower % percentile of this density
( )
1

l
t jt jt
j
P s f
=
=
76
2011 Emese Lazar, ICMA Centre
Monte Carlo VaR
Risk
factor 1
Portfolio(Rf1, Rf2,,Rfn)
Risk
factor 2
.
.
.
Risk
factor n
-VaR%
77
2011 Emese Lazar, ICMA Centre
Properties
Characteristics:
Assumes normal distribution (can be changed)
Relies on parameters estimated from historical data
Advantages: used mostly for options portfolios!
It can capture different market behaviours
It can model complicated dependencies
It can model the tails of the payoff distributions
It can model extreme events
Disadvantages:
Computer intensive
Assumes a standard distribution
Assumes stable market conditions
What parameters to use?
78
2011 Emese Lazar, ICMA Centre
Applications
When risk factors have non-normal distributions so analytical
solutions are not available (e.g. credit-related factors)
For non-standard risk factors (e.g. mortgages)
When the dependency of portfolios on risk factors is more
complex (e.g. options)
For diversified portfolios with many products
When simulations are preferred
79
2011 Emese Lazar, ICMA Centre
Example Monte Carlo VaR (portfolio level)
How much is the simulated VaR of a stock price after a period of
time T?
We assume the stock price follows:
where Wis a Brownian motion
Discretizing we get:
where is a standard normal variable
We divide the time period 0 T into n small steps of length t
We assume the starting stock price is S
0
= 1
By simulating n standard normal variables we get one final stock price S
T
By redoing this simulation thousands of times we get an entire simulated
density for S
T
The VaR
MC
%, h, t
will be the lower % percentile of this density
( ) ( )
( )
2
2 / t t
S t t S t e
+
+ =
( ) ( ) dS t /S t dt dW = +
80
2011 Emese Lazar, ICMA Centre
Example
0
50
100
150
200
0 10 20 30 40 50 60 70 80 90 100
81
2011 Emese Lazar, ICMA Centre
Example
0%
1%
2%
3%
4%
5%
6%
7%
8%
0 50 100 150 200 250 300
-VaR1%
82
2011 Emese Lazar, ICMA Centre
5. Methodology comparison
Normal Linear VaR and MC VaR, based on the same assumptions, should
lead to similar VaR estimates in linear portfolios
Linear and MC VaR are based on too restrictive assumptions (normality)
Historical VaR is not based on normality, but it assumes that future
behaviour will be the same as the past
Historical VaR Linear VaR Monte Carlo VaR
Risk factor distribution No assumption Normal Normal
P&L distribution Empirical Normal Simulated
Requires covariance matrix No Yes Yes
Risk factor i.i.d Yes Yes Yes
83
2011 Emese Lazar, ICMA Centre
6. Advantages of VaR
1. Simple & easy, it expresses risk in one number
2. It tells you how much is the expected maximum loss
3. It can be extended to other types of risks (credit, operational)
It gives a unified framework for risk management
4. It can be broken down for units of the firm
5. It can take account of specific risk
6. VaRs can be compared across units
7. VaRs can be aggregated by using correlations
Useful for capital allocation
84
2011 Emese Lazar, ICMA Centre
Disadvantages of VaR
1. It assumes stable market conditions, constant vol. & corr.
2. It only captures short-term risks
3. Many models assume that returns are i.i.d./normally distributed
4. It does not take into account all sources of risk (interest rates, volatility,
liquidity)
5. Different models lead to different estimates (even if they make the same
assumptions)
6. It is based on end-of-the-day positions (it does not take into account intra-
day movements)
7. It is not sub-additive: VaR
P1+P2
VaR
P1
+ VaR
P2
(it can be super-additive)
It is not a coherent measure
85
2011 Emese Lazar, ICMA Centre
7. Expected shortfall
Average of the losses worse than VaR
ES
%, h, t
= E[X| X VaR
%, h, t
]
Interpretation: it is the average loss that occurs in the %
most extreme cases (worst losses)
e.g., out of 1000 losses, the 5% VaR is the smallest of the 50 largest losses,
whilst ES is the average of the 50 largest losses
It is a coherent measure:
ES
a*P1+b*P2
a * ES
P1
+ b * ES
P2
Also called conditional VaR or expected tail loss (ETL)
86
8. Types of VaR
Stand-alone VaR
Marginal VaR
Incremental VaR
87
Stand-alone VaR
All the VaR measures discussed so far are stand-alone VaR
estimates
This is the VaR of a component (of a portfolio) taken
individually/in isolation, ignoring that this is part of a portfolio
As seen before, stand-alone VaR is not additive (except normal
linear VaR when the components are perfectly correlated), but
they can be aggregated using correlations:
Stand-alone VaR can be used to compare the performance of
different activities
2011 Emese Lazar, ICMA Centre
2 2
1 2 12 1 2
2
total
VaR VaR VaR VaR VaR = + +
88
Marginal VaR
Marginal VaR assigns a proportion of total VaR to each
component (or factor)
It measures the relative risk contribution of a specific
component (or risk factor) to the total VaR, taking into
account the correlation between the components (or factors)
P&L = s
i
RF
i,t
We also have VaR = s
i
f
i
with f
i
= VaR / s
i
We define VaR
i
= s
i
f
i
the i
th
marginal VaR
By definition, for linear portfolios only marginal normal linear
VaR is additive (otherwise only approximately)! VaR = VaR
i
Thus, it can be used for capital allocation purposes
2011 Emese Lazar, ICMA Centre
89
Incremental VaR
Incremental VaR is defined as the VaR impact of a small
new trade
This is the change in VaR for a given change in sensitivities
(for example, greeks)
We know that VaR = s
i
f
i
with f
i
= VaR / s
i
A new trade / transaction / addition to the portfolio will
change all the sensitivities s
i
= s
i
*
s
i
for all i
This will lead to a change in VaR:
VaR = VaR
*
VaR = s
i
f
i
This is incremental VaR
2011 Emese Lazar, ICMA Centre
90
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 3
VaR models part II.
91
2011 Emese Lazar, ICMA Centre
Contents
1. Backtesting
2. Stress testing
3. Model risk
92
2011 Emese Lazar, ICMA Centre
1. Backtesting
test of the performance of the risk
estimation model
Two types:
(A) Compares ex-ante VaR estimates with ex-post values of
actual P&L (assuming daily trading)
Make sure the P&L series is relevant (e.g. they can be affected by
accounting systems or mistaken entries)
(B) Compares ex-ante VaR estimates with ex-post values of a
hypothetical P&L (assuming portfolio is left unchanged)
What hypothetical P&L? (e.g. eliminate the effects of day trading)
93
2011 Emese Lazar, ICMA Centre
Backtesting
Methodology:
Compare the frequency of exceeding VaR
%
with % for
different values
Christoffersen, P.F. (1998): Evaluating Interval Forecasts,
International Economic Review 39(4), 841-862
Reasons that might lead to wrong conclusions:
The use of inappropriate / insufficient data
The use of inappropriate parameters (or incorrect estimation)
Incomplete consolidation of trading positions
Inaccurate mapping of trades to risk-equivalent positions
If the test fails to validate the model Find the problem!
94
2011 Emese Lazar, ICMA Centre
Unconditional Test
Indicator function:
Count the exceptions:
Percentage of exceptions:
Test statistic:
Test: if then the VaR model is bad

1
0
t t ;
t ;
, r VaR
I
, otherwise
<

( )

2

1
2 1
1
T n n
LR log ~

| |
| | | |

|
=
| |
|
\ \
\

1
T
t ;
t
n I
=
=

n /T =
( )
2

1
a%
LR >
95
2011 Emese Lazar, ICMA Centre
Example
-1.5
-1
-0.5
0
0.5
1
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
S&P 500 returns
5% VaR
96
2011 Emese Lazar, ICMA Centre
Example
T = 3866; = 5%; a = 10%
n
5%
= 177

5%
= 177 / 3866 = 4.6%

2
(1)
10%
= 2.71
LR
5%
<
2
(1)
10%
the VaR model is good
( )
3866 177 177
2
5
1 0 05 0 05
2 1 46 1
1 0 046 0 046
%
. .
LR log . ~
. .

| |

| | | |
= = |
| |
|

\ \
\
97
2011 Emese Lazar, ICMA Centre
2. Stress testing
A method to quantify potential extreme
negative future outcomes in a portfolio
Motivation:
1. VaR only works if its underlying assumptions are satisfied,
what if they are not?
stress situations are characterized by chain reactions and
the standard theory breaks down
2. To understand the aftermath of stress conditions
Stress testing is required for internal risk modelling
98
2011 Emese Lazar, ICMA Centre
Roles
To enhance the assessment of risk (supplement to VaR)
To compute the capital necessary to absorb
potential large losses
To identify the steps the firm can take to reduce its
risk
To lead to better decisions
99
2011 Emese Lazar, ICMA Centre
Types
Scenarios requiring no simulations (historical)
Scenarios requiring simulations
Testing current portfolios against past crashes
Testing based on assumptions about vols & correlations
Scenarios built with special purpose
These are made with a specific portfolio in mind
(e.g. in a specific country or that depends on a specific counterparty)
100
2011 Emese Lazar, ICMA Centre
Most common stress scenarios
Related to asset classes:
Stocks: - 1987 crash
- widening of credit spreads
- hypothetical crashes
Interest rates: bond market crash in 1994
Commodities: Middle East crisis
Related to geographic regions:
Emerging markets: Asian crisis
FX: USD weakening/strengthening
101
2011 Emese Lazar, ICMA Centre
The object of stress tests
Portfolios most general
Trading desks
Traders
Individual positions
these can be aggregated, but it is not preferred
102
2011 Emese Lazar, ICMA Centre
(A) Historical scenarios
Scenario shocks taken from historical events
Required by Basel
Methodology:
1. Choose an event period (a crisis period or a statistical choice)
- it can be more than one event
- the start and end date are not obvious, but usually the
peaks occur at the start date and the trough at the end date
- it is assumed that positions cannot be traded or hedged
2. Specify shock factors (generally relative; IR - absolute)
- often do PCA & shocks to first 3 components; it reduces
dimension & factors are orthogonal
- for missing shocks: use proxies or interpolations
103
2011 Emese Lazar, ICMA Centre
(B) Hypothetical scenarios
3 types:
Modifying the covariance matrix
Creating events (specifying factor shocks)
Sensitivity analysis
+ hybrid models
104
2011 Emese Lazar, ICMA Centre
(B.1) Covariance matrix
This method modifies the covariance matrix
based on that covariances change in stressful markets (2 states);
however, increased correlations dont always lead to losses
Attention! Correlation matrices always have to be such that variances
are always positive (the covariance matrix is positive definite)
If you change one value from the correlation matrix, the entire matrix
will change
Alternative: change the returns, not the correlations
Stress tests can be constructed from this modified covariance
matrix in several ways
e.g. we compute what loss is obtained by a one-standard-deviation change in the
portfolio value, then multiply this by a number (as many st. dev. changes as desired)
105
2011 Emese Lazar, ICMA Centre
Covariance matrix
r = returns V = covariance matrix
We want to find an equivalent set of returns for stressful markets
r
*
= ? W= stress covariance matrix
Solution:
C = Cholesky matrix of V (C triangular so that V = C C)
D = Cholesky matrix of W (D triangular so that W = D D)
r* = D C
1
r
106
2011 Emese Lazar, ICMA Centre
(B.2) Create an event
Based on specifying hypothetical shocks to the market factors
get ideas from historical prices
We have to specify which no-arbitrage relationships hold
(these are assumptions based on history)
this will help in specifying chain reactions
Example:
war disruption of oil production spike in oil price
spike in energy products price expectation of
future inflation increases
107
2011 Emese Lazar, ICMA Centre
Example - liquidity
Liquidity is linked to the mechanisms that tie together
institutions
Leverage & risk management have an important role:
Collateral requirements
Risk management policies (e.g. risk limits)
Position transparency & risk disclosures (herd behaviour)
OTC derivatives markets (synthetic hedges)
These provide safety in normal periods, but can be a
potential danger in crisis periods (due to liquidity problems)
lack of liquidity increased prices losses
108
2011 Emese Lazar, ICMA Centre
(B.3) Sensitivity analysis
Simple artificial portfolio shocks are created
At most a few risk factors are shocked
Several strengths of shocks are considered (volatility determines the range)
Correlation is ignored
They only provide a partial picture
Care with interpretation!
E.g. a 10% drop in equity prices
109
2011 Emese Lazar, ICMA Centre
Regulators Recommendations
To calculate risk capital:
8% for equity indices & FX rates
15% for commodities
100 basis points for the short maturity interest rates
60 basis points for the long maturity interest rates
Volatility shifts (to the volatility term structure):
Large shifts (30%) for very short maturities
Smaller shift for longer maturities
(8% for 1 year and 3% for 5 years)
110
2011 Emese Lazar, ICMA Centre
(C) Algorithmic
The goal is to create a search algorithm to identify the
worst outcome for the portfolio within some feasible set
Tasks:
Only include relationships that are relevant for identification
The algorithm has to be able to identify the worst outcome
The feasible set should include only plausible outcomes
Attention: the rule that stronger shock more loss does not apply!
111
2011 Emese Lazar, ICMA Centre
(C.1) Factor-push
It pushes the risk factors in the direction that results in a
portfolio loss
Ignores correlations!
Steps:
1. Choose a push magnitude (x times st. dev.) m
2. Create two shocks: s
1
= m and s
2
= m
3. Compute the portfolio value in case of these two shocks
4. Compare the two portfolio values and the shock that leads
to smaller portfolio value is chosen for the stress test
5. Repeat this for all risk factors
6. Revalue portfolio when all shocks are applied simultaneously
112
2011 Emese Lazar, ICMA Centre
(C.2) Maximum loss
It is the set of changes in the risk factors that result in the
greatest portfolio risk, subject to feasibility constraints
dependent on the structure of the portfolio
not dependent on the economic environment
Takes account of correlations!
Steps:
1. Specify feasible scenarios: that have a likelihood > a small number
we have a probability associated with the loss!
2. Specify joint probabilities for scenarios
3. Search the space of feasible scenarios for the set of risk
factors that will lead to the largest loss
113
2011 Emese Lazar, ICMA Centre
Example
114
2011 Emese Lazar, ICMA Centre
Comparison
Approach Description Advantages Disadvantages
Historical Repeat crisis Taken from history Many possible shocks
scenarios Difficult to interpret
No guarantee that
this is the worst case
Hypothetical Covariance matrix Easy Mixed empirical support
scenarios Create event Flexible No guarantee that
this is the worst case
Sensitivity analysis Detailed Limited risk information
Algorithmic Factor push Minimal qualitative No guarantee that
elements this is the worst case
Ignores correlations
Maximum loss Identifies worst case Assumes data from normal
in feasible set periods are relevant
Computationally intensive
115
2011 Emese Lazar, ICMA Centre
3. Model risk
the possibility of wrong estimates due to
the use of an inappropriate model
VaR estimates depend on many factors
One important factor is the model used
A model is a framework that enables the computation of an
output based on assumptions
it is a representation of reality
If the model is not a good representation of market
circumstances, then the VaR estimate can be wrong
116
2011 Emese Lazar, ICMA Centre
Model risk
It is the difference between:
The risk estimate based on a model
True risk
It can never be eliminated, but it should be reduced
to an acceptable level
117
2011 Emese Lazar, ICMA Centre
Sources of model risk
Incorrect model specifications
Wrong assumptions (normal distribution / constant volatility)
Missing factors
Misspecified relationships
Transaction costs and liquidiy factors
Incorrect model application
Implementation risk (clean the input)
Other sources:
Incorrect parameters
Programming problems
Data problems (lack of data, incorrect data etc.)
118
2011 Emese Lazar, ICMA Centre
Solutions
Be aware of model risk
Identify, evaluate & check the assumptions
Choose the simplest reasonable model
Backtest & stress-test
Estimate model risk quantitatively
Re-evaluate models periodically
119
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 4
Advanced VaR models
120
2011 Emese Lazar, ICMA Centre
Contents
1. Motivation
2. Volatility
2.1 Historical
2.2 EWMA
2.3 GARCH
3. Correlation
3.1 Historical
3.2 EWMA
3.3 GARCH
3.4 Size reduction - PCA
4. Non-normal distributions
121
2011 Emese Lazar, ICMA Centre
1. Motivation
Returns: conditionally & unconditionally non-normal
Models: based on conditional normal distribution (in general)
Solution: Having a time-varying volatility AND/OR conditional
non-normality can capture unconditional non-normality (skewness
and excess kurtosis)
S&P 500 Returns
Number of observations 3863
Mean 0.00033
Standard deviation (daily) 1.00%
Volatility (annual) 15.83%
Skewness -0.095
Excess kurtosis 4.19
Number of observations below the 65
lower bound of 99% confidence interval (vs. 39 expected)
122
2011 Emese Lazar, ICMA Centre
Time-varying volatility





