Capital requirements

Statistical estimation of risk measures

Risk allocation

Financial Risk Management

Lecture 2. Market Risk

Thierry Roncalli?

?
University of Paris-Saclay

September, 2020

Thierry Roncalli Financial Risk Management (Lecture 2) 1 / 117

 Capital requirements
Statistical estimation of risk measures
Risk allocation

Agenda

Lecture 1: Introduction to Financial Risk Management

Lecture 2: Market Risk
Lecture 3: Credit Risk
Lecture 4: Counterparty Credit Risk and Collateral Risk
Lecture 5: Operational Risk
Lecture 6: Liquidity Risk
Lecture 7: Asset Liability Management Risk
Lecture 8: Model Risk
Lecture 9: Copulas and Extreme Value Theory
Lecture 10: Monte Carlo Simulation Methods
Lecture 11: Stress Testing and Scenario Analysis
Lecture 12: Credit Scoring Models

Thierry Roncalli Financial Risk Management (Lecture 2) 2 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Most important dates

19 October 1987: Stock markets crashed and the Dow Jones

Industrial Average index dropped by more than 20% in the day
1988: Publication of the Basel I Accord
1990s: Japanese asset price bubble
1994: Bond market massacre
October 1994: Publication of RiskMetrics by J.P. Morgan
January 1996: Amendment to incorporate market risks (Basel I)
2004: Measuring market risks is the same in Basel II
2008: Global Financial Crisis (GFC)
2009: Basel 2.5
January 2019: Revision of market risk in Basel III (also known as the
fundamental review of the trading book or FRTB)

Thierry Roncalli Financial Risk Management (Lecture 2) 3 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Definition

According to the Basel Committee, market risk is defined as “the risk of

losses (in on- and off-balance sheet positions) arising from movements in
market prices. The risks subject to market risk capital requirements
include but are not limited to:
default risk, interest rate risk, credit spread risk, equity risk, foreign
exchange (FX) risk and commodities risk for trading book
instruments;
FX risk and commodities risk for banking book instruments.”

Portfolio Fixed Income Equity Currency Commodity Credit

Trading X X X X X
Banking X X

⇒ trading book 6= banking book

Thierry Roncalli Financial Risk Management (Lecture 2) 4 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The Basel I/II framework

To compute the capital charge, banks have the choice between two
approaches:
1 the standardized measurement method (SMM)
2 the internal model-based approach (IMA)
⇒ Banks quickly realized that they can sharply reduce their capital
requirements by adopting internal models

Thierry Roncalli Financial Risk Management (Lecture 2) 5 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Standardized measurement method (SMM)

Five main risk categories:

1 Interest rate risk
2 Equity risk
3 Currency risk
4 Commodity risk
5 Price risk on options and derivatives

For each category, a capital charge is computed to cover:

the general market risk
the specific risk

Thierry Roncalli Financial Risk Management (Lecture 2) 6 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Standardized measurement method (SMM)

The capital charge K is equal to the risk exposure E times the capital
charge weight K :
K=E ·K

For the specific risk, the risk exposure corresponds to the notional of
the instrument, whether it is a long or a short position
For the general market risk, long and short positions on different
instruments can be offset

Thierry Roncalli Financial Risk Management (Lecture 2) 7 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The case of equity risk

The capital charge for specific risk is 4% if the portfolio is liquid and
well-diversified and 8% otherwise
For the general market risk, the risk weight is equal to 8% and applies
to the net exposure

Remark
Under Basel 2.5, the capital charge for specific risk is set to 8% whatever
the liquidity of the portfolio

Thierry Roncalli Financial Risk Management (Lecture 2) 8 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The case of equity risk

Example
We consider a $100 mn short exposure on the S&P 500 index futures
contract and a $60 mn long exposure on the Apple stock.

The capital charge for specific risk is1 :

KSpecific = 100 × 4% + 60 × 8% = 4 + 4.8 = 8.8

The net exposure is −$40 mn. We deduce that the capital charge for the
general market risk is:

KGeneral = |−40| × 8% = 3.2

It follows that the total capital charge for this equity portfolio is $12 mn.

1 We
assume that the S&P 500 index is liquid and well-diversified, whereas the
exposure on the Apple stock is not diversified.
Thierry Roncalli Financial Risk Management (Lecture 2) 9 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The case of interest rate risk (specific risk)

For government instruments, the capital charge weights are:

AAA A+ BB+
Below
Rating to to to NR
B−
AA− BBB− B−
Maturity 0−6M 6M−2Y 2Y+
K 0% 0.25% 1.00% 1.60% 8% 12% 8%
In the case of other instruments (PSE, banks and corporates), the
capital charge weights are:
AAA BB+
Below
Rating to to NR
BB−
BBB− BB−
Maturity 0−6M 6M−2Y 2Y+
K 0.25% 1.00% 1.60% 8% 12% 8%

Thierry Roncalli Financial Risk Management (Lecture 2) 10 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The case of interest rate risk (specific risk)

Example
We consider a trading portfolio with the following exposures: a long
position of $50 mn on Euro-Bund futures, a short position of $100 mn on
three-month T-Bills and a long position of $10 mn on an investment grade
(IG) corporate bond with a three-year residual maturity.

⇒ Why the capital charge for specific risk is equal to $0, $0 and $160 000?