123
2011 Emese Lazar, ICMA Centre
Conditional non-normality





124
2011 Emese Lazar, ICMA Centre
2. Volatility
THE most important input into risk measurement models
time-varying
autocorrelated (volatility clustering)
very high in times of market crashes
S&P 500 returns
-1.5
-1
-0.5
0
0.5
1
1991 1993 1995 1997 1999 2001 2003 2005
125
2011 Emese Lazar, ICMA Centre
2.1. Historical volatility
Volatility is estimated as the standard deviation of the returns
over a given period of time (the average of squared returns)
It assumes constant volatility, but it results in time-varying
volatility when the window is rolled over time
-0.131
0.155
0.0085
-0.005
0.1303 =STDEV(K4:K103)
0.0353 =STDEV(K5:K104)
-0.051 =STDEV(K6:K105)
4%
6%
8%
10%
12%
14%
16%
18%
Jan-04 Jul-04 Jan-05 Jul-05
1 month hist vol
3 months hist vol
1 year hist vol
126
2011 Emese Lazar, ICMA Centre
Time horizon
What time horizon to use?
- Subjective choice, depends on objectives
- It depends on the time horizon of your forecast
if you forecast for the next x months, then it is useful to use
the last x months of data
- Generally, at least 1 year of data is required (3-5 year data is
recommended)
127
2011 Emese Lazar, ICMA Centre
Interpretation
1. Interpretation of historical volatility
- It gives the expected range for future volatility
1. compute volatility using a period without extreme returns;
this will give the lower boundary of volatility
2. compute volatility using a period with many extreme returns;
this will give the upper boundary of volatility
- Disadvantage: Extreme returns have the same effect on each
volatility estimate for a period equal with the window length
128
2011 Emese Lazar, ICMA Centre
2.2. EWMA volatility
EWMA = Exponentially Weighted Moving Average
Volatility is estimated as the weighted average of the squared
returns:
In recursive form:
This model puts more weight on more recent observations
Returns have an immediate impact on the next volatility estimate
after which their effect slowly diminishes over time
Suitable for short horizon forecasts only (because forecasts are constant)
=

=

1
2 1 2
1
i
i t
i
t
r ) (

2
1
2
1
2
1

+ =
t t t

r ) (

129
2011 Emese Lazar, ICMA Centre
EWMA volatility
4%
6%
8%
10%
12%
14%
16%
18%
Jan-04 Jul-04 Jan-05 Jul-05
1 month hist vol
EWMA vol with lambda = 0.9
EWMA vol with lambda = 0.98
-0.181 0.0251
-0.221 0.0258
-0.045 0.0282
-0.276 =lambda*L6+(1-lambda)*K6^2
-0.027 =lambda*L7+(1-lambda)*K7^2
-0.172 =lambda*L8+(1-lambda)*K8^2
130
2011 Emese Lazar, ICMA Centre
The smoothing constant
Lambda is the smoothing constant between 0 and 1
Interpretation: In the recursive form equation we have 2 terms
The first one (1 ) gives the intensity of reaction to market event
The second ( ) gives the persistence in volatility
The influence of on volatility:
If is high then the volatility is smoother, undisturbed by extreme returns
and the longer and smaller the effect of extreme returns
If is low then the volatility is spikier, reacting fast to extreme returns
Which to use:
It is difficult to estimate by standard statistical methods
Generally it is between 0.85 (for short-term forecasts) and 0.98 (for long-
term forecast); RiskMetrics suggests 0.94 for daily data
131
2011 Emese Lazar, ICMA Centre
2.3. GARCH volatility
GARCH = Generalized Autoregressive Conditional
Heteroscedasticity
Todays volatility depends on yesterdays volatility (persistence )
and yesterdays squared returns (reaction )
, , > 0
Similar to EWMA, but - here we have a constant
- the sum of and is less than 1
The long-term variance is
2
= / (1 )
(the variance converges towards this)
Suitable for longer horizons as the volatility is assumed to be time-varying
2 2 2
1 1

t t t

r

= + +
132
2011 Emese Lazar, ICMA Centre
GARCH volatility
4%
6%
8%
10%
12%
14%
16%
18%
Jan-04 Jul-04 Jan-05 Jul-05
1 month hist vol
EWMA vol with lambda = 0.94
GARCH vol
= 0.0001; = 0.05; = 0.94
133
2011 Emese Lazar, ICMA Centre
Application of EWMA & GARCH for VaR
Historical VaR:
Returns are standardised using EWMA/GARCH volatility
z
t
= r
t
/
t
VaR
*
%, h, t
is historical VaR for the standardised returns z
t
VaR
%, h, t
= VaR
*
%, h, t
*
t
Normal Linear VaR: similar to above (or use linear VaR
formula for each day)
Monte Carlo VaR:
At every time step the EWMA/GARCH volatility
t
is forecasted
A variable z
t
~ N(0,1) is simulated, and r
t
=
t
* z
t
Repeat this for every step, all paths density VaR
134
2011 Emese Lazar, ICMA Centre
Comparison of models
Skewness estimates got worse (but that was not the purpose)
Excess kurtosis got better!
Percentile fit also improved!
Models with volatility clustering & time-varying volatility
GARCH seems to be the best alternative
Still not good enough!
S&P 500 Returns - standardized by vol: Constant Hist (1 m) EWMA (0.94) GARCH
Number of observations = 3863
Skewness -0.095 -0.413 -0.371 -0.369
Excess kurtosis 4.19 2.36 2.07 2.01
Number of observations below the lower 65 78 70 60
bound of 99% confidence interval (39 expected)
135
2011 Emese Lazar, ICMA Centre
3. Correlation
V = D C D
V = covariance matrix
D = diagonal matrix of standard deviations
C = correlation matrix
2
1 12 1 1 12 1
2
2 21 2 21 2 2
2
1 2
1 2
0 0 1
0 0 1
0 0

n n
n n
n n n
n n n
....... ....... .......
....... ....... .......
. . . . . .
. ... ..... . . ... ..... . . ... ..... .
. . . . . .
....... ..
.......
| |
| |
|
|
|
|
|
|
|
= |
|
|
|
|
|
|
|
|
\
\
1
2
0 0
0 0
1 0 0
n
.......
.......
. .
. ... ..... .
. .
..... .......
| | | |
| |
| |
| |
| |
| |
| |
| |
| |
\ \
136
2011 Emese Lazar, ICMA Centre
Portfolios
w = (w
1
, , w
n
) portfolio weights
r = (r
1
, ,r
n
) asset returns
R = portfolio return (R = w
1
r
1
+ + w
n
r
n
) = ?
R = w r
V = covariance matrix

2
= portfolio variance =?

2
= w V w
137
2011 Emese Lazar, ICMA Centre
Example

2
(w
1
r
1
+ w
2
r
2
) = w
1
2

1
2
+ w
2
2

2
2
+ 2 w
1
w
2

12

2
=
= w V w
If
1
= 20%,
2
= 25% and
12
= 0.5 then the portfolio variance is:
So the volatility is 0.0364 = 19.08%
( )
2
1 12 1 2 1
1 2
2
2
12 1 2 2


w
w w
w
| |
| |
|
|
|
\
\
( ) ( )
0 04 0 025 0 6 0 6
0 6 0 4 0 034 0 04 0 0364
0 025 0 0625 0 4 0 4
. . . .
. . . . .
. . . .
| | | | | |
= =
| | |
\ \ \
138
2011 Emese Lazar, ICMA Centre
Correlation
THE most important input into multivariate risk models
time-varying
autocorrelated
very high in times of market crashes
0.4
0.5
0.6
0.7
0.8
0.9
0% 20% 40% 60% 80% 100%
Lower tail
C
o
r
r
e
l
a
t
i
o
n

(
S
P
,

F
T
S
E
)
139
2011 Emese Lazar, ICMA Centre
Correlation
Considerations for modelling:
Correlations are most often extracted from variance and
covariance estimates
The model must guarantee that correlation is always
between 1 and 1
Covariance matrices need to be positive semi-definite to ensure
that volatilities can be defined (eigenvalues are positive)
Correlation assumes that all relationships are linear
e.g. x and x
2
are uncorrelated, but not independent when x is normal
140
2011 Emese Lazar, ICMA Centre
3.1. Historical correlation
Correlation is estimated as the correlation coefficient of the returns
over a given period of time
It assumes constant correlation, but it results in time-varying
correlation when the window is rolled over time
+ same discussion as for hist. vol.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Jan-02 Jul-02 Jan-03
1 month hist corr
3 month hist corr
1 year hist corr
0.21373 0.12261
-0.0595 0.29569
0.35891 -0.0788
0.00703 0.11976
0.14632 0.26049 =CORREL(K3:K103,L3:L103)
0.03393 0.14182 =CORREL(K4:K104,L4:L104)
0.18738 -0.2152 =CORREL(K5:K105,L5:L105)
141
2011 Emese Lazar, ICMA Centre
3.2. EWMA correlation
Correlation is estimated based on the covariance which is the
weighted average of the cross-product of the returns and the
variances are also estimated using the EWMA model:
In recursive form:
+ same discussion as for EWMA vol.
=

=

1
2 1
1
12
1
i
i t , i t ,
i
t ,
r r ) (

1 12 1 2 1 1 12
1

+ =
t , t , t , t ,

r r ) (

t , t ,
t ,
t ,

2 1
12
12

=
142
2011 Emese Lazar, ICMA Centre
3.3. GARCH correlation
Several alternatives
Usually they estimate the time-varying variance and covariance
equations, and based on these the time-varying correlation
process is built
+ similar discussion as for GARCH vol.
2 2 2
1 1
12 12 12 1 1 2 1 12 12 1
1 2

i ,t i i i ,t i i ,t
,t ,t ,t ,t

r i ,

r r


= + + =
= + +
143
2011 Emese Lazar, ICMA Centre
3.4. Size reduction - PCA
The risk of a large portfolio with many different
types of investments has to be assessed, where some
risk factors might have very similar behaviour
Classically, this would involve the computation and
modelling of a large covariance matrix
Is it possible to approximate this large covariance
matrix using a smaller one?
2 ways:
Regressions: CAPM, APT
PCA
144
2011 Emese Lazar, ICMA Centre
Principal Component Analysis
Used when many series with similar behaviour are being analysed
Instead of modelling directly the series, the main (independent)
sources that move the series are extracted and these are analysed
The original series can be expressed with very good
approximation as linear combination of these factors
The covariance matrix of the original series can be expressed
with the help of the variances of the principal components
The advantages are:
1. size reduction: the original data can have even hundreds
of series, but usually up to 3 or 4 factors are analysed;
2. correlation can be approximated by modelling the
orthogonal factors
145
2011 Emese Lazar, ICMA Centre
PCA - Applications
Fixed income products portfolios with exposures to yield curve
PCA applied to the yield curve (to the covariance
or correlation matrix of changes in the interest rates)
r
i,t
= w
i1
P
1,t
+ w
i2
P
2,t
+ w
i3
P
3,t
Commodity portfolios with exposures to futures prices at
different maturities
PCA applied to returns on commodity futures
Derivatives portfolios with exposures to the volatility surface
PCA applied to implied volatilities (to the covariance or
correlation matrix of changes in the difference between implied
volatility from ATM volatility for different maturities)
(
I
i, K