Thierry Roncalli Financial Risk Management (Lecture 2) 11 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The case of interest rate risk (general market risk)

Two methods:
Maturity approach
Duration approach (price sensitivity with respect to a change in yield)

Thierry Roncalli Financial Risk Management (Lecture 2) 12 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Internal model-based approach

The use of an internal model is conditional upon the approval of the
supervisory authority:
Qualitative criteria
Independent risk control unit
Daily reports
Daily risk management
Etc.
Quantitative criteria
The value-at-risk (VaR) is computed on a daily basis with a 99%
confidence level. The minimum holding period of the VaR is 10
trading days. If the bank computes a VaR with a shorter holding
period, it can use the square-root-of-time rule
Relevant risk factors
Sample period: at least one year
The value of the multiplication factor depends on the quality of the
internal model with a range between 3 and 4. The quality of the
internal model is related to its ex-post performance measured by the
backtesting procedure
Stress testing & Backtesting
Thierry Roncalli Financial Risk Management (Lecture 2) 13 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The square-root-of-time rule

The holding period to define the capital is 10 trading days. For that,
banks can compute the one-day VaR and converts it to a ten-day VaR:
√
VaRα (w ; ten days) = 10 × VaRα (w ; one day)

Thierry Roncalli Financial Risk Management (Lecture 2) 14 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Required capital

The required capital at time t is equal to:

60
!
1 X
Kt = max VaRt−1 , (3 + ξ) · VaRt−i
60
i=1

where VaRt is the 10-day value-at-risk calculated at time t and ξ is the

penalty coefficient (0 ≤ ξ ≤ 1)

Thierry Roncalli Financial Risk Management (Lecture 2) 15 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Required capital

Figure: Calculation of the required capital with the VaR

Thierry Roncalli Financial Risk Management (Lecture 2) 16 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Backtesting
Definition
Backtesting consists of verifying that the internal model is consistent with
a 99% confidence level
⇒ For instance, we expect that the realized loss exceeds the VaR figure
once every 100 observations on average

Table: Value of the penalty coefficient ξ for a sample of 250 observations

Number of
Zone ξ
exceptions
Green 0–4 0.00
5 0.40
6 0.50
Yellow 7 0.65
8 0.75
9 0.85
Red 10+ 1.00

Thierry Roncalli Financial Risk Management (Lecture 2) 17 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Statistical approach of backtesting

We note w the portfolio, VaRα (w ; h) the value-at-risk calculated at time
t − 1, and Lt (w ) the daily loss at time t:

Lt (w ) = −Πt (w ) = MtMt−1 − MtMt

By definition, we have:

Pr {Lt (w ) ≥ VaRα (w ; h)} = 1 − α

Let et be the random variable which is equal to 1 if there is an exception

and 0 otherwise. et is a Bernoulli random variable with parameter p:

p = Pr {et = 1} = Pr {Lt (w ) ≥ VaRα (w ; h)} = 1 − α

Pt2
Let Ne (t1 ; t2 ) = t=t1 et be the number of exceptions for the period
[t1 , t2 ]. We assume that the exceptions are independent across time.
Main result
Ne (t1 ; t2 ) is a binomial random variable B (n; 1 − α)
Thierry Roncalli Financial Risk Management (Lecture 2) 18 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Statistical approach of backtesting

Table: Probability distribution (in %) of the number of exceptions (n = 250

trading days)

α = 99% α = 98%
m Pr {Ne = m} Pr {Ne ≤ m} Pr {Ne = m} Pr {Ne ≤ m}
0 8.106 8.106 0.640 0.640
1 20.469 28.575 3.268 3.908
2 25.742 54.317 8.303 12.211
3 21.495 75.812 14.008 26.219
4 13.407 89.219 17.653 43.872
5 6.663 95.882 17.725 61.597
6 2.748 98.630 14.771 76.367
7 0.968 99.597 10.507 86.875
8 0.297 99.894 6.514 93.388
9 0.081 99.975 3.574 96.963
10 0.020 99.995 1.758 98.720

Thierry Roncalli Financial Risk Management (Lecture 2) 19 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Statistical approach of backtesting

Figure: Color zones of the backtesting procedure (α = 99%)

Thierry Roncalli Financial Risk Management (Lecture 2) 20 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The Basel 2.5 framework

The required capital becomes:

Kt = KVaR
t + KSVaR
t + KSRC
t + KIRC
t + KCRM
t

where KVaR
t is the VaR capital and KSRC
t (Basel II), and:
KSVaR
t is the Stressed VaR
KIRC
t is the incremental risk charge (IRC), which measures the
impact of rating migrations and defaults
KCRM
t is the comprehensive risk measure (CRM), which
corresponds to a supplementary capital charge for credit exotic
trading portfolios

Thierry Roncalli Financial Risk Management (Lecture 2) 21 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The stressed VaR

Definition
The stressed VaR has the same characteristics than the traditional VaR
(99% confidence level and 10-day holing period), but the model inputs are
“calibrated to historical data from a continuous 12-month period of
significant financial stress relevant to the bank’s portfolio”.

⇒ This implies that the historical period to compute the SVaR is

completely different than the historical period to compute the VaR2

2 For
instance, a typical period is the 2008 year which both combines the subprime
mortgage crisis and the Lehman Brothers bankruptcy
Thierry Roncalli Financial Risk Management (Lecture 2) 22 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

The Basel III framework

Banks have the choice between two approaches for computing the capital
charge:
1 a standardized method (SA-TB3 )
2 an internal model-based approach (IMA)
⇒ SMM is replaced by SA-TB and IMA is revisited
Remark
Contrary to the previous framework, the SA-TB method is very important
even if banks calculate the capital charge with the IMA method. Indeed,
the bank must implement SA-TB in order to meet the output (or capital)
floor requirement, which is set at 72.5% in January 2027:
IMA SA-TB


Kt = max Kt , 72.5% × Kt

3 TB means trading book

Thierry Roncalli Financial Risk Management (Lecture 2) 23 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

SA-TB

The standardized capital charge is the sum of three components:

1 sensitivity-based capital requirement
2 the default risk capital (DRC)
3 the residual risk add-on (RRAO)

Some comments:
The first component must be viewed as the pure market risk and is
the equivalent of the capital charge for the general market risk
The second component captures the jump-to-default risk (JTD) and
replaces the specific risk
The last component captures specific risks that are difficult to
measure in practice

Thierry Roncalli Financial Risk Management (Lecture 2) 24 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Sensitivity-based capital requirement

We have:
K = KDelta + KVega + KCurvature
⇒ a capital charge for delta, vega and curvature risks

7 risk classes:
1 General interest rate risk (GIRR)
2 Credit spread risk(CSR) on non-securitization products
3 Credit spread risk(CSR) on non-correlation trading portfolio
(non-CTP)
4 Credit spread risk(CSR) on correlation trading portfolio (CTP)
5 Equity risk
6 Commodity risk
7 Foreign exchange risk

Thierry Roncalli Financial Risk Management (Lecture 2) 25 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Delta and vega risk components

We first begin to calculate the weighted sensitivity of each risk factor
Fj :
WSj = Sj · RWj
where Sj and RWj are the net sensitivity of the portfolio with respect
to the risk factor and the risk weight of Fj
Second, we calculate the capital requirement for the risk bucket Bk :
v  
u
u X X
2
KBk = tmax  WSj + ρj,j 0 WSj WSj 0 , 0
u
j j 0 6=j

where Fj ∈ Bk .
Finally, we aggregate the different buckets for a given risk class:
sX X
Delta/Vega 2
K = KBk + γk,k 0 WSBk WSBk 0
k k 0 6=k
P
where WSBk = j∈Bk WSj is the weighted sensitivity of the bucket
Bk .
Thierry Roncalli Financial Risk Management (Lecture 2) 26 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Delta and vega risk components

The capital requirement for delta and vega risks can be viewed as a
Gaussian risk measure with the following parameters:
1 the sensitivities Sj of the risk factors that are calculated by the bank;
2 the risk weights RWj of the risk factors;
3 the correlation ρj,j 0 between risk factors within a bucket;
4 the correlation γk,k 0 between the risk buckets.

Thierry Roncalli Financial Risk Management (Lecture 2) 27 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Curvature risk component

The curvature risk uses a similar methodology, but it is based on two

adverse scenarios: (1) the risk factor is shocked upward and (2) the risk
factor is shocked downward

The curvature risk is close to the gamma risk that we encounter in the
theory of options

Thierry Roncalli Financial Risk Management (Lecture 2) 28 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Practical computation of dela, vega and curvature risks

Three steps:
1 defining the risk factors
2 calculating the sensitivities
3 calculating the risk-weighted sensitivities WSj

Thierry Roncalli Financial Risk Management (Lecture 2) 29 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Defining the risk factors

⇒ The Basel Committee gives a very precise list of risk factors by asset
classes

For instance, the equity delta risk factors are the equity spot prices and the
equity repo rates, the equity vega risk factors are the implied volatilities of
options, and the equity curvature risk factors are the equity spot prices

In the case of the interest rate risk class (GIRR), the risk factors include
the yield curve4 , a flat curve of market-implied inflation rates for each
currency and some cross-currency basis risks

4 The
risk factors correspond to the following tenors of the yield curve: 3M, 6M, 1Y,
2Y, 3Y, 5Y, 10Y, 15Y, 20Y and 30Y.
Thierry Roncalli Financial Risk Management (Lecture 2) 30 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Calculating the sensitivities

The equity delta sensitivity of the instrument i with respect to the equity
risk factor Fj is given by:

Si,j = ∆i (Fj ) · Fj

where ∆i (Fj ) measures the (discrete) delta of the instrument i by

shocking the equity risk factor Fj by 1%:

Pi (1.01 · Fj ) − Pi (Fj ) Pi (1.01 · Fj ) − Pi (Fj )

Si,j = · Fj =
1.01 · Fj − Fj 0.01

Remark
If the instrument corresponds to a stock, the sensitivity is exactly the
price of this stock when the risk factor is the stock price, and zero
otherwise
If the instrument corresponds to an European option on this stock, the
sensitivity is the traditional delta of the option times the stock price
Thierry Roncalli Financial Risk Management (Lecture 2) 31 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Calculating the sensitivities

For the vega sensitivity, we have:

Si,j = υ i (Fj ) · Fj

where Fj is the implied volatility and υ i (Fj ) is the vega of the instrument

Thierry Roncalli Financial Risk Management (Lecture 2) 32 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Calculating the risk-weighted sensitivities

We use the figures given in BCBS (2019) for the risk weight RWj , the
correlation ρj,j 0 and the correlation γk,k 0

Thierry Roncalli Financial Risk Management (Lecture 2) 33 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Internal model-based approach

A trading desk is “an unambiguously defined group of traders or trading

accounts that implements a well-defined business strategy operating within
a clear risk management structure”.

⇒ Internal models are implemented at the trading desk level, meaning

that some trading desks are approved for the use of internal models, while
other trading desks must use the SA-TB approach

Thierry Roncalli Financial Risk Management (Lecture 2) 34 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

Main differences with Basel I/II

The value-at-risk at the 99% confidence level is replaced by the expected
shortfall at the 97.5% confidence level. Moreover, the 10-day holding
period is not valid for all instruments

Expected shortfall
The expected shortfall is the average loss beyond the value-at-risk

Thierry Roncalli Financial Risk Management (Lecture 2) 35 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

Impact of the liquidity

v !2
u 5 r
uX hk − hk−1
ESα (w ) = t ESα (w ; hk )
h1
k=1

ESα (w ; h1 ) is the expected shortfall of the portfolio w at horizon 10

days by considering all risk factors
ESα (w ; hk ) is the expected shortfall of the portfolio w at horizon hk
days by considering the risk factors Fj that belongs to the liquidity
class k
hk is the horizon of the liquidity class k, which is given below:
Liquidity class k 1 2 3 4 5
Liquidity horizon hk 10 20 40 60 120

Thierry Roncalli Financial Risk Management (Lecture 2) 36 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

Liquidity buckets
1 Interest rates (specified currencies and domestic currency of the
bank), equity prices (large caps), FX rates (specified currency pairs).
2 Interest rates (unspecified currencies), equity prices (small caps) and
volatilities (large caps), FX rates (currency pairs), credit spreads (IG
sovereigns), commodity prices (energy, carbon emissions, precious
metals, non-ferrous metals).
3 FX rates (other types), FX volatilities, credit spreads (IG corporates
and HY sovereigns).
4 Interest rates (other types), IR volatility, equity prices (other types)
and volatilities (small caps), credit spreads (HY corporates),
commodity prices (other types) and volatilities (energy, carbon
emissions, precious metals, non-ferrous metals).
5 Credit spreads (other types) and credit spread volatilities, commodity
volatilities and prices (other types).