I
i, ATM
)
t
= w
i1
P
1,t
+ w
i2
P
2,t
+ w
i3
P
3,t
146
2011 Emese Lazar, ICMA Centre
Example
Crude oil prices
10
15
20
25
30
35
40
1990 1992 1994 1996 1998 2000 2002
m2 m3 m4
m5 m6 m7
m8 m9 m10
m11 m12
147
2011 Emese Lazar, ICMA Centre
Example
From this dataset we extract 3 (orthogonal) principal
components , these will explain 93% of the total variation in the
dataset
We model the variances of the principal components by fitting
GARCH models to these factors (zero correlations for the PCs)
OGARCH model
From these we can backup volatilities and correlations for the
original series
148
2011 Emese Lazar, ICMA Centre
Example
OGARCH volatilities
0%
2%
4%
6%
8%
10%
1990 1992 1994 1996 1998 2000 2002
m2 m3 m4
m5 m6 m7
m8 m9 m10
m11 m12
149
2011 Emese Lazar, ICMA Centre
Example
OGARCH correlations of m2
0.2
0.4
0.6
0.8
1
1990 1992 1994 1996 1998 2000 2002
m2 m3 m4
m5 m6 m7
m8 m9 m10
m11 m12
150
2011 Emese Lazar, ICMA Centre
First PC
This is called TREND
It explains the part of the differences among the series that is
constant in time
e.g. if the trend increases, then all data series will increase with the same amount
Trend in the
yield curve
Time
S
e
r
i
e
s
Maturity
Y
i
e
l
d
151
2011 Emese Lazar, ICMA Centre
Second PC
This is called TILT
It explains the part of the differences among the series that
linearly increases for less correlated series
Slope of the
yield curve
Time
S
e
r
i
e
s
Maturity
Y
i
e
l
d
152
2011 Emese Lazar, ICMA Centre
Third PC
This is called CURVATURE
It explains the part of the differences among the series that
increases with less correlated series, but decreases again for very
uncorrelated series
Convexity of the
yield curve
Time
S
e
r
i
e
s Maturity
Y
i
e
l
d
153
2011 Emese Lazar, ICMA Centre
PCA Methodology
Input: X
(T * k)
= k series of normalized returns
Output: P
(T * k)
= k principal components
out of these only the first 2 or 3 will be relevant
V
*
(T * k)
= the estimate of the covariance
matrix of the returns
this will also give the correlation matrix
154
2011 Emese Lazar, ICMA Centre
PCA Methodology
V
(k * k)
= XX / T covariance matrix of X
W
(k * k)
matrix of eigenvectors of V

(k * k)
diagonal matrix of eigenvalues of V:
This gives:
VW= W
V = WW since W
-1
= W
tr() = tr(V) = k = sum of eigenvalues
155
2011 Emese Lazar, ICMA Centre
PCA Methodology
P
(T * k)
= XW principal components
these are linear combinations of the returns
(the weights are the columns of W):
p
m
= w
1m
x
1
+ w
2m
x
2
+ .+ w
km
x
k
This also means:
X = P W the returns are also linear combinations of the PCs
(the weights are the rows of W):
x
i
= w
i1
p
1
+ w
i2
p
2
+ ...... + w
ik
p
k
The covariance matrix of P:
PP/T = W' [(XX)/T] W = W'VW = W'W =
This means that principal components are orthogonal
156
2011 Emese Lazar, ICMA Centre
PCA Methodology
We get:
The variance of the i
th
PC is the i
th
eigenvalue
The proportion of the total variation in X explained by the
m
th
PC is:

m
/(sum of all eigenvalues) =
m
/k
The first r (2/3/4) PCs explain around 95% of the variation of
X, so these are enough to model the returns and its variance
P
*
the first r columns of P = the first r PCs (trend , tilt, curvature)

*
the covariance matrix of P*
W
*
the first r columns of W
Model P
*
and/or
*
(OGARCH: do GARCH on P*)
Get X
*
= P
*
W
*
and V
*
= W
*

*
W
*

157
2011 Emese Lazar, ICMA Centre
PCA - Eigenvectors
1 2 3 4 5 6 7
Eigenvector 1
Eigenvector 2
Eigenvector 3
158
2011 Emese Lazar, ICMA Centre
PCA - Eigenvectors
1 1 2 3
2 1 2 3
7 1 2 3
1 1
2
0 5 0 25 0 5
0 5 0 1
X =0.5 PC - 0.25 PC - 0.5 PC
0 5 0 25 1 5
X =0.5 PC +0 PC +1 PC
0 5 0 5 1 3
0 5 0 75 0 8
X =0.5 PC +1.25 PC - 1 PC
0 5 1 0
0 5 1 25 1
if PC increases with 1, then X increases with 0.5,
X inc
. . .
.
. . .
. . .
. . .
.
. .





=

*
7
2 1
2 7
reases with 0.5,, X increases with 0.5,
if PC increases with 1, then X decreases with 0.25,
X doesn't change,, X increases with 1.25, etc.
159
2011 Emese Lazar, ICMA Centre
PCA - Eigenvectors
If
*
the covariance matrix of P
*
is
Based on these the correlations can also be computed
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
2 2 2
2
1
2 2 2
2
2
2 2 2
2
7
12
67
0 04 0 0
0 0 016 0
0 0 0 01
0 5 0 04 0 25 0 016 0 5 0 01
0 5 0 04 0 0 016 1 0 01
0 5 0 04 1 25 0 016 1 0 01
0 5 0 5 0 04 0 25 0 0 016 0 5 1 0 01
0 5 0 5 0 04 1 1 25 0 016 0
* * * '
.
.
.
. . . . . .
. . . .
. . . . .
. . . . . . .
. . . . .


= =



= + +
= + +
= + +
= + +
= + +
V W W

* *

( ) 1 0 01 .
160
2011 Emese Lazar, ICMA Centre
4. Non-normal distributions
Central limit theorem: (law of large numbers)
The sums and averages of any random variables follow a
normal distribution
Popularity of normal distributions
But the presence of skewness and excess kurtosis (caused by herd-
behaviour etc.) necessitates the use of alternative distributions
Still, extreme events can be explained by different models, but
generally they cannot be predicted
161
2011 Emese Lazar, ICMA Centre
Non-normal distributions
(1) Students t distribution
(2) Normal mixture distribution
f (x) = p *
1
(x) + (1 p) *
2
(x)
-1 0 1
Standard normal
Normal mixture
162
2011 Emese Lazar, ICMA Centre
Normal mixture distribution
0
0.5
1
1.5
2
-1 -0.5 0 0.5 1
Normal1
Normal mixture
Normal (with the
same variance as the
normal mixture)
Normal2
163
2011 Emese Lazar, ICMA Centre
Normal mixture distribution
There is no analytical formula to compute the linear
VaR for the normal mixture density
Normal:
F( VaR
%, x days
) = %
Normal mixture:
p * F
1
( VaR
%, x days
) + (1 p) * F
2
( VaR
%, x days
) = %
In Excel the Solver facility can be used to compute the
linear VaR for the normal mixture density
164
2011 Emese Lazar, ICMA Centre
Example
Example: the returns follow a normal mixture distribution with
2 components.
Normal 1:
1
= 8%
1
= 5% p = 90%
Normal 2:
2
= 50%
2
= 20% 1 p = 10%
VaR
NM, 5%
= 0.12 Solver
VaR
N1, 5%
= 0.065
VaR
N2, 5%
= 0.297
But 0.9 * (0.065) + 0.1 * (0.297) = 0.088 0.12
So p * VaR
N1, %
+ (1 p ) * VaR
N2, %
VaR
NM, %
165
2011 Emese Lazar, ICMA Centre
Example
0
1
2
3
4
5
6
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4
Normal 1
Normal 2
Normal mixture
166
2011 Emese Lazar, ICMA Centre
Simulating normal mixture distribution
Two random uniform [0,1] series have to be simulated: x
1
and x
2
The first one is used to decide which normal distribution should be
used:
if 0 < x
1
< p then use the first normal in the mixture i = 1
otherwise use the second one i = 2
The second one is used to draw a random normal variable from the
normal distribution chosen:
x = F
1
(
i
,
2
i
, x
2
)
167
2011 Emese Lazar, ICMA Centre
Market Risk
Applied Market Risk
Lecture 5
Equities and foreign exchange
168
2011 Emese Lazar, ICMA Centre
General framework
Linear portfolios with = 0:
One asset portfolio VaR:
Historical VaR: data on portfolio value histogram for P&L VaR
Normal Linear VaR: VaR = Z

Monte Carlo VaR: normal dist. P&L simulated histogram VaR

Many assets portfolio VaR: P&L = w r
Historical VaR: data on asset values histogram for P&L VaR
Normal Linear VaR: VaR = Z

w V w
Monte Carlo VaR: normal dist. assets simulated histogram VaR
Risk factor VaR: P&L = sensitivity RF
Historical VaR: data on risk factors histogram for P&L VaR
Normal Linear VaR: VaR = Z

s V
RF
s
Monte Carlo VaR: normal dist. RF simulated histogram VaR
169
2011 Emese Lazar, ICMA Centre
RF VaR for a linear portfolio of many assets
Historical VaR: P&L Sensitivities & RFs
past evolution
Result: historical density for the P&L: P&L
t
= s
i
RF
i,t
VaR
Normal Linear VaR: P&L Sensitivities & RFs
assume normal
Result: P&L also normal: = s V
RF
s VaR = Z

MC VaR: P&L Sensitivities & RFs

simulate normal
Result: simulated density for P&L: P&L = s
i
RF
i,t
VaR
Note: we assume = 0, constant sensitivities, and ignore specific risk
170
2011 Emese Lazar, ICMA Centre
Part I - Equities
Contents
1. Traditional risk measures
2. Risk factors & mappings
3. Application for VaR
3.1 Historical VaR
3.2 Linear VaR
3.3 Monte Carlo VaR
171
2011 Emese Lazar, ICMA Centre
Equities
If a long history of data is available (min. 2 years)
then all VaR methodologies can be used (except
Monte Carlo VaR if the portfolio has many assets)
If there is missing data (e.g. only the past 3 months
are available) then the Historical VaR methodology
needs to be modified (the other ones can be used)
How to modify Historical VaR?
How to modify MC VaR for large portfolios?
172
2011 Emese Lazar, ICMA Centre
Introduction
For linear assets we have that (assuming constant sensitivities):
Y
j,t
=
j
+
1,j
RF
1t
++
k,j
RF
kt
+
j,t
The variance of asset j can be decomposed into:

2
Yj
=
j
V
RF

j
+
2
j
where V
RF
is the covariance matrix for the RFs
The covariance matrix for several assets is:
V
Y
= V
RF
+ V

For a portfolio with investments w, the portfolios total variance

can be written as:
w V
Y
w = w V
RF
w + w V

w
The RFs will give the systematic or Risk Factor VaR, the error
term will give the Specific VaR, and the two together will give
the Total VaR
173
2011 Emese Lazar, ICMA Centre
VaR for equity portfolios
Historical VaR: P&L Sensitivities & RFs
past evolution
Result: historical density for the P&L: P&L
t
= w
i

i
RF
i,t
VaR
Normal Linear VaR: P&L Sensitivities & RFs
assume normal
Result: P&L also normal: = w V
RF
w VaR = Z

MC VaR: P&L Sensitivities & RFs

simulate normal
Result: simulated density for P&L: P&L = w
i

i
RF
i,t
VaR
Note: we assume = 0, constant sensitivities, and ignore specific risk
174
2011 Emese Lazar, ICMA Centre
1. Traditional risk measures - CAPM
Capital Asset Pricing Model (CAPM)
a simple example of a factor model
Y
j,t
=
j
+
j
X
t
+
j,t
Y
j,t
= a stock or portfolio return
X
t
= market index return in excess of r
f
(RF)

j
,
j
= constants (
j
= sensitivity)

j
= the error term with volatility
j
(specific risk)
Given historical data on X and Y
j
, we can estimate
j
,
j
and
j
175
2011 Emese Lazar, ICMA Centre
Beta
It measures how the stock responds to changes in the factor: (1%
fall in the factor is expected to be matched by a % fall in the stock)
If is insignificantly different from zero the factor (the index)
has no statistical effect on the stock
If is insignificantly different from 1 then changes in the
factor are matched exactly by changes in the stock
Stocks with < 1 are regarded as 'low risk' investments,
> 1 are regarded as 'high risk'
176
2011 Emese Lazar, ICMA Centre
Irreducible risk
only represents the irreducible part of portfolio risk
This risk that cannot be hedged away by
holding a large and diversified portfolio
ignores the risks arising from
movements in the underlying risk factors:
i.e. it ignores the variance of the index returns
the specific risks of a portfolio:
i.e. it ignores the variance of the error
Portfolio beta is: = w
j

j
177
2011 Emese Lazar, ICMA Centre
Variance decomposition
The variance of the stock returns can be
decomposed into:

2
Yj
=
j
2

2
X
+
2
j
Terminology:

Yj
is the total risk

X
is the systematic risk

j
is the specific risk
178
2011 Emese Lazar, ICMA Centre
Example
A stock has a beta of 0.8. The market volatility is 18%
and the stocks total volatility is 21%. How much is
the specific risk of the stock?
Solution:
(0.21)
2
= (0.8)
2
(0.18)
2
+
2

So the specific risk is

= 15%
179
2011 Emese Lazar, ICMA Centre
Traditional risk measures - APT
Multifactor Models (APT)
a simple example of a factor model
Y
j,t
=
j
+
1,j
X
1t
++
k,j
X
kt
+
j,t
=
j
+
j
X
t
+
jt
Y
j,t
= a stock or portfolio return
X
it
= macro/micro variables low correlation! (RF)

j
,
i,j
= constants (
j,i
= sensitivities)

j
= the error term (specific risk)
Given historical data on X
i
and Y
j
, we can estimate
j
,
i,j
and
j
The parameters
i,j
measure the sensitivities to the risk factors
180
2011 Emese Lazar, ICMA Centre
APT
Portfolio parameters:
i
= w
j

i,j
Example:
Y
1,t
=
1
+
1,1
X
1t
+
2,1
X
2t
+
t
Y
2,t
=
2
+
1,2
X
1t
+
2,2
X
2t
+
t
If we buy 100 of stocks Y
1
and 200 of stocks Y
2
then the sensitivity of the portfolio to the Xs will be:

1
= 100
1,1
+ 200
1,2

2
= 100
2,1
+ 200
2,2
181
2011 Emese Lazar, ICMA Centre
Variance decomposition
The variance of the stock returns can be
decomposed into:

2
Yj
=
j
V
X

j
+
2
j
where V
X
is the covariance matrix of the risk factors
Terminology:

Yj
is the total risk

j
V
X

j
is the systematic risk

j
is the specific risk
182
2011 Emese Lazar, ICMA Centre
Notation for many assets
Stock returns:
Y
t
= + X
t
+
t
Variance decomposition for Y:
V
Y
= V
X
+ V

Portfolio:
P&L
t
= w Y
t
= w + w X
t
+ w
t
= w +s X
t
(ignoring specific risk)

2
= w V
Y
w = w V
X
w + w V

w
= s V
X
s (ignoring specific risk)
s = w sensitivities
183
2011 Emese Lazar, ICMA Centre
2. Risk factors & mappings
Risk factors:
Market indices
Macro- and microeconomic variables
Mapping the portfolios to the risk factors:
Define factor model (e.g. CAPM / APT)
Estimate sensitivities (betas)
184
2011 Emese Lazar, ICMA Centre
3.1. Application: Historical VaR for APT
Y
j,t
=
j
+
j
X
t
+
j,t
Y
t
= + X
t
+
t
P&L
t
= w Y
t
= w + w X
t
+ w
t
u
t
v
t
Procedure:
We have a historical distribution for X
t
,
so historical distribution for u
t
We know that
t
follows a N(0,
2

) density,
so v
t
follows a normal density
This will give the approximated historical distribution for P&L
t
,
so now we can compute VaR
This procedure is called convolution
185
2011 Emese Lazar, ICMA Centre
Convolution
Convolution = getting the distribution of u + v, when the
distributions of u and v are known
Terminology: we say that the distribution of u + v is the
convolution of the distribution of u and the distribution of v
Example: if u and v follow a normal distribution, then u + v also
follows a normal distribution
186
2011 Emese Lazar, ICMA Centre
3.2. Application: Normal Linear VaR for CAPM
Assume your portfolio consists of a large number of stocks.
To compute the VaR we first need to compute the volatility
of the portfolio
Traditionally, for this we would need a large covariance matrix
for which we have to estimate thousands of correlations
The CAPM model offers a shortcut for the Systematic VaR:
VaR
%, h
= Z
%
*
m,h
* w w
= Z
%
*
m,h
* w
j

j
where
m,h
is the h-day volatility of the market and
w is the vector amounts invested in each stock
is the vector of asset sensitivities
187
2011 Emese Lazar, ICMA Centre
Example of CAPM
A portfolio consists of 10 mil investment in stock A
with a sensitivity relative to the FTSE of
1
= 0.8 and
20 mil into stock B with a sensitivity of
2
= 1.2.
Assuming that the FTSE has volatility 20% compute
the 1% 20-day Systematic VaR (ignoring specific risk).
VaR
1%, 20 days
= Z
1%
*
m, 20 days
* (w
1

1
+ w
2

2
)
= 2.33 * 20/250 * 0.2 * (10 * 0.8 + 20 * 1.2)
= 2.33 * 20/250 * 0.2 * 32
= 4.217 mil
188
2011 Emese Lazar, ICMA Centre
Example of CAPM
The individual systematic VaRs are:
VaR
1, 1%, 20 days
= Z
1%
*
m, 20 days
* 0.8 * 10
= 1.054 mil
VaR
2, 1%, 20 days
= Z
1%
*
m, 20 days
* 1.2 * 20
= 3.163 mil
We can see that in this case only the systematic VaR is additive:
VaR
1%, 20 days
= VaR
1, 1%, 20 days
+ VaR
2, 1%, 20 days
189
2011 Emese Lazar, ICMA Centre
Total VaR
However, if the error term is taken into consideration, then the
total VaR can be decomposed into:
(Total VaR)
2
= (Systematic VaR)
2
+ (Specific VaR)
2
This relationship holds only if the returns and error terms follow a
normal distribution (only for linear VaR).
Since w V
Y
w = w V
X
w + w V

w
Systematic VaR
%, h
= Z
%
*
m,h
* w w
Total VaR
%, h
= Z
%
* w V
Y
w
Specific VaR
%, h
= Z
%
* w V

w
190
2011 Emese Lazar, ICMA Centre
Application: Normal Linear VaR for APT
Risk decomposition:

2
Yj
=
j
V
X

j
+
2

where
j
is the vector of sensitivities of Y
j
V
X
is the covariance matrix of the risk factors
The variance of the error term can be ignored by
diversification so the variance can be approximated as

2
Yj

j
V
X

j
The covariances can be approximated as:

Yi Yj

i
V
X

j
191
2011 Emese Lazar, ICMA Centre
Application: Linear VaR for APT
The covariance matrix can be approximated by:
V
Y
= V
X
=
2
1 1 2 1 11 21 1 11 12 1
2
12 22 2 21 2 1 2 2 2
2
1 2
1 2




X X X X Xm n m
n X X X X Xm
m m nm
X Xm X Xm Xm
....... ....... .......
....... .......
. . . .
. ... ..... . . ... ..... .
. . . .
.......
.......
| |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\
\
2 2
1 2

m
n n nm
.......
. .
. ... ..... .
. .
.......
| |
|
|
|
|
|
|
|
|
\
192
2011 Emese Lazar, ICMA Centre
APT
Assume your portfolio consists of a large number of stocks
To compute the VaR we first need to compute the volatility of
the portfolio
Traditionally, for this we would need a large covariance matrix
for which we have to estimate thousands of correlations
The APT model offers a shortcut for systematic VaR:
VaR
%, h
= Z
%
* w V
X,h
w
where V
X,h
is the h-day covariance matrix of the RFs
w is the vector amounts invested in each stock
is the matrix of asset sensitivities
193
2011 Emese Lazar, ICMA Centre
Example of APT
You invest 10 mil into stock A, 20 mil into stock B and 30
mil into stock C. You use an APT model with two factors, these
are the FTSE and the inflation. Compute the 1% 20-day
Systematic VaR. Known are the sensitivities of the stocks to the
two factors and the covariance matrix of the factors:
VaR
1%, 20 days
= Z
1%
* 20/250 * w V
X
w
= 2.33 * 20/250 * 12.46
= 8.21 mil
0 8 0 4
0 04 0 002
0 9 0 2
0 002 0 01
1 2 0 1
X
. .
. .
. . ' V
. .
. .
| |
| |
|
= =
|
|
\
|

\
194
2011 Emese Lazar, ICMA Centre
Example of APT
The individual VaRs are:
VaR
1, 1%, 20 days
= Z
1%
* 20/250 * 10 *
1
V
X

1
= 1.11
VaR
2, 1%, 20 days
= Z
1%
* 20/250 * 20 *
2
V
X

2
= 2.41
VaR
3, 1%, 20 days
= Z
1%
* 20/250 * 30 *
3
V
X

3
= 4.73
We can see that in this case the Systematic VaR is not additive:
VaR
1%, 20days
< VaR
1, 1%, 20days
+ VaR
2, 1%, 20days
+ VaR
3, 1%, 20days
195
2011 Emese Lazar, ICMA Centre
3.3. Application: MC VaR
Assume a distribution for the risk factors and for the error
term(s) generally a normal distribution
Simulate observations for the risk factors; simultaneously
simulate observations for the error terms
These will give a simulated density for the portfolio P&L, then
simply compute the required percentile of the simulated density,
this will give the VaR
Advantage of CAPM/APT: Size reduction (small # of RFs)
196
2011 Emese Lazar, ICMA Centre
Part II Foreign Exchange
Contents
1. Traditional risk measures
2. Risk factors & mappings
3. Application for VaR
197
2011 Emese Lazar, ICMA Centre
1. Traditional risk measures
CAPM & APT with sensitivities defined relative to the
foreign index
Exposure to the FX rate where the sensitivities are
given by the amounts invested
198
2011 Emese Lazar, ICMA Centre
2. Risk factors & mappings
Risk factors (2 types):
Market indices, macro- and microeconomic variables
Exchange rates
Mapping the portfolios to the two types of risk factors:
Define factor model (e.g. CAPM / APT) and estimate
sensitivities (e.g. betas) Equity VaR
The sensitivities to the exchange rates are given by the
amounts invested FX VaR
Considering FX-equity correlations Total VaR
199
2011 Emese Lazar, ICMA Centre
Returns on foreign investments
You have P and invest it at time 0 in Germany buying shares
of stock DB
Assume the return on DB, in Euros, from time 0 to time T is r
Assume the exchange rate at time 0 is x
0
per 1
At time T you sell your shares and exchange it at rate x
T
per
1, and the return on the FX rate is r
FX
. Thus: x
T
= x
0
* e
r
FX
T
Time 0: P = P/x
0
Time T: (P/x
0
) * e
rT
= (P/x
0
) * e
rT
* x
T
=
P * e
rT
* (x
T
/x
0
) =
P * e
(r + r
FX
) T
Thus the total interest earned is r + r
FX
200
2011 Emese Lazar, ICMA Centre
Returns for foreign investments
Assume an APT model (it can be CAPM as well) for each
investment j in currency i:
Y
i
j,t
=
i
j
+
i
j
X
i
t
+
i
jt
+ r
FX, it
where
i
j
+
i
j
X
i
t
+
i
jt
is the return on the stock according to APT
r
FX, it
represents the return on the exchange rate
Thus the sensitivity to the FX rate i is given by the investment in
the foreign stock, expressed in the home currency, w
i
j, FX
= w
i
j, equity
Variance decomposition:
2
total, Yj
= V
i
systematic, j
+
2
j
where V
systematic, j
is the variance of
i
j
X
i
t
+ r
FX, it
201
Matrices
We group all investments in one country together and compute
the total investment (in equity & FX) in this country and the
average betas of this investment relative to the APT model used
in this country
w
i
equity
= w
i
j, equity
w
i
FX
= w
i
j, FX

i
= (w
i
j, equity
/ w
i
equity
)
i
j

i
= (w
i
j, equity
/ w
i
equity
)
i
j

i
= (w
i
j, equity
/ w
i
equity
)
i
j
Also, we construct a large vector which has all risk factors in all
currencies and define the betas relative to this large vector of risk
factors by using a sensitivity of 0 for an investment in a specific
country relative to the APT factors in all other countries
2011 Emese Lazar, ICMA Centre
202
Matrices
X
t
= [X
1
t
X
2
t

...
X
K
t
]
[
1
0 ... 0 ]
= [0
2
... 0 ]
...
[0 0 ...
K
]
w
equity
= [w
1
equity
w
2
equity ...
w
K
equity
]
w
FX
= [w
1
FX
w
2
FX ...
w
K
FX
]
V
X, h
and V
FX, h
are the covariance matrices of the risk factors
and the FX rates
2011 Emese Lazar, ICMA Centre
203
2011 Emese Lazar, ICMA Centre
Matrices
Weights:
Covariances (Systematic):
where Cov
equity j, FX i
=
equity j, FX i

equity i

FX j
equity
total
FX

=


w
w
w
, ,
,
, ,
h equity FX
systematic h
equity FX FX h
Cov
Cov

=


X
'V
V
V
204
2011 Emese Lazar, ICMA Centre
3.1. Applications: Historical VaR
Y
i
j,t
=
i
j
+
i
j
X
i
t
+
i
jt
+ r
FX, it
Y
t
= + X
t
+
t
+ r
FX,t
where elements represents investments in different countries
P&L
t
= w Y
t
= w + w X
t
+ w r
FX,t
+ w
t
u
t
v
t
Procedure:
We have a historical distribution for X
t
and r
FX, it
so historical distribution for u
t
We know that
t
follows a N(0,
2

) density,
so v
t
follows a normal density
This will give the approximated historical distribution for P&L
t
,
so now we can compute VaR using convolution
205
2011 Emese Lazar, ICMA Centre
3.2. Applications: Linear VaR
Equity VaR: VaR
l
equity, %, h
= Z
%
*
equity, h

equity, h
= w
equity
V
X, h
w
equity
FX VaR: VaR
l
FX, %, h
= Z
%
*
FX, h

FX, h
= w
FX
V
FX, h
w
FX
Total VaR: VaR
l
total, %, h
= Z
%
*
total, h
(systematic)
total, h
= w
total
V
total, h
w
total
Remember that VaRs do not add up, so we have that:
VaR
total
VaR
equity
+ VaR
FX
206
2011 Emese Lazar, ICMA Centre
Example 1
An American investor holds 1mil in a UK stock. The
exchange rate is 0.7 /$ and the volatility of the
exchange rate is 20%. The stock has a of 0.8 relative to
the FTSE, which has a volatility of 15%. The correlation
between the FTSE returns and the FX rate is 0.5. What
is the 10-day 1% total VaR, ignoring specific risk?
I
1
= 1 mil / 0.7 = $ 1.43 mil
VaR
equity, 1%, 10 days
= Z
1%
* 10/250 * 1.43 * 0.8 * 0.15
= $ 79,886
207
2011 Emese Lazar, ICMA Centre
Example 1
VaR
FX, 1%, 10 days
= Z
1%
* 10/250 * 1.43 * 0.2
= $ 133,143
VaR
total, 1%, 10 days
= Z
1%
* 10/250 * 1.43 *
[ (0.8 * 0.15)
2
+ 0.2
2
+ 2 * (0.8 * 0.15) * 0.2 * 0.5]
= $ 186,400
VaR
total, 1%, 10 days
< VaR
equity, 1%, 10 days
+ VaR
FX, 1%, 10 days
208
2011 Emese Lazar, ICMA Centre
Example 2
A British investor holds 1 mil in a German stock A that has a
of 0.9 relative to the DAX and 2 mil in an American stock B
that has a of 1.1 relative to the S&P 500. The volatility of the
DAX is 20%, the volatility of the S&P 500 is 18% whilst the /
and /$ exchange rates have a volatility of 15% and 25%,
respectively. The two indices have a correlation of 0.8, the DAX
has a correlation of 0.5 with the / exchange rate and the S&P
500 index has a correlation of 0.1 with the /$ FX rate, whilst the
indices are not correlated with foreign FX rates. We assume that
the two exchange rates have correlation 0.1. What are the 10-day
1% equity, FX and total VaRs, ignoring specific risk?
209
2011 Emese Lazar, ICMA Centre
Example 2
I
1
= 1 mil I
2
= 2 mil

1
= 0.9
2
= 1.1

DAX
= 20%
S&P
= 18%

/
= 15%
/$
= 25%

DAX, S&P
= 0.8

DAX, /
= 0.5
DAX, /$
= 0

S&P, /
= 0
S&P, /$
= 0.1

/$, /
= 0.1
Z
1%
= 2.33
210
2011 Emese Lazar, ICMA Centre
Example 2
w
equity
= [1 2 ] = [0.9 0 ]
[0 1.1]
V
equity, 10
= V
X,10
= 10/250 * [0.9 0 ] [0.2
2
0.2*0.18*0.8] [0.9 0 ]
[0 1.1] [0.2*0.18*0.8 0.18
2
] [0 1.1]
= [0.0013 0.00114]
[0.00114 0.00157]

equity, 10
= w
equity
V
equity, 10
w
equity
= 0.11
VaR
l
equity, 1%, 10
= Z
1%
*
equity, 10
= 2.33 * 0.11 = 256,221
w
FX
= [1 2 ]
V
FX, 10
= 10/250 * [0.15
2
0.15 * 0.25 * 0.1]
[0.15 * 0.25 * 0.1 0.25
2
]
= [ 0.0009 0.00015]
[0.00015 0.0025 ]

FX, 10
= w
FX
V
FX, 10
w
FX
= 0.107
VaR
l
FX, 1%, 10
= Z
1%
*
FX, 10
= 2.33 * 0.107 = 249,473
211
2011 Emese Lazar, ICMA Centre
Example 2
w
total
= [1 2 1 2]
V
total, 10
= [0.0013 0.00114 0.9*0.2*0.15*0.5/25 0 ]
[0.00114 0.00157 0 1.1*0.18*0.25*0.1/25 ]
[0.9*0.2*0.15*0.5/25 0 0.0009 0.00015 ]
[0 1.1*0.18*0.25*0.1/25 0.00015 0.0025 ]
= [0.0013 0.00114 0.00054 0 ]
[0.00114 0.00157 0 0.0002 ]
[0.00054 0 0.0009 0.00015 ]
[0 0.0002 0.00015 0.0025 ]

total, 10
= w
total
V
total, 10
w
total
= 0.162
VaR
l
total, 1%, 10
= Z
1%
*
total, 10
= 377,231
212
2011 Emese Lazar, ICMA Centre
Example 2
VaR
equity, 1%, 10 days
= Z
1%
* 10/250 *
[(1*0.9*0.2)
2
+(2*1.1*0.18)
2
+2 * 1*0.9*0.2 * 2 * 1.1 * 0.18 * 0.8]
= 256,221
VaR
FX, 1%, 10 days
= Z
1%
* 10/250 *
[(1*0.15)
2
+ (2*0.25)
2
+ 2 * 1 * 0.15 * 2 * 0.25 * 0.1]
= 249,473
VaR
total, 1%, 10 days
= Z
1%
* 10/250 *
[(1*0.9*0.2)
2
+(2*1.1*0.18)
2
+ 2 * 1*0.9*0.2 * 2 * 1.1 * 0.18 * 0.8
+(1*0.15)
2
+ (2*0.25)
2
+2 * 1 * 0.15 * 2 * 0.25 * 0.1
+2*1*0.9*0.2*1*0.15*0.5 + 2*2*1.1*0.18*2*0.25*0.1 ]
= 377,231
VaR
total, 1%, 10 days
< VaR
equity, 1%, 10 days
+ VaR
FX, 1%, 10 days
213
Market Risk
Lecture 6
Fixed income products
214
Contents
1. Introduction
2. Traditional risk measures
3. Risk factors & mappings
4. Application for VaR
4.1 Historical VaR
4.2 Linear VaR
4.3 Monte Carlo VaR
215
1. Introduction
For IR products (or cash flows) the risk factors are
the interest rates
For VaR calculations, quite often we ignore the
nonlinearities (convexity) of the yield curve
For a linear portfolio we assume a linear relationship
between the P&L (or cash flow, or change in the
value of an IR product) and the change in interest
rates (yield curve)
P&L
t
= Sensitivity
i
* IR
i, t
216
RF VaR for a linear portfolio IR products
Historical VaR: P&L Sensitivities & RFs=IRs
past evolution
Result: historical density for the P&L: P&L
t
= s
i
IR
i,t
VaR
Normal Linear VaR: P&L Sensitivities & RFs=IRs
assume normal
Result: P&L also normal: = s V
IR
s VaR = Z

MC VaR: P&L Sensitivities & RFs=IRs

simulate normal
Result: simulated density for P&L: P&L = s
i
IR
i,t
VaR
Note: we assume = 0, constant sensitivities, and ignore specific risk
217
2. Traditional risk measures
(A) Duration:
D = t * [ PV
t
/ PV
t
]
It is the average time that the CFs are received
It is a measure of interest rate risk
(e.g. if interest rates rise, then Duration goes down)
(B) Modified duration:
D
*
= D / (1 + y) y = IR at D
218
Traditional risk measures
(C) Present value of a basis point (PV01 or PVBP):
PV01 = D
*
/ 10,000
This gives the relative change in the PV(Investment)
when the entire yield curve goes up with 1 bp
(D) Dollar value of a basis point (DV01 or DVBP):
DV01 = PV01 * PV(Investment) = Sensitivity
This gives the absolute change in the PV when
the entire yield curve goes up with 1 bp
P&L
t
= 10,000 DV01
i
IR
i,t
219
Example
Time CF Yield PV RelPV Time * RelPV Yield
*
PV
*
1 10 11.622% 8.96 9.07% 0.091 11.632% 8.96
2 10 11.175% 8.09 8.19% 0.164 11.185% 8.09
3 10 10.785% 7.35 7.45% 0.223 10.795% 7.35
4 10 10.573% 6.69 6.78% 0.271 10.583% 6.69
5 10 10.472% 6.08 6.16% 0.308 10.482% 6.07
6 10 10.388% 5.53 5.60% 0.336 10.398% 5.52
7 10 10.301% 5.03 5.10% 0.357 10.311% 5.03
8 10 10.215% 4.59 4.65% 0.372 10.225% 4.59
9 10 10.131% 4.20 4.25% 0.382 10.141% 4.19
10 110 10.051% 42.21 42.75% 4.275 10.061% 42.18
Total 98.73 Duration 6.780 Total
*
98.67
6.78 10.32% Modified duration 6.145 PV
*
- PV -0.0607
PVBP -0.0006 PV * PVBP -0.0607
220
3. Risk factors & mappings
For cash flows the risk factors are interest rates for
different (pre-defined) maturities
Cash flow mappings need to be made to compute the
individual sensitivities to the risk factors (IRs)
If the CF is received at a vertex, then sensitivity = DV01
What to do when the CF is received between two vertices?
find an equivalent set of CF with pre-defined maturities
When there are too many risk factors (maturities),
PCA has to be applied to reduce their number
221
Cash flow characteristics
1. Amount of CF
2. Time t when it is received
3. Interest rate at time t
4. Volatility of interest rate at time t
On the other hand, we know:
Given interest rates at pre-defined time points (yield curve)
Their volatilities (usually smaller volatilities for large maturities)
What are the equivalent CFs at these vertices?
} Unknown
222
Example
1 2 3 4 5 6
c
1
c
2
c
3
c
4
c
5
c
6
q
1
, s
1
q
2
, s
2
q
3
, s
3
q
4
, s
4
q
5
, s
5
q
6
, s
6
x
1
x
2
x
3
x
4
x
5
x
6
r
1
,
1
r
2
,
2
r
3
,
3
r
4
,
4
r
5
,
5
r
6
,
6
223
Example
Quite often this can be reduced to several mappings of
the form:
t
1
t t
2
C
q, s
Cx
1
Cx
2
r
1
, 1 r
2
,
2
224
Equivalence of CF
x
1
, x
2
= ?
Criteria:
1. the sum of CF amounts stays constant
2. the PV of the CFs stays constant
3. the variance stays constant
4. the duration stays constant
5. the interest rate sensitivity stays constant
225
Equivalence of CF
The interest rate at time t can be approximated with:
q = r
1
+ (r
2
r
1
) * [ (t t
1
) / (t
2
t
1
) ]
The volatility at time t can be approximated with:
s =
1
+ (
2

1
) [ (t t
1
) / (t
2
t
1
) ]
It is possible to work with:
(A) Discrete compounding
(B) Continuous compounding
226
(A) Discrete compounding
The discount factors are (PV = CF * ):
= (1 + q)
t

1
= (1 + r
1
)
t1

2
= (1 + r
2
)
t2
The PV01s are (PV = PV01 * PV when r = 1bp):
PV01 = t / (10,000 * (1 + q))
PV01
1
= t
1
/ (10,000 * (1 + r
1
))
PV01
2
= t
2
/ (10,000 * (1 + r
2
))
227
(B) Continuous compounding
The discount factors are (PV = CF * ):
= exp( q * t)

1
= exp( r
1
* t
1
)

2
= exp( r
2
* t
2
)
The PV01s are (PV = PV01 * PV when r = 1bp):
PV01 = exp ( 0.0001 * t) 1
PV01
1
= exp ( 0.0001 * t
1
) 1
PV01
2
= exp ( 0.0001 * t
2
) 1
228
Example
C = 1 mil at time t = 1.25
Map this CF for t
1
= 1 and t
2
= 2 x
1
, x
2
= ?
VaR
1%, 10 days
= ?
r
1
= 4% and r
2
= 6% (discrete compounding)

1
= 2%,
2
= 3% and = 0.9
We get:
q = 4.5%
s = 2.25%
= 0.95,
1
= 0.96 and
2
= 0.89
PV01 = 0.00012, PV01
1
= 0.000096, PV01
2
= 0.00019
229
Criteria 1
1. CF invariant: x
1
+ x
2
= 1
Example: The following three CF mappings are all CF
invariant:
(a) x
1
= 0.5, x
2
= 0.5
(b) x
1
= 0.9, x
2
= 0.1
(c) x
1
= 0.2 , x
2
= 0.8
Note: Infinite solutions! (1 equation with 2 unknowns)
230
Criteria 2
2. PV invariant: y
1
+ y
2
= 1 y
i
= x
i
* (
i
/)
Proof: C * = C * x
1
*
1
+ C * x
2
*
2
1 = x
1
*
1
/ + x
2
*
2
/
Example: The following three CF mappings are all PV
invariant:
(a) y
1
= 0.5, y
2
= 0.5 x
1
= 0.49, x
2
= 0.53
(b) y
1
= 0.9, y
2
= 0.1 x
1
= 0.89, x
2
= 0.11
(c) y
1
= 0.2, y
2
= 0.8 x
1
= 0.20, x
2
= 0.85
Note: Infinite solutions! (1 equation with 2 unknowns)
% of PV distributed
to each vertex
231
Criteria 3
3. Volatility (variance) invariant:
s
2
= y
1
2

1
2
+ y
2
2

2
2
+2 y
1
y
2

2
= the correlation between the interest rates
Example: The following three CF mappings are all volatility
consistent:
(a) y
1
= 0.5, y
2
= 0.44 x
1
= 0.49, x
2
= 0.46
(b) y
1
= 0.9, y
2
= 0.16 x
1
= 0.89, x
2
= 0.17
(c) y
1
= 0.2, y
2
= 0.63 x
1
= 0.20, x
2
= 0.67
Note: Infinite solutions! (1 equation with 2 unknowns)
232
Criteria 3.1
3.1. PV & Volatility invariant:
y
1
+ y
2
= 1
s
2
= y
1
2

1
2
+ y
2
2

2
2
+2 y
1
y
2

2
Example: The solution is:
y
1
= 0.69, y
2
= 0.31 x
1
= 0.68, x
2
= 0.32
Note: Unique solution! (system of 2 equations with 2 unknowns)
233
Criteria 4
4. Duration invariant: t (y
1
+ y
2
) = t
1
y
1
+ t
2
y
2
Proof: t = (t
1
* C * x
1
*
1
+ t
2
* C * x
2
*
2
)/ (C * x
1
*
1
+
+ C * x
2
*
2
)
t (y
1
+ y
2
) = t
1
* y
1
+ t
2
* y
2
Example: The following CF mappings are all duration
invariant:
(a) y
1
= 0.5, y
2
= 0.17 x
1
= 0.49, x
2
= 0.18
(b) y
1
= 0.9, y
2
= 0.3 x
1
= 0.89, x
2
= 0.32
(c) y
1
= 0.2, y
2
= 0.07 x
1
= 0.20, x
2
= 0.07
Note: Infinite solutions! (1 equation with 2 unknowns)
234
Criteria 4.1
4.1. PV & Duration invariant:
y
1
+ y
2
= 1
t (y
1
+ y
2
) = t
1
y
1
+ t
2
y
2
This can be expressed as:
y
1
= (t
2
t) / (t
2
t
1
)
y
2
= (t t
1
) / (t
2
t
1
)
Example: The solution is:
y
1
= 0.75, y
2
= 0.25 x
1
= 0.74, x
2
= 0.27
Note: Unique solution! (system of 2 equations with 2 unknowns)
Note: The solution satisfies
t = t
1
x
1

1
+ t
2
x
2

2
235
Criteria 5
5. Interest-rate sensitivity (or VaR) invariant:
Idea: Want to find the mapped CFs so that the change in the PV
of the original CF when r
1
or r
2
increases with 1 bp is the same
as the change in the PV of the mapped CF.
5.1. Assuming that there are only parallel shifts in the yield curve
when r
1
increases with 1 bp, then so does r
2
5.2. Assuming that each interest rate can move independently
when r
1
increases with 1 bp, then r
2
stays unchanged
236
Criteria 5.1
5.1. Interest-rate sensitivity invariant approximate method:
the change in the PV of the original CF when the entire YC
increases with 1 bp is the same as the change in the PV of
the mapped CF
Lets assume continuous compounding
When r
1
and r
2
increase with 1 bp the changes in the PV of the original and
mapped CFs are:
Original: DV01
*
= C * [exp( (q + 0.0001) t) exp( q t)]
Mapped: DV01 = C * x
1
* [exp( (r
1
+ 0.0001) t
1
) exp( r
1
t
1
)]
+ C * x
2
* [exp( (r
2
+ 0.0001) t
2
) exp( r
2
t
2
)]
DV01
*
= DV01 x
1
, x
2
Note: Infinite solutions! (1 equation with 2 unknowns)
237
Criteria 5.1
This expression reduces to:
[exp( 0.0001 t) 1] =
= x
1