Thierry Roncalli Financial Risk Management (Lecture 2) 37 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

How to calculate the expected shortfall for a period of stress?

!
(full,current)
(reduced,stress) ESα (w ; h)
ESα (w ; h) = ESα (w ; h) · min (reduced,current)
,1
ESα (w ; h)

where ES(full,current)
α is the expected shortfall based on the current period
(reduced,current)
with the full set of risk factors, ESα is the expected shortfall
based on the current period with a restricted set of risk factors and
(reduced,stress)
ESα is the expected shortfall based on the stress period with
the restricted set of risk factors
Remark
The previous formula assumes that there is a proportionality factor
between the full set and the restricted set of risk factors:

ES(full,stress)
α (w ; h) ES(reduced,stress)
α (w ; h)
≈
ES(full,current)
α (w ; h) (reduced,current)
ESα (w ; h)
Thierry Roncalli Financial Risk Management (Lecture 2) 38 / 117
 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

Example
In the table below, we have calculated the 10-day expected shortfall for a
given portfolio:

Set of Liquidity class

Period
risk factors 1 2 3 4 5
Full Current 100 75 34 12 6
Reduced Current 88 63 30 7 5
Reduced Stress 112 83 47 9 7

Thierry Roncalli Financial Risk Management (Lecture 2) 39 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

Table: Scaled expected shortfall

Full Reduced Reduced Full/Stress Full

k Sck
Current Current Stress (not scaled) Stress
1 1 100.00 88.00 112.00 127.27 127.27
2 √1 75.00 63.00 83.00 98.81 98.81
3 √2 48.08 42.43 66.47 53.27 75.33
4 √2 16.97 9.90 12.73 15.43 21.82
5 6 14.70 12.25 17.15 8.40 20.58
Total 135.80 117.31 155.91 180.38
p
The scaling factor is equal to Sck = (hk − hk−1 ) /h1 , the scaled
expected shortfall is equal to ES?α (w ; hk ) = Scq
k · ESα (w ; hk ) and the
P5 ? 2
total expected shortfall is given by ESα (w ) = k=1 (ES α (w ; h k ))

Thierry Roncalli Financial Risk Management (Lecture 2) 40 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for modellable risk factors

The final step for computing the capital requirement (also known as the
‘internally modelled capital charge’) is to apply this formula:
5
X
IMCC = % · IMCCglobal + (1 − %) · IMCCk
k=1

where:
% is equal to 50%
IMCCglobal is the stressed ES calculated with the internal model and
cross-correlations between risk classes
IMCCk is the stressed ES calculated at the risk class level (interest
rate, equity, foreign exchange, commodity and credit spread)

Thierry Roncalli Financial Risk Management (Lecture 2) 41 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Other capital requirements

Concerning non-modellable risk factors, the capital requirement is

based on stress scenarios, that are equivalent to a stressed expected
shortfall SES
The default risk capital (DRC) is calculated using a value-at-risk
model with a 99.9% confidence level with the same default
probabilities that are used for the IRB approach

Thierry Roncalli Financial Risk Management (Lecture 2) 42 / 117

 Capital requirements The Basel I/II framework
Statistical estimation of risk measures The Basel 2.5 framework
Risk allocation The Basel III framework

Capital requirement for the market risk

For eligible trading desks, we have:
P60 P60 !
mc IMCCt−i + SESt−i
KIMA
t = max IMCCt−1 + SESt−1 , i=1 i=1
+DRC
60

where mc = 1.5 + ξ and 0 ≤ ξ ≤ 0.5

Table: Value of the penalty coefficient ξ in Basel III

Number of
Zone ξ
exceptions
Green 0–4 0.00
5 0.20
6 0.26
Amber 7 0.33
8 0.38
9 0.42
Red 10+ 0.50

Thierry Roncalli Financial Risk Management (Lecture 2) 43 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Coherent risk measures

We note R (w ) as the risk measure of portfolio w
Coherent risk measure
1 Subadditivity
R (w1 + w2 ) ≤ R (w1 ) + R (w2 )
2 Homogeneity
R (λw ) = λR (w ) if λ ≥ 0
3 Monotonicity

if w1 ≺ w2 , then R (w1 ) ≥ R (w2 )

4 Translation invariance

if m ∈ R, then R (w + m) = R (w ) − m

⇒ Translation invariance implies that:

R (w + R (w )) = R (w ) − R (w ) = 0
Thierry Roncalli Financial Risk Management (Lecture 2) 44 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Some risk measures

The portfolio’s loss is equal to L (w ) = −Pt (w ) Rt+h (w )

Volatility of the loss

R (w ) = σ (L (w )) = σ (w )

Standard deviation-based risk measure

R (w ) = SDc (w ) = E [L (w )] + c · σ (L (w )) = −µ (w ) + c · σ (w )

Value-at-risk

R (w ) = VaRα (w ) = inf {` : Pr {L (w ) ≤ `} ≥ α}

Expected shortfall
Z 1
1
R (w ) = ESα (w ) = E [L (w ) | L (w ) ≥ VaRα (w )] = VaRu (w ) du
1−α α

Thierry Roncalli Financial Risk Management (Lecture 2) 45 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The value-at-risk is not always subadditive

Example
We consider a $100 defaultable zero-coupon bond, whose default
probability is equal to 200 bps. We assume that the recovery rate R is a
binary random variable with Pr {R = 0.25} = Pr {R = 0.75} = 50%.