1
[exp( 0.0001 t
1
) 1] + x
2

2
[exp( 0.0001 t
2
) 1]
We use the following approximation: exp(a) 1 + a
We get:
[ 0.0001 t] = x
1

1
[ 0.0001 t
1
] + x
2

2
[ 0.0001 t
2
]
This reduces to:
t = x
1

1
t
1
+ x
2

2
t
2
or t = y
1
t
1
+ y
2
t
2
This expression is satisfied by Criteria 4.1 PV & Duration invariant mapping, so
PV & Duration invariant mapping also preserves sensitivities to interest rates
with an approximation, assuming that the yield curve changes using parallel
shifts only
Only when the YC has parallel shift only; this is not exactly true for discrete compounding
238
Criteria 5.2
5.2. Interest-rate sensitivity invariant exact method:
Want to find the mapped CFs so that the change in the PV of
the original CF when r
1
or r
2
increases with 1 bp is the same as
the change in the PV of the mapped CF.
When r
1
or r
2
increase with 1 bp then the interest rate at time t
would be:
q
*
= r
1
+ 0.0001 + (r
2
(r
1
+ 0.0001)) * [ (t t
1
) / (t
2
t
1
) ]
q
**
= r
1
+ (r
2
+ 0.0001 r
1
) * [ (t t
1
) / (t
2
t
1
) ]
Example: q
*
= 4.5075% and q
**
= 4.5025%
In the following we differentiate between discrete and continuous
compounding:
239
Criteria 5.2
(A) Discrete compounding:
When r
1
increases with 1 bp the changes in the PV of the original and
mapped CFs are:
Original: DV01
1
*
= C * (q
*
q) * 10,000 * * PV01
Mapped: DV01
1
= C * x
1
*
1
* PV01
1
DV01
1
*
= DV01
1
x
1
When r
2
increases with 1 bp the changes in the PV of the original and
mapped CFs are:
Original: DV01
2
*
= C * (q
**
q) * 10,000 * * PV01
Mapped: DV01
2
= C * x
2
*
2
* PV01
2
DV01
2
*
= DV01
2
x
2
240
Criteria 5.2
Example discrete compounding:
DV01
1
*
= 89.71 * = 84.9
DV01
1
= 96.15 * x
1
*
1
DV01
1
*
= DV01
1
x
1
= 0.92
DV01
2
*
= 29.9 * = 28.3
DV01
2
= 188.68 * x
2
*
2
DV01
2
*
= DV01
2
x
2
= 0.17
The solution is:
x
1
= 0.92, x
2
= 0.17
Note: Unique solution!
241
Criteria 5.2
(B) Continuous compounding:
When r
1
increases with 1 bp the changes in the PV of the original and
mapped CFs are:
Original: DV01
1
*
= C * [exp( q
*
t) exp( q t)]
Mapped: DV01
1
= C * x
1
* [exp( (r
1
+ 0.0001) t
1
) exp( r
1
t
1
)]
DV01
1
*
= DV01
1
x
1
When r
2
increases with 1 bp the changes in the PV of the original and
mapped CFs are:
Original: DV01
2
*
= C * [exp( q
**
t) exp( q t)]
Mapped: DV01
2
= C * x
2
* [exp( (r
2
+ 0.0001) t
2
) exp( r
2
t
2
)]
DV01
2
*
= DV01
2
x
2
242
Criteria 5.2
Example continuous compounding:
q
*
= 4.5075% q
**
=4.5025%
DV01
1
*
= C * [exp( q
*
t) exp( q t)]= 88.6
DV01
1
= C * x
1
* [exp( (r
1
+ 0.0001) t
1
) exp( r
1
t
1
)] = 96.07 * x
1
DV01
1
*
= DV01
1
x
1
= 0.92
DV01
2
*
= C * [exp( q
**
t) exp( q t)]= 29.5
DV01
2
= C * x
2
* [exp( (r
2
+ 0.0001) t
2
) exp( r
2
t
2
)] = 177.4 * x
2
DV01
2
*
= DV01
2
x
2
= 0.17
The solution is:
x
1
= 0.92, x
2
= 0.17
Note: Unique solution!
243
Mapping
Note: When mapping multiple CFs to multiple vertices (more than
two vertices) then it is possible to do it in a way that more than
two criteria are satisfied.
Example: When mapping 2 CFs to 3 vertices then it is possible to
do it in a PV-duration-volatility invariant manner.
However, when doing interest rate sensitivity invariant
mapping, then there is always a unique solution
Another advantage of VaR-invariant mapping is that
DV01s can be summed at each vertex!
244
Sensitivities
1 2 3 4 5 6
c
1
c
2
c
3
c
4
c
5
c
6
q
1
, s
1
q
2
, s
2
q
3
, s
3
q
4
, s
4
q
5
, s
5
q
6
, s
6
DV01
1
DV01
2
DV01
3
DV01
4
DV01
5
DV01
6
r
1
,
1
r
2
,
2
r
3
,
3
r
4
,
4
r
5
,
5
r
6
,
6
245
Summary
The P&L is:
P&L = 10,000 * DV01
i
* r
i
This expression will be used to compute all 3 VaRs
The volatility is (for linear VaR):
= 10,000 * s V s * h/250
where s = [DV01
1
, , DV01
n
]
V = the covariance matrix of the IRs
The Normal Linear VaR estimate is:
VaR = Z
%
*
246
4.1 Application for VaR Historical VaR
For a portfolio of IR sensitive instruments we
compute the cumulative sensitivities (DVBP) for
each vertex
Then we consider the historical series for changes in
the IRs at each vertex (r
i,t
)
Thus we can construct a historical series of P&L:
P&L
t
= 10,000 * DV01
i
* r
i,t
The VaR will be the % quantile of the distribution
of the P&L
t
series
247
4.2 Application for VaR Normal Linear VaR
For a single CF:
VaR
t, %, h
= 10,000 * Z
%
* (h/250) * DV01
t
* r
t
*
t
where r
t
= interest rate
- Z
%
= lower % percentile of the standard normal density

t
= volatility of r
t
(relative)
For multifactor (assume 2 factors) model:
VaR
%, h
= 10,000 * Z
%
* (h/250) *
(DV01
1
* r
1
*
1
)
2
+ (DV01
2
* r
2
*
2
)
2
+
2 * * (DV01
1
* r
1
*
1
)* (DV01
2
* r
2
*
2
)
248
Normal Linear VaR
Example:
Find the 1% 10-day VaR:
Discrete compounding:
VaR
1%, 10 days
= 10,000 * Z
1%
* (10/250) *
(84.9 * 0.04 * 0.02)
2
+ (28.3 * 0.05 * 0.03)
2
+ 2 * 0.9 * (84.9 * 0.04 * 0.02)* (28.3 * 0.05 * 0.03)
= 539
Continuous compounding:
VaR
1%, 10 days
= 10,000 * Z
1%
* (10/250) *
(88.6 * 0.04 * 0.02)
2
+ (29.5 * 0.05 * 0.03)
2
+ 2 * 0.9 * (88.6 * 0.04 * 0.02)* (29.5 * 0.05 * 0.03)
= 563
249
Example 2 continuous compounding
CF of 1 mil maturing in 2 years and 9 months.
Interest rates: 6% for 2 years (with volatility of 1%) and 6.5%
for 3 years (with volatility of 1.25%) with correlation 0.9.
Compute the 5% 1-year VaR.
Solution:
The interest rate for 2.75 years is:
q = r
2
* ( 3 2.75) + r
3
* (2.75 2) = 6.375%
An increase of 1bp in r
2
would result in an increase in q to:
q
*
=(r
2
+ 1bp) * ( 3 2.75) + r
3
* (2.75 2) = 6.378%
An increase of 1bp in r
3
would result in an increase in q to:
q
**
= r
2
* ( 3 2.75) + (r
3
+ 1bp) * (2.75 2) = 6.383%
250
Example 2 continuous compounding
The change in the PV for an 1bp change in r
2
and r
3
are
(original CF):
DV01
2
*
= 1 mil * (e
2.75 * 0.06378
e
2.75 * 0.06375
) = 57.69
DV01
3
*
= 1 mil * (e
2.75 * 0.06383
e
2.75 * 0.06375
) = 173.07
Now we find the CFs in year 2 and 3 equivalent to these values:
DV01
2
= 1 mil * x
2
* (e
2 * 0.0601
e
2 * 0.06
)
DV01
3
= 1 mil * x
3
* (e
3 * 0.0651
e
3 * 0.065
)
The solution is: x
2
= 0.325 and x
3
= 0.701
This means that, to keep interest rate sensitivities the same, the
original CF of 1 mil in 2.75 years can be replaced by
325,274 in 2 years and 701,202 in 3 years.
251
Example continuous compounding
VaR will be:
VaR
5%, 1 year
= 10,000 * 1.645 *
(DV01
2
* r
2
*
2
)
2
+ (DV01
3
* r
3
*
3
)
2
+
2 * * (DV01
2
* r
2
*
2
)* (DV01
3
* r
3
*
3
)
= 802
252
4.3 Application for VaR Monte Carlo VaR
For a portfolio of IR sensitive instruments we compute
the cumulative sensitivities (DVBP) for each vertex
Then, assuming a distribution (multivariate normal) for
the changes in the IRs, we simulate thousands of
observations for changes in the IRs at each vertex (r
i,t
)
(PCA can be used in case there are many IRs)
Thus we can construct a simulated series of P&L:
P&L
t
= 10,000 * DV01
i
* r
i,t
The VaR will be the % quantile of the simulated
distribution of the P&L
t
series
253
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 7
Equity and interest rate derivatives
254
2011 Emese Lazar, ICMA Centre
Contents
Part I. Equity derivatives
1. Traditional risk measures
2. Risk factors & mappings
3. Application for VaR
3.1 Historical VaR
3.2 Monte Carlo VaRmappings
4. Some named equity derivatives: basket option, index options, forwards
Part II. Interest rate derivatives
5. Traditional risk measures
6. Risk factors & mappings
7. Application for VaR
7.1 Historical VaR
7.2 Monte Carlo VaRmappings
8. Some named interest rate derivatives: swaps, swaptions
255
2011 Emese Lazar, ICMA Centre
1. Traditional risk measures
Risk factors: underlying price, volatility, interest rates, time to
maturity
The greeks measure the sensitivities of the option price to the
risk factors
The purpose is to reduce the sensitivities to different risk
factors (make them zero). Hedging a portfolio is making the
sensitivities to the main risk factors equal with zero
Traditionally traders had limits expressed in greeks, but the
problem with this is (1) they cannot be aggregated and (2)
they are only sensitivities, they ignore the risk of the risk
factors themselves
256
2011 Emese Lazar, ICMA Centre
Simple option pricing
Black-Scholes model:
Assume S follows a geometric Brownian motion
diffusion:
dS / S = dt + dW
W is a Wiener process (Brownian motion)
Discretizing using a small step t, we get:
S / S = t + with ~ N(0, t)
S = S t + S
257
2011 Emese Lazar, ICMA Centre
Simple option pricing
The BS price of a European call/put option with strike K
and maturity T at any time t [0, T] is:
where (.) is the cumulative standard normal distribution,
q is the dividend yield, r is the risk free rate and:
( )
( )
( )
( )
1 2
q T t r T t
c Se d Ke d

=
( )
( )
( )
( )
2 1
r T t q T t
p Ke d Se d

=
( )
( )
1 2 1
1

2

q T t
r T t
ln M Se
d T t d d T t M
T t Ke


= + = =

258
2011 Emese Lazar, ICMA Centre
Delta
The delta of an option is the sensitivity of the option
price to changes in the underlying price S:
= f / S
Properties:
- Delta is positive for calls and negative for puts
- Its shape becomes steeper with time to maturity
Delta call
0.0
0.2
0.4
0.6
0.8
1.0
55 65 75 85 95 105 115 125 135 145
Underlying
Delta put
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
55 65 75 85 95 105 115 125 135 145
Underlying
259
2011 Emese Lazar, ICMA Centre
Delta hedging
For every long option we have to short units of the underlying
(E.g. if = 0.6, then for 10 long call options we should go short 6 shares)
In a portfolio of many options each having a delta
i
and the
number of options being w
i
, the delta of the portfolio is the
weighted average of the deltas:
= w
i

i
This means that to make the portfolio delta-neutral (i.e. insensitive to
small movements in the underlying price), a total of underlying assets
have to be shorted
260
2011 Emese Lazar, ICMA Centre
Gamma
The gamma of an option measures the sensitivity of
delta to changes in the underlying price:
= / S =
2
f / S
2
Properties:
- Gamma is generally positive
- Gamma is largest for ATM options
- Gamma decreases towards expiry
(except ATM options)
Gamma
0.00
0.01
0.02
0.03
55 65 75 85 95 105 115 125 135 145
Underlying
261
2011 Emese Lazar, ICMA Centre
Gamma hedging
To gamma-hedge an option a position has to be taken in
another option so that the gamma of the portfolio is zero
Again, we have that the gamma of a portfolio is a weighted
average of individual gammas:
= w
i

i
To gamma-neutralize a portfolio of options, we take another
option/portfolio so that the gamma of the hedged portfolio
will be zero
Delta-gamma hedging: after gamma hedging, the portfolio
is delta hedged
262
2011 Emese Lazar, ICMA Centre
Vega
The vega of an option measures the sensitivity of the
option price to changes in the volatility (units):
= f /
Properties:
- Vega is always positive
- Vega is largest for ATM options
- Vega decreases towards expiry
Vega
0
5
10
15
20
25
55 65 75 85 95 105 115 125 135 145
Underlying
263
2011 Emese Lazar, ICMA Centre
Vega hedging
To vega-hedge an option a position has to be taken in another
option so that the vega of the portfolio is zero
The vega of a portfolio is a weighted average of individual vegas:
= w
i

i
To vega-neutralize a portfolio of options, we take another
option/portfolio so that the vega of the hedged portfolio will be
zero
Vega-gamma hedging: after vega hedging, the portfolio is delta
hedged
264
2011 Emese Lazar, ICMA Centre
Theta
The theta of an option measures the change in its
price over time:
= f / t
Properties:
- Theta is generally negative
- Theta is largest in absolute value for ATM options
- Theta decreases in absolute value as maturity approaches
(except ATM options)
265
2011 Emese Lazar, ICMA Centre
Rho
The rho of an option measures its sensitivity to
changes in the risk-free interest rate:
= f / r
Properties:
- Rho is positive for calls and negative for puts
266
2011 Emese Lazar, ICMA Centre
Kappa
The kappa of an option measures the sensitivity of
vega to changes in volatility:
= / =
2
f /
2
Properties:
- It is generally positive, but it can be negative as well when
close to ATM
267
2011 Emese Lazar, ICMA Centre
Approximating greeks
Greeks are generally used for hedging, as seen before,
but they can be useful for risk management as well
The greeks, especially for portfolios or complex
products, can be approximated by first or second
order finite differences:
For example:
[ f (S + ) f (S )] / 2
[ f (S + ) 2 f (S) + f (S )] /
2
268
2011 Emese Lazar, ICMA Centre
Position greeks
We define the following concepts (position greeks):
Position delta
P
= * N
Position gamma
P
= * N
Position vega
P
= * N etc.
Position greek = Greek * # of options
These are computed by multiplying the option greeks
with the number of the particular option in the
portfolio
These are additive for a portfolio, and after adding
them up we get the portfolios position greeks
269
2011 Emese Lazar, ICMA Centre
Example: delta-gamma hedging
For hedging use a call option O4 with = 0.3; = 0.015
Compute the net position gamma of the portfolio:

P
= 100 * 0.04 + 200 * 0.01 + (-300) * 0.03 = 3
For hedging we need to buy (+3) / 0.015 = 200 calls O4
The net position delta of the new portfolio is:

P
= 100 * 0.5 + 200 * (-0.2) + (-300) * 0.4 + 200 * 0.3 =
= 50 so we need to buy 50 shares
Option Stock Type Amount Delta Gamma
O1 A Call 100 0.5 0.04
O2 A Put 200 0.2 0.01
O3 A Call -300 0.4 0.03
270
2011 Emese Lazar, ICMA Centre
Dollar delta & gamma
We define the following concepts (dollar greeks):
Dollar delta
$
=
P
* S
Dollar gamma
$
=
P
* S
2
Dollar delta = Position delta * Price of underlying
Dollar gamma = Position gamma * (Price of underlying)
2
They give the sensitivity to asset returns
These are additive for a portfolio
271
2011 Emese Lazar, ICMA Centre
Implied volatility
Implied volatility = the volatility for which the Black-
Scholes price equals the market price (inverse of price)
It is a forecast of the average volatility over the life of
the option, using market option data (forward looking)
It is the wrong number that, when plugged into the
wrong model, will give the right price
It is different for different strikes K & maturities T
Volatility smile is (K) for each maturity T
Equivalent to this, the smile is (K/S)
272
2011 Emese Lazar, ICMA Centre
Implied volatility
15%
20%
25%
30%
0.6 0.8 1 1.2 1.4
Moneyness (K/S)
Volatility smile
273
2011 Emese Lazar, ICMA Centre
Volatility surface
Representing the volatility smiles (K/S) for each
maturity
volatility surface (K, T)
Properties:
It is steeper for short-term options
It changes in time
Volatility term structure: how the volatility smile
changes for different maturities
274
2011 Emese Lazar, ICMA Centre
Volatility surface
0
.
8
0
.
8
6
0
.
9
2
0
.
9
8
1
.
0
4
1
.
1
1
.
1
6
1
.
2
2
1
.
2
8
m3
m6
m9
m12
m15
m18
0%
5%
10%
15%
20%
25% Volatility
Moneyness
Maturity
275
2011 Emese Lazar, ICMA Centre
2. Risk factors & mappings
Risk factors:
Underlying
Volatility of underlying
Sensitivities: Greeks
Mapping the portfolio to the risk factors:
Define factor model (e.g. delta-gamma approximation)
Calculate sensitivities (e.g. using finite differences)
276
2011 Emese Lazar, ICMA Centre
Factor model
We know the evolution in time for the risk factors
(underlying price, volatility etc.)
We dont know the time evolution of the derivatives
prices (because pricing is rather complex), and we
dont know the evolution for the portfolios value
Thus, we have to approximate changes in portfolios
value (profit & losses)
This is done using Taylors expansion
277
2011 Emese Lazar, ICMA Centre
Taylors expansion
The delta-gamma-vega-theta approximation to the
change in the portfolio is (using position greeks):
P
P
* S +
P
* [S ]
2
+
P
* +
P
t
The delta-gamma approximation to the change in the
portfolio is:
P
P
* S +
P
* [S ]
2
278
2011 Emese Lazar, ICMA Centre
Taylors expansion using returns
The delta-gamma-vega-theta approximation to the
change in the portfolio is (using dollar greeks):
P
$
* r +
$
* r
2
+
P
* +
P
t
The delta-gamma approximation to the change in the
portfolio is:
P
$
* r +
$
* r
2
279
2011 Emese Lazar, ICMA Centre
Taylors expansion
The change in the value of the portfolio can be
approximated using the greeks, so it speeds up Monte
Carlo simulations
If a portfolio is delta-gamma hedged, then, for small
changes in the underlying, the portfolio value will not
change (assuming that the other factors like volatility, interest
rate etc. remain unchanged)
Similar reasoning for other types of hedging
It is better to use Taylors formula based on returns
280
2011 Emese Lazar, ICMA Centre
3. Application for VaR
Never use linear VaR because option portfolios are
highly non-normal
Use historical or Monte Carlo VaR
281
2011 Emese Lazar, ICMA Centre
Application for VaR
To compute P&L at every time point we need an
estimate of the new value of the portfolio.
Re-pricing the portfolio (especially if it has exotic
options) is rather complex & time-consuming
Taylors formula for delta-gamma-vega-theta hedging
(or some other type of hedging) is used to approximate
the change in the portfolios value
This will speed up the P&L calculation and thus the
VaR estimation
282
2011 Emese Lazar, ICMA Centre
Application for VaR
Different approach in the following situations:
Option type:
1. Vanilla options
2. Exotic options
Hedging:
A. No hedging
B. Type of hedging: (e.g. delta, delta-gamma, etc.)
Static:
Dynamic
283
2011 Emese Lazar, ICMA Centre
Vanilla options
Specifications:
Liquid
Simple pricing methods
Black-Scholes greeks are good approximations
284
2011 Emese Lazar, ICMA Centre
Exotic options
Specifications:
Illiquid
Complex pricing models (+ control variate technique)
Prices depend on demand & supply
The risks associated with these can be separated into:
a part that can be hedged with vanilla options and
a part that cannot be hedged using vanilla options
Greeks can be approximated
285
2011 Emese Lazar, ICMA Centre
3.1. Historical VaR
2 types:
Full revaluation
Based on sensitivities to risk factors
Historical VaR calculation based on full revaluation
takes a long time and as such it is rarely used in
practice; different numerical methods can be used to
approximate changes in the option prices
286
2011 Emese Lazar, ICMA Centre
Historical VaR based on revaluation
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Given are the time series for the risk factors, from these we get
the returns and changes in volatility
3. At every step compute new values for the risk factors, and thus
compute the new value of the portfolio (using option pricing
models); this will give the change in the portfolios value P
4. Do step 3 for all time steps historical histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
287
2011 Emese Lazar, ICMA Centre
Historical VaR based on sensitivities
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Given are the time series for the risk factors (prices), from these
we get the returns and changes in volatility
3. At every step approximate the change in the portfolios value P
using Taylors expansion (if the portfolio is hedged, then
accommodate for this in the formula)
4. Do step 3 for all time steps historical histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
288
2011 Emese Lazar, ICMA Centre
Changes in index implied volatility
How to map
I
, K
to RF? depends on strike & maturity
get rid of strike & maturity dependence
We assume that
I
, K
=
I
, ATM
,only model ATM vol (vega bucketing)
If there is a long history for
I
, ATM
for all maturities then we use this
for historical simulation
If the data on
I
, ATM
is sparse, then it is not practical to model the
evolution in time for the index implied volatilities for all maturities
(N) and N risk factors
In practice we model (changes in) only one implied volatility, for
example only the 3-month ATM volatility
It is possible to fit a regression to the volatility term structure:

I
, ATM
=
I

*
I
3m, ATM
+
I

289
2011 Emese Lazar, ICMA Centre
Changes in stock implied volatility
How to map
S
, K
to RF? depends on strike & maturity of option
get rid of strike, maturity & stock dep.
We assume that
S
, K
=
S
, ATM
,only model ATM vol
If there is a long history for
S
, ATM
for all maturities then we use
this for historical simulation
If the data on
S
, ATM
is sparse, but
S
3m, ATM
is available then run
regressions to estimate sensitivities to changes in the 3m vol, so we are
modelling only one implied volatility

S
, ATM
=
S

*
I
3m, ATM
+
S

If the data on
S
3m, ATM
is sparse, then estimate sensitivities to the
index implied vol

S
3m, ATM
=
S
*
I
3m, ATM
+
S
In this case we would have
S
, ATM

S
*
S

*
I
3m, ATM
290
2011 Emese Lazar, ICMA Centre
Other maturities
For options with non-standard maturities we do linear
interpolation
For example, if the option has maturity with
1
< <
2
and the
time series for
S
1, ATM
and
S
2, ATM
are known, then at every
time point:

S
, ATM
= (
2
) / (
2

1
)
S
1, ATM
+
+ (
1
) / (
2

1
)
S
2, ATM
291
2011 Emese Lazar, ICMA Centre
3.2. Monte Carlo VaR
2 types:
Full revaluation
Based on sensitivities to risk factors
Monte Carlo VaR calculation based on full
revaluation takes an extremely long time and as such
it is almost never used in practice; different
numerical methods can be used to approximate
changes in the option prices and thus shorten the
time required
292
2011 Emese Lazar, ICMA Centre
Monte Carlo VaR based on revaluation
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Simulate (correlated?) asset returns and changes in implied volatility
3. Compute new values for the risk factors, and thus compute
the new value of the portfolio (using option pricing models);
this will give the change in the portfolios value P
4. Do steps 2 & 3 for thousands of times simulated
histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
293
2011 Emese Lazar, ICMA Centre
Monte Carlo VaR based on sensitivities
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Simulate (correlated?) asset returns and changes in implied volatility
3. At every step approximate the change in the portfolios value
P using Taylors expansion (if hedging, then accommodate for
this in the formula)
4. Do steps 2 & 3 for thousands of times simulated histogram
for P&L = P
5. The VaR will be the lower % percentile of this histogram
294
2011 Emese Lazar, ICMA Centre
Simulating stock prices
For accuracy we simulate the log-price process:
The discrete-time version is:
true even for large t
This can be simulated by drawing a new at each step
( )
( )
2
2 d ln S t / dt dW = +
( ) ( )
( )
2
2 ln S t t ln S t / t t + = +
( ) ( )
( )
2
2 / t t
S t t S t e
+
+ =
295
2011 Emese Lazar, ICMA Centre
Simulating correlated variables
Some models require the simulation of correlated
random variables; this can be obtained by simulating
independent variables first and then transforming these
into correlated ones e.g., for 2 variables we have:
The Cholesky matrix of a covariance matrix multiplied
by the independent series will give correlated series
1 1
2
2 1 2

1
x
x x
=
= +
296
2011 Emese Lazar, ICMA Centre
Practical considerations
Use more realistic models for simulations
Stochastic (time-varying) volatility, with negative
correlation between returns and volatility
and/or
Accommodate for the possibility of jumps in the prices
Variance reduction techniques:
Antithetic variables
Importance sampling
Stratified sampling
Moment matching
Quasi-random sequences
297
2011 Emese Lazar, ICMA Centre
4.1 Basket options
These are options written on several underlying assets
Different deltas & gammas (relative to changes in each
underlying asset ) PLUS cross-gammas (relative to changes
in pairs of underlying assets)

ij
=
i
/ S
j
=
2
f / S
i
S
j
Different vegas
Taylors expansion becomes:
P
i
P
* S
i
+
ij
P
* [S
i
] [S
j
] +
i
P
*
i
+
P
t
The same treatments applies to portfolios of options written
on different underlying assets (except for cross-gammas)
298
Basket options
Dollar greeks
Dollar delta
i
$
=
i
P
* S
i
Dollar gamma
ij
$
=
ij
P
* S
i
* S
j
Taylors expansion:
P
i
$
* r
i
+
ij
$
* r
i
* r
j
+
i
P
*
i
+
P
t
2011 Emese Lazar, ICMA Centre
299
4.2 Index options
When some of the options are written on an index, then define
the greeks as total greeks, taking point values into account
For example, options on the FTSE 100 have a point value of pv
= 10, which means that every option is written on 10 * the
index, so when the index increases with x, then S = (10 * x)
When computing dollar delta, then the dollar delta has to be
multiplied by the point values to obtain the total dollar delta and
the dollar gammas by the point values of both underlying assets
Dollar delta
i
$
=
P
* S
i
* pv
i
Dollar gamma
ij
$
=
P
* S
i
* S
j
* pv
i
* pv
j
2011 Emese Lazar, ICMA Centre
300
2011 Emese Lazar, ICMA Centre
4.3 Futures and forwards
They can be written on any underlying (equities, commodities,
exchange rates, etc.)
Risk factors: (1) underlying asset prices &
(2) spot interest rates
Instead (because certain events affect the spot market but not the
futures market) it is better to map futures positions to
(1) constant maturity futures prices
(these price series have to be constructed artificially, using linear interpolation) &
(2) forward interest rates
This mapping can be done similar to CF mapping (e.g. variance
invariant mapping); also PCA can be used due to high correlations
among futures prices
301
2011 Emese Lazar, ICMA Centre
Interest rate derivatives
Part II. Interest rate derivatives
302
2011 Emese Lazar, ICMA Centre
5. Traditional risk measures
Risk factors: interest rates & their volatilities
The greeks measure the sensitivities of the option price to the
interest rates (+ position greeks)
Delta measures the sensitivity of the option price to changes in the
interest rate separate for each maturity
Instead of delta most often DVO1 =
P
/ 10,000 is used!
Gamma measures the sensitivity of the deltas to changes in the
interest rates separate for each (i,j) maturities; used less often
Vega measures the sensitivity of the option price to changes in the
volatility of interest rates separate for each maturity
Hedging (making the sensitivity to a RF equal with zero) is
also important
303
2011 Emese Lazar, ICMA Centre
6. Risk factors & mapping
Risk factors:
Interest rates (yield curve)
Volatility of interest rates
Sensitivities: Greeks
Mapping the portfolio to the risk factors:
Define factor model (e.g. delta-gamma approximation)
Calculate sensitivities (e.g. using finite differences)
304
2011 Emese Lazar, ICMA Centre
Factor model Taylors expansion
The delta-gamma-vega-theta approximation to the
change in the portfolio is (using portfolio greeks):
P
i
P
* r
i
+
ij
P
* [r
i
] [r
j
] +
i
P
*
i
+
P
t
The DV01-vega approximation to the change in the
portfolio is:
P 10,000 DV01
i
* r
i
+ +
i
P
*
i
305
2011 Emese Lazar, ICMA Centre
7.1 Historical VaR
2 types:
Full revaluation
Based on sensitivities to risk factors
Historical VaR calculation based on full revaluation
takes a long time and as such it is rarely used in
practice; different numerical methods can be used to
approximate changes in the option prices
306
2011 Emese Lazar, ICMA Centre
Historical VaR based on revaluation
Steps:
1. Start with a hedged/unhedged portfolio of IR derivatives
2. Given are the time series for the risk factors, from these we get
changes in the yield curve (and changes in implied volatility)
3. At every step compute new values for the risk factors, and thus
compute the new value of the portfolio (using IR derivative
pricing models); this will give the change in the portfolios value
P
4. Do step 3 for all time steps historical histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
307
2011 Emese Lazar, ICMA Centre
Historical VaR based on sensitivities
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Given are the time series for the risk factors, from these we get
changes in the yield curve (and changes in implied volatility)
3. At every step approximate the change in the portfolios value P
using Taylors expansion (if the portfolio is hedged, then
accommodate for this in the formula)
4. Do step 3 for all time steps historical histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
308
2011 Emese Lazar, ICMA Centre
Implementation
Often only parallel shifts in the yield curve are modelled r
i
=r
We assume that
i, , K
=
i,
, ignore the strike price for
implied volatilities
If there is a long history for
i,
for all maturities then we use
this for historical simulation
If the data on
i,
is sparse, then we run a regression