D =0 L=0

Pr = 98%

100

R = 75% L = 25
Pr = 2% Pr = 50%

D =1

Pr = 50%
R = 25% L = 75

⇒ F (0) = Pr {L ≤ 0} = 98%, F (25) = Pr {Li ≤ 25} = 99% and

F (75) = Pr {Li ≤ 75} = 100%
Thierry Roncalli Financial Risk Management (Lecture 2) 46 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The value-at-risk is not always subadditive

The 99% value-at-risk is equal to $25, and we have:
25 + 75
ES99% (L) = E [L | L ≥ 25] = = $50
2
We now consider two zero-coupon bonds with iid default times:
L1 = 0 L1 = 25 L1 = 75
L2 = 0 96.04% 0.98% 0.98% 98.00%
L2 = 25 0.98% 0.01% 0.01% 1.00%
L2 = 75 0.98% 0.01% 0.01% 1.00%
98.00% 1.00% 1.00%

We deduce that the probability distribution function of L = L1 + L2 is:

` 0 25 50 75 100 150
Pr {L = `} 96.04% 1.96% 0.01% 1.96% 0.02% 0.01%
Pr {L ≤ `} 96.04% 98% 98.01% 99.97% 99.99% 100%

It follows that VaR99% (L) = 75 and:

75 × 1.96% + 100 × 0.02% + 150 ∗ 0.01%
ES99% (L) = = $75.63
1.96% + 0.02% + 0.01%
Thierry Roncalli Financial Risk Management (Lecture 2) 47 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Value-at-risk

Definition
The value-at-risk VaRα (w ; h) is defined as the potential loss which the
portfolio w can suffer for a given confidence level α and a fixed holding
period h:

Pr {L (w ) ≤ VaRα (w ; h)} = α ⇔ VaRα (w ; h) = F−1

L (α)

Three parameters are necessary to compute this risk measure:

the holding period h, which indicates the time period to calculate the
loss;
the confidence level α, which gives the probability that the loss is
lower than the value-at-risk;
the portfolio w , which gives the allocation in terms of risky assets and
is related to the risk factors.

Thierry Roncalli Financial Risk Management (Lecture 2) 48 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Expected shortfall

Definition
The expected shortfall ESα (w ; h) is defined as the expected loss beyond
the value-at-risk of the portfolio:

ESα (w ; h) = E [L (w ) | L (w ) ≥ VaRα (w ; h)]

We notice that ESα (w ; h) ≥ VaRα (w ; h)

Thierry Roncalli Financial Risk Management (Lecture 2) 49 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Three methods
Let (F1 , . . . , Fm ) be the vector of risk factors. We assume that there is a
pricing function g such that:
Pt (w ) = g (F1,t , . . . , Fm,t ; w )
We deduce that the expression of the random loss is equal to:
L (w ) = Pt (w ) − g (F1,t+h , . . . , Fm,t+h ; w ) = ` (F1,t+h , . . . , Fm,t+h ; w )
where ` is the loss function. We have:
d α (w ; h) = F̂−1 (α) = −F̂−1 (1 − α)
VaR L Π

and: Z 1
c α (w ; h) = 1
ES F̂−1
L (u) du
1−α α

1 the historical (or empirical or non-parametric) VaR/ES

2 the analytical (or parametric or Gaussian) VaR/ES
3 the Monte Carlo (or simulated) VaR/ES
Thierry Roncalli Financial Risk Management (Lecture 2) 50 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Historical methods

Two approaches:
order statistic approach
kernel approach

Let (F1,s , . . . , Fm,s ) be the vector of risk factors observed at time s < t. If
we calculate the future P&L with this historical scenario, we obtain:

Πs (w ) = g (F1,s , . . . , Fm,s ; w ) − Pt (w )

If we consider nS historical scenarios (s = 1, . . . , nS ), the empirical

distribution F̂Π is described by the following probability distribution:

Π (w ) Π1 (w ) Π2 (w ) ··· ΠnS (w )
ps 1/nS 1/nS 1/nS

Thierry Roncalli Financial Risk Management (Lecture 2) 51 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Order statistic approach

Theorem (HFRM, page 67)

Let X1 , . . . , Xn be a sample from a continuous distribution F. Suppose
√ for a given scalar α ∈ ]0, 1[, there exists a sequence {an } such that
that
n (an − nα) → 0. We can show that:

√
 
α (1 − α)
n X(an :n) − F−1 (α) → N 0, 2 −1


f (F (α))

⇒ F̂−1 (α) = X(nα:n)

If ns = 1 000, F̂−1 (90%) is the 900th order statistic
If ns = 2 00, F̂−1 (90.5%) is the 181th order statistic

Thierry Roncalli Financial Risk Management (Lecture 2) 52 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Order statistic approach

Figure: Density of the quantile estimator (Gaussian case)

Thierry Roncalli Financial Risk Management (Lecture 2) 53 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

We calculate the order statistics associated to the P&L sample

{Π1 (w ) , . . . , ΠnS (w )}:

min Πs (w ) = Π(1:nS ) ≤ Π(2:nS ) ≤ · · · ≤ Π(nS −1:nS ) ≤ Π(nS :nS ) = max Πs (w )

s s

It follows that:
VaRα (w ; h) = −Π(nS (1−α):nS )

Thierry Roncalli Financial Risk Management (Lecture 2) 54 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

Remark
If nS (1 − α) is not an integer, we consider the interpolation scheme:



VaRα (w ; h) = − Π(q:nS ) + (nS (1 − α) − q) Π(q+1:nS ) − Π(q:nS )

where q = qα (nS ) = bnS (1 − α)c is the integer part of nS (1 − α).

In the case where we use 250 historical scenarios, the 99% value-at-risk is
the mean between the second and third largest losses:



VaR99% (w ; h) = − Π(2:250) + (2.5 − 2) Π(3:250) − Π(2:250)
1

= − Π(2:250) + Π(3:250)
2
1

= L(249:250) + L(248:250)
2

Thierry Roncalli Financial Risk Management (Lecture 2) 55 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

Example
We consider a portfolio composed of 10 stocks Apple and 20 stocks
Coca-Cola. The current date is 2 January 2015.

Remark
Data are available at
http: // www. thierry-roncalli. com/ download/ frm-data1. xlsx

Thierry Roncalli Financial Risk Management (Lecture 2) 56 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

The mark-to-market of the portfolio is:

Pt (w ) = 10 × P1,t + 20 × P2,t

where P1,t and P2,t are the stock prices of Apple and Coca-Cola. We
assume that the market risk factors corresponds to the daily stock returns
R1,t and R2,t . We deduce that the P&L for the scenario s is equal to:

Πs (w ) = 10 × P1,s + 20 × P2,s − Pt (w )
| {z }
g (R1,s ,R2,s ;w )

where Pi,s = Pi,t × (1 + Ri,s ) is the simulated price of stock i for the
scenario s.

Thierry Roncalli Financial Risk Management (Lecture 2) 57 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

Table: Computation of the market risk factors R1,s and R2,s

Apple Coca-Cola
s Date
Price R1,s Price R2,s
1 2015-01-02 109.33 −0.95% 42.14 −0.19%
2 2014-12-31 110.38 −1.90% 42.22 −1.26%
3 2014-12-30 112.52 −1.22% 42.76 −0.23%
4 2014-12-29 113.91 −0.07% 42.86 −0.23%
5 2014-12-26 113.99 1.77% 42.96 0.05%
6 2014-12-24 112.01 −0.47% 42.94 −0.07%
7 2014-12-23 112.54 −0.35% 42.97 1.46%
8 2014-12-22 112.94 1.04% 42.35 0.95%
9 2014-12-19 111.78 −0.77% 41.95 −1.04%
10 2014-12-18 112.65 2.96% 42.39 2.02%

Thierry Roncalli Financial Risk Management (Lecture 2) 58 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

We calculate the historical risk factors. For instance, we have:

109.33
R1,1 = − 1 = −0.95%
110.38
We calculate the simulated prices. For instance, in the case of the 9th
scenario, we obtain:

P1,s = 109.33 × (1 − 0.77%) = $108.49

P2,s = 42.14 × (1 − 1.04%) = $41.70

We then deduce the simulated mark-to-market

MtMs (w ) = g (R1,s , R2,s ; w )

Thierry Roncalli Financial Risk Management (Lecture 2) 59 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

Table: Computation of the simulated P&L Πs (w )

Apple Coca-Cola
s Date MtMs (w ) Πs (w )
R1,s P1,s R2,s P2,s
1 2015-01-02 −0.95% 108.29 −0.19% 42.06 1 924.10 −12.00
2 2014-12-31 −1.90% 107.25 −1.26% 41.61 1 904.66 −31.44
3 2014-12-30 −1.22% 108.00 −0.23% 42.04 1 920.79 −15.31
4 2014-12-29 −0.07% 109.25 −0.23% 42.04 1 933.37 −2.73
5 2014-12-26 1.77% 111.26 0.05% 42.16 1 955.82 19.72
23 2014-12-01 −3.25% 105.78 −0.62% 41.88 1 895.35 −40.75
69 2014-09-25 −3.81% 105.16 −1.16% 41.65 1 884.64 −51.46
85 2014-09-03 −4.22% 104.72 0.34% 42.28 1 892.79 −43.31
108 2014-07-31 −2.60% 106.49 −0.83% 41.79 1 900.68 −35.42
236 2014-01-28 −7.99% 100.59 0.36% 42.29 1 851.76 −84.34
242 2014-01-17 −2.45% 106.65 −1.08% 41.68 1 900.19 −35.91
250 2014-01-07 −0.72% 108.55 0.30% 42.27 1 930.79 −5.31

Thierry Roncalli Financial Risk Management (Lecture 2) 60 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the value-at-risk

If we rank the scenarios, the worst P&Ls are −84.34, −51.46, −43.31,
−40.75, −35.91 and −35.42. We deduce that the daily historical VaR is
equal to:
1
VaR99% (w ; one day) = (51.46 + 43.31) = $47.39
2
If we assume that mc = 3, the corresponding capital charge represents
23.22% of the portfolio value:
VaR
√
Kt = 3 × 10 × 47.39 = $449.54

Thierry Roncalli Financial Risk Management (Lecture 2) 61 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the expected shortfall

Since the expected shortfall is the expected loss beyond the value-at-risk,
it follows that the historical expected shortfall is given by:
S n
1 X
ESα (w ; h) = 1 {Ls ≥ VaRα (w ; h)} · Ls
qα (nS ) s=1

or:
S n
1 X
ESα (w ; h) = − 1 {Πs ≤ − VaRα (w ; h)} · Πs
qα (nS ) s=1

where qα (nS ) = bns (1 − α)c is the integer part of ns (1 − α).

Computation of the ES
We have:
qα (nS )
1 X
ESα (w ; h) = − Π(i:nS )
qα (nS )
i=1

Thierry Roncalli Financial Risk Management (Lecture 2) 62 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the expected shortfall

We have:
84.34 + 51.46
ES99% (w ; one day) = = $67.90
2
and:
84.34 + 51.46 + 43.31 + 40.75 + 35.91 + 35.42
ES97.5% (w ; one day) = = $48.53
6

We remind that VaR99% (w ; one day) = $47.39.