i,
=

*
i
+

This means that we model (changes in) only one implied

volatility, for example only the 3-month volatility
309
2011 Emese Lazar, ICMA Centre
7.2 Monte Carlo VaR
2 types:
Full revaluation
Based on sensitivities to risk factors
Monte Carlo VaR calculation based on full
revaluation takes an extremely long time and as such
it is almost never used in practice; different
numerical methods can be used to approximate
changes in the option prices and thus shorten the
time required
310
2011 Emese Lazar, ICMA Centre
Monte Carlo VaR based on revaluation
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Simulate changes in the yield curve (and changes in implied volatility)
3. Compute new values for the risk factors, and thus compute
the new value of the portfolio (using IR derivative pricing
models); this will give the change in the portfolios value P
4. Do steps 2 & 3 for thousands of times simulated
histogram for P&L = P
5. The VaR will be the lower % percentile of this histogram
311
2011 Emese Lazar, ICMA Centre
Monte Carlo VaR based on sensitivities
Steps:
1. Start with a hedged/unhedged portfolio of options
2. Simulate changes in the yield curve (and changes in implied volatility)
3. At every step approximate the change in the portfolios value
P using Taylors expansion (if hedging, then accommodate for
this in the formula)
4. Do steps 2 & 3 for thousands of times simulated histogram
for P&L = P
5. The VaR will be the lower % percentile of this histogram
312
2011 Emese Lazar, ICMA Centre
8.1 Swaps
FRA: exchange of two cash flows (single-period)
Swap: exchange of two cash flow streams (multi-period)
At time t = 0: PV of both cash flows are same (for fixed coupons
and floating coupons)
e.g.: 1mil notional, 2y maturity
leg1 = 6m Libor rate(L1, L2, L3), leg2 = 5% swap rate (f)
t=0 t = 6m t = 12m t = 18m t = 24m
L1
f
L2
f
L3
f
313
Swaps
Their treatment is very similar to CF mapping; they can be
decomposed into a series of CFs
Risk factors: Libor interest rates
Sensitivities: DV01
Best mapping method: interest rate sensitive mapping
Historical, Normal linear or Monte Carlo VaR can be
computed
2011 Emese Lazar, ICMA Centre
314
2011 Emese Lazar, ICMA Centre
8.2 Interest rate options
Caplet = call option on Libor
payoff = max (Libor fixed, 0)
Floorlet = put option on Libor
payoff = max (fixed Libor,0)
Cap = sum of caplets (multi-period)
Floor = sum of floorlets (multi-period)
315
2011 Emese Lazar, ICMA Centre
Swaptions
They are options on swaps
payoff = max (swap rate swap strike rate, 0)
Thus, they are defined by two maturities: one for the option
and one for the swap
Example: a 3 month option on a 2 year swap; this gives you
the right to enter into a 2 year swap after 3 months, with a
given swap strike rate
If after 3 months, the 2-year swap rate is higher than the swap
strike rate, then the option holder will exercise the option and
will enter into a swap with the swap strike rate (better than
market rate).
If after 3 months, the 2-year swap rate is lower than the swap
strike rate, then the option holder will not exercise the option
and will enter into a swap at market swap rate.
316
2011 Emese Lazar, ICMA Centre
Swaptions
Risk factors: (1) swaps yield curve
(2) swaption volatilities
Swaps yield curve:
Contains swap rates for different maturities
Similar to the treasury yield curve
Swaption volatility matrix:
Contains implied volatilities for different pairs of (option
maturity, underlying maturity)
317
Swaptions
Sensitivities: greeks
Delta:
i
measures the sensitivity of swaption price to
changes in returns r
i
DV01 can be used instead of delta
Gamma:
ij
measures the sensitivity of
i
to changes in r
j
Vega:
ij
measures the sensitivity of swaption price to
changes in
ij
Theta: measures the sensitivity of swaption price to
changes in time
It is best to compute historical VaR based on sensitivities (or
with full revaluation, but using some numerical approximations)
318
2011 Emese Lazar, ICMA Centre
Swaptions Taylors expansion
The delta-gamma-vega-theta approximation:
P
i
P
* r
i
+
ij
P
* [r
i
] [r
j
] +
ij
P
*
ij
+
P
t
The DV01-vega approximation to the change in the
portfolio is:
P 10,000 DV01
i
* r
i
+ +
ij
P
*
ij
319
Swaptions changes in volatilities
Since the swaption volatility matrix contains many elements
(they are all risk factors), it makes sense to reduce the number
of risk factors
In practice we model (changes in) only one swaption
implied volatility, for maturity (m*,n*) (the most liquid one)
How to map
i, j
to rf? depends on two maturities
get rid of maturity dependence
It is possible to fit a regression to the volatility term structure:

i, j
=
i, j
*
m*, n*
+
Thus we have only one volatility risk factor,
m*, n*
320
2011 Emese Lazar, ICMA Centre
Market Risk
Lecture 8
Fund management, banking & non-financial firms
321
2011 Emese Lazar, ICMA Centre
Contents
1. Funds
1.1 Risk assessment
1.2 Risk control
2. Banks
2.1 Risk assessment
2.2 Risk control
3. Non-financial firms
3.1 Risk assessment
3.2 Risk control
322
2011 Emese Lazar, ICMA Centre
1. Funds
Activity: to generate returns for investors by taking market risk
Income: % of asset value managed + % of profits
Types: - pension funds
- hedge funds
- specialist funds (UK equity, index tracker)
Risk factors: primary risks: market prices & volatilities
secondary risks: liquidity, interest rates
affect mainly assets but rarely liabilities as well!
Risk taker & beneficiary: investors
Regulation: directed at protecting investors
323
2011 Emese Lazar, ICMA Centre
1.1. Risk assessment
1. Ex-ante: standard models (more difficult)
2. Ex-post: - statistical analysis
- relative to a benchmark (traded or not traded asset)
- active return (return benchmark)
(1 + r
A
) * (1 + r
B
) = (1 + r)
r
B
= w
Bi
r
i
r = w
i
r
i
r
A
= (w
i
w
Bi
) r
i
- tracking error: (measures the risk of the relative returns)
for this the benchmark is irrelevant!
- relative (active) risk: (measures the relative risk of the returns)
doesnt distinguish between over- and under-performance
Concentrate on the medium term (monthly)!
2
A
STE r =
( )
2
A
TE r r =
324
2011 Emese Lazar, ICMA Centre
Example of TE & STE
(1 + c) * (1 + r
Index
) = (1 + r
Fund
) r
A
= c
TE = 0
STE = T * c
2
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
Index
Fund
60
70
80
90
100
110
120
Index
Fund
325
2011 Emese Lazar, ICMA Centre
Downside risk metrics
Lower partial moments: (regret, semi-st. dev., semi-skewness & semi-kurtosis)
only returns less than a benchmark (often = ) are considered
LPM
,
= E [ |min(0, X )|

]
1/
= 1, 2, 3, 4
Percentile risk metrics: probability of under-performing a benchmark
PBU
x
= P (X < x)
Benchmark VaR: Benchmark VaR

= P * x * B
where P = value of fund; B = zero-coupon bond value
x = such that PBU
x
=
Other measures: - Sharpe ratio doesnt take account of high moments
- Information ratio: excess return / tracking error
326
2011 Emese Lazar, ICMA Centre
Example
Daily profit and losses over the past 4 years (1000 obs.) in increasing
order: 95, 90, 80, 75, 70, 65, 64, 62, 61, 61, 59,
PBU
60
= P (X < 60) = 10 / 1000 = 1%
LPM
1, 60
= E [ |min(0, X (60))| ]
= E [35, 30, 20, 15, 10, 5, 4, 2, 1, 1, 0,]
= 12.3 * 1% + 0 * 99% = 0.123
LPM
2, 60
= (E [ |min(0, X (60))|
2
])
1/2
= (E [(35)
2
, (30)
2
, (20)
2
, (15)
2
, (10)
2
, (5)
2
, (4)
2
, (2)
2
, (1)
2
,(1)
2
, 0,])
1/2
= (289.7 * 1% + 0 * 99%)
1/2
= 2.897

= 1.7 etc.
327
2011 Emese Lazar, ICMA Centre
1.2. Risk control
Selective hedging: against a specific source of risk
- Disadvantage: protects only in several situations (e.g. small movements)
Momentary hedging: against risks that are feared for a short time only
Managing for a RAPM target: in line with investors objectives
(e.g. Sharpe ratio, Information ratio)
Capital protection: they guarantee a minimum return (built using derivatives)
- Disadvantage: fixed size & maturity
Solution (?): Constant proportional portfolio insurance
Example: 90% in bonds + 20% in index future (borrowing 10%)
Compliance and accountability: they have to comply with regulations
and act according to stated objectives
328
2011 Emese Lazar, ICMA Centre
2. Banks
Activity: 1. to collect deposits and offer loan (banking book)
2. investments (trading book)
3. to generate profit by taking market risk
illiquid assets and liabilities with indeterminate maturities
Income: difference between the rates on loans and deposits
Risk factors: primary risks: market prices, interest rates, FX etc.
secondary risks: volatility, underlying-futures spread, dividend
affect both assets and liabilities!
more difficult to be identified in the banking book
Risk taker & beneficiary: mostly banks
Regulation: directed at protecting clients
329
2011 Emese Lazar, ICMA Centre
2.1. Risk assessment
Depends on risk factors chosen and their dynamics; in which
banks enjoy a high level of freedom.
Models:
1. Statistical models: SV, GARCH future value of market factors
2. Pricing models: BS pricing and sensitivities of instruments
3. Risk aggregation models: VaR
Concentrate on the short term (daily or 10 days)!
330
2011 Emese Lazar, ICMA Centre
2.2. Risk control
Banks have freedom to:
Define deposit & loan interest rates
Adjust liabilities to match risk of assets
Use derivatives for hedging (50% chance the hedging will make money)
Hedging: of primary & secondary risks mostly using derivatives
rebalancing is important
Managing for a RAPM target: (e.g. Sharpe ratio, Information ratio)
Compliance:
Traders limits traditionally set in terms of exposures (greeks);
new regulation is towards setting these in terms of VaR
331
2011 Emese Lazar, ICMA Centre
3. Non-financial firms
Activity: 1. Manufacturing / Trading / Services
2. To generate profit for shareholders by taking risk
Income: differs
Risk factors: primary risks: interest rates, FX, commodity etc.
secondary risks: volatility, underlying-futures spread, dividend
can affect both assets and liabilities!
difficult to identify them
problem: accounting standards can be misleading
Risk taker & beneficiary: mostly the firms / shareholders
(pass the risks over to financial firms?)
Regulation
332
2011 Emese Lazar, ICMA Centre
3.1. Risk assessment
Horizon of interest: - long-term
- short-term
Risks: - transaction FX risk: affects the firm within the horizon of interest
- translation FX risk: affects the firm beyond the horizon of interest
Models: not VaR
Decision analysis methods: max V
firm
= f (factors)
to maximize the value of the firm, find:- significant factors
- sensitivities to the factors
- probability distributions of factors
- risk attitude of stockholders
- possible decisions
- optimal decision
333
2011 Emese Lazar, ICMA Centre
3.2. Risk control
Hedging: Alternative solutions are preferable over financial hedges
because financial hedges are standardized, can be illiquid and need
dynamic cash-flow
e.g. a foreign asset should be funded in the currency of that unit
financial hedges preferred for short-term transaction risks
Disadvantages: - expensive
- derivatives are for short term, risks for long term in general
- hedging strategies lose money in 50% of the cases
- mismatch between exposure and risk hedged
Managing to maximize risk-adjusted firm value
Compliance
334
2011 Emese Lazar, ICMA Centre
Summary
Market risk measurement & management is a growing field
Risk managing unit separate from risk taking unit
Look beyond VaR
Risk identification:
Finding hidden market risks is very important
Accounting standards are be misleading; use market data
Risk assessment:
Depends on objectives
Depends on assumptions
Model: some VaR model
RAPM
Risk control/mitigation:
Constraints: regulations, risk levels
Depends on objectives