Thierry Roncalli Financial Risk Management (Lecture 2) 63 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Analytical methods

We speak about analytical value-at-risk when we are able to find a

closed-form formula of F−1
L (α)

Thierry Roncalli Financial Risk Management (Lecture 2) 64 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Gaussian value-at-risk

2
Suppose that L (w ) ∼ N µ (L) , σ (L) . In this case, we have
Pr L (w ) ≤ F−1

L (α) = α or:

F−1
   −1 
L (w ) − µ (L) (α) − µ (L) F (α) − µ (L)
Pr ≤ L =α⇔Φ L
=α
σ (L) σ (L) σ (L)
We deduce that:
F−1
L (α) − µ (L)
= Φ−1 (α) ⇔ F−1
L (α) = µ (L) + Φ −1
(α) σ (L)
σ (L)
The expression of the value-at-risk is then:
VaRα (w ; h) = µ (L) + Φ−1 (α) σ (L)

if α = 99%, we obtain:
VaR99% (w ; h) = µ (L) + 2.33 × σ (L)
Thierry Roncalli Financial Risk Management (Lecture 2) 65 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Gaussian value-at-risk

Example
We consider a short position of $1 mn on the S&P 500 futures contract.
We estimate that the annualized volatility σ̂SPX is equal to 35%

The portfolio loss is equal to L (w ) = N × RSPX where N is the exposure

amount (−$1 mn) and RSPX is the (Gaussian) return of the S&P 500
index. We deduce that the annualized loss volatility is σ̂ (L) = |N| × σ̂SPX .
The value-at-risk for a one-year holding period is:

VaR99% (w ; one year) = 2.33 × 106 × 0.35 = $815 500

By using the square-root-of-time rule, we deduce that:

815 500
VaR99% (w ; one day) = √ = $50 575
260

Thierry Roncalli Financial Risk Management (Lecture 2) 66 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Gaussian expected shortfall

By definition, we have:

ESα (w ) = E [L (w ) | L (w ) ≥ VaRα (w )]
Z ∞
1
= xfL (x) dx
1 − α FL (α)
−1

where fL and FL are the density and distribution functions of the loss L (w )

The Gaussian expected shortfall of the portfolio w is equal to:

−1


φ Φ (α)
ESα (w ) = µ (L) + σ (L)
(1 − α)

Thierry Roncalli Financial Risk Management (Lecture 2) 67 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Proof

∞  2 !
x − µ (L)
Z
1 x 1
ESα (w ) = √ exp − dx
1−α µ(L)+Φ−1 (α)σ(L) σ (L) 2π 2 σ (L)

−1
With the variable change t = σ (L) (x − µ (L)), we obtain:
Z ∞  
1 1 1
ESα (w ) = (µ (L) + σ (L) t) √ exp − t 2 dt
1 − α Φ−1 (α) 2π 2
Z ∞  
µ (L) ∞ σ (L) 1
= [Φ (t)]Φ−1 (α) + √ t exp − t 2 dt
1−α (1 − α) 2π Φ−1 (α) 2
  ∞
σ (L) 1 2
= µ (L) + √ − exp − t
(1 − α) 2π 2 Φ−1 (α)
 
σ (L) 1 −1
 2
= µ (L) + √ exp − Φ (α)
(1 − α) 2π 2

Thierry Roncalli Financial Risk Management (Lecture 2) 68 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Gaussian VaR vs Gaussian ES

The value-at-risk and the expected shortfall are both a standard

deviation-based risk measure. They coincide when the scaling parameters
cVaR = Φ−1 (αVaR ) and cES = φ Φ−1 (αES ) / (1 − αES ) are equal.

Table: Scaling factors cVaR and cES

α (in %) 95.0 96.0 97.0 97.5 98.0 98.5 99.0 99.5

cVaR 1.64 1.75 1.88 1.96 2.05 2.17 2.33 2.58
cES 2.06 2.15 2.27 2.34 2.42 2.52 2.67 2.89

Thierry Roncalli Financial Risk Management (Lecture 2) 69 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Linear factor models

Pn
When g (Ft ; w ) = i=1 wi Pi,t , the random P&L is equal to:

Π (w ) = Pt+h (w ) − Pt (w )
Xn Xn
= wi Pi,t+h − wi Pi,t
i=1 i=1
Xn
= wi (Pi,t+h − Pi,t )
i=1

We assume that the asset returns are the risk factors :

Pi,t+h = Pi,t (1 + Ri,t+h )

where Ri,t+h is the asset return between t and t + h. In this case, we

obtain:
Xn
Π (w ) = wi Pi,t Ri,t+h
i=1

Thierry Roncalli Financial Risk Management (Lecture 2) 70 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The covariance model

Let Rt be the vector of asset returns. We note Wi,t = wi Pi,t the wealth
invested (or the nominal exposure) in asset i and Wt = (W1,t , . . . , Wn,t ).
It follows that:
X n
Π (w ) = Wi,t Ri,t+h = Wt> Rt+h
i=1

If we assume that Rt+h ∼ N (µ, Σ), we deduce that µ (Π) = Wt> µ and
σ 2 (Π) = Wt> ΣWt . Therefore, the expression of the value-at-risk is:
q
VaRα (w ; h) = −Wt> µ + Φ−1 (α) Wt> ΣWt

Thierry Roncalli Financial Risk Management (Lecture 2) 71 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Example
We consider the Apple/Coca-Cola example. The nominal exposures are
$1 093.3 (Apple) and $842.8 (Coca-Cola). The estimated standard
deviation of daily returns is equal to 1.3611% for Apple and 0.9468% for
Coca-Cola, whereas the cross-correlation is equal to 12.0787%. It follows
that:

σ 2 (Π) = Wt> ΣWt

 2  2
0.9468
1.3611
= 1 093.32 × + + 842.82 ×
100 100
12.0787 1.3611 0.9468
2× × 1 093.3 × 842.8 × ×
100 100 100
= 313.80

We deduce that the 99% daily value-at-risk is equal to:

−1
√
VaR99% (w ; one day) = Φ (0.99) 313.80 = $41.21

Thierry Roncalli Financial Risk Management (Lecture 2) 72 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The factor model

CAPM (HFRM, pages 76-77)

APT (HFRM, page 77 and Exercise 2.4.5 page 119)
Application to a bond portfolio (HFRM, pages 77-80)

Thierry Roncalli Financial Risk Management (Lecture 2) 73 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Some other topics

Volatility forecasting EWMA, GARCH and SV models (HFRM, pages

80-83 and Section 10.2.4 page 664)
Other probability distributions (HFRM, pages 84-90)
Cornish-Fisher approximation (HFRM, pages 85-87)

VaRα (w ; h) = µ (L) + Z (α; γ1 (L) , γ2 (L)) × σ (L)

where:
1 2
1 3

1 3

2
Z (α; γ1 , γ2 ) = zα + zα − 1 γ1 + zα − 3zα γ2 − 2zα − 5zα γ1
6 24 36
and zα = Φ−1 (α)

Thierry Roncalli Financial Risk Management (Lecture 2) 74 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Monte Carlo methods

We assume a given probability distribution H for the risk factors:

(F1,t+h , . . . , Fm,t+h ) ∼ H

We simulate nS scenarios of risk factors and calculate the simulated

P&L Πs (w ) for each scenario s
We calculate the empirical quantile using the order statistic approach

⇒ The Monte Carlo VaR/ES is a historical VaR/ES with simulated

scenarios or the Monte Carlo VaR/ES is a parametric VaR/ES for which
it is difficult to find an analytical formula

Thierry Roncalli Financial Risk Management (Lecture 2) 75 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Identification of risk factors

We consider a portfolio containing wS stocks and wC call options on this

stock. We note St and Ct the stock and option prices at time t. We have:

Π (w ) = wS (St+h − St ) + wC (Ct+h − Ct )

If we use asset returns as risk factors, we get:

Π (w ) = wS St RS,t+h + wC Ct RC ,t+h

where RS,t+h and RC ,t+h are the returns of the stock and the option for
the period [t, t + h]

⇒ Two risk factors: RS,t+h and RC ,t+h ?

Thierry Roncalli Financial Risk Management (Lecture 2) 76 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Identification of risk factors

The problem is that the option price Ct is a non-linear function of the

underlying price St :
Ct = fC (St )
This implies that:

Π (w ) = wS St RS,t+h + wC (fC (St+h ) − Ct )

= wS St RS,t+h + wC (fC (St (1 + RS,t+h )) − Ct )

⇒ One risk factor: RS,t+h ?

Thierry Roncalli Financial Risk Management (Lecture 2) 77 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The Black-Scholes formula

The price of the call option is equal to:
CBS (St , K , Σt , T , bt , rt ) = St e (bt −rt )τ Φ (d1 ) − Ke −rt τ Φ (d2 )
where:
St is the current price of the underlying asset
K is the option strike
Σt is the volatility parameter,
T is the maturity date
bt is the cost-of-carry5
rt is the interest rate
the parameter τ = T − t is the time to maturity
The coefficients d1 and d2 are defined as follows:
1 √ √
 
1 St
d1 = √ ln + bt τ + Σt τ and d2 = d1 − Σt τ
Σt τ K 2
5 The
cost-of-carry depends on the underlying asset. We have bt = rt for
non-dividend stocks and total return indices, bt = rt − dt for stocks paying a continuous
dividend yield dt , bt = 0 for forward and futures contracts and bt = rt − rt? for foreign
exchange options where rt? is the foreign interest rate.
Thierry Roncalli Financial Risk Management (Lecture 2) 78 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Identification of risk factors

We can write the option price as follows:

Ct = fBS (θcontract ; θ)

where θcontract are the parameters of the contract (strike K and maturity
T ) and θ are the other parameters

St is obviously a risk factor

If Σt is not constant, the option price may be sensitive to the
volatility risk
The option may be impacted by changes in the interest rate or the
cost-of-carry
⇒ The choice of risk factors depends on the derivative contract (volatility
risk, dividend risk, yield curve risk, correlation risk, etc.)

Thierry Roncalli Financial Risk Management (Lecture 2) 79 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Methods to calculate VAR and ES risk measures

1 The method of full pricing (option repricing)

2 The method of sensitivities (delta-gamma-vega approximation)
3 The hybrid method

Thierry Roncalli Financial Risk Management (Lecture 2) 80 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The method of full pricing

We recall that the P&L of the s th scenario has the following expression:

Πs (w ) = g (F1,s , . . . , Fm,s ; w ) − Pt (w )

In the case of the previous example, the P&L becomes then:


wS St Rs + wC (fC (St (1 + Rs ) ; Σt ) − Ct ) with one risk factor
Πs (w ) =
wS St Rs + wC (fC (St (1 + Rs ) , Σs ) − Ct ) with two risk factors

where the pricing function is:

fC (S; Σ) = CBS (S, K , Σ, T − h, bt , rt )

Remark
In the model with two risk factors, we have to simulate the underlying
price and the implied volatility. For the single factor model, we use the
current implied volatility Σt instead of the simulated value Σs .

Thierry Roncalli Financial Risk Management (Lecture 2) 81 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

Example
We consider a long position on 100 call options with strike K = 100. The
value of the call option is $4.14, the residual maturity is 52 days and the
current price of the underlying asset is $100. We assume that Σt = 20%
and bt = rt = 5%. The objective is to calculate the daily 99% VaR and
the daily 97.5% ES with 250 historical scenarios, whose first nine values
are the following:
s 1 2 3 4 5 6 7 8 9
Rs −1.93 −0.69 −0.71 −0.73 1.22 1.01 1.04 1.08 −1.61
∆Σs −4.42 −1.32 −3.04 2.88 −0.13 −0.08 1.29 2.93 0.85

Remark
Data are available at
http: // www. thierry-roncalli. com/ download/ frm-data1. xlsx

Thierry Roncalli Financial Risk Management (Lecture 2) 82 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

⇒ The implied volatility is equal to 20%

For the first scenario, Rs is equal to −1.93% and St+h is equal to

100 × (1 − 1.93%) = 98.07. The residual maturity τ is equal to 51/252
years. It follows that:
  r
1 98.07 51 1 51
d1 = p ln + 5% × + × 20% × = −0.0592
20% × 51/252 100 252 2 252
r
51
d2 = −0.0592 − 20% × = −0.1491
252
We deduce that:
51 51
Ct+h = 98.07 × e (5%−5%) 252 × Φ (−0.0592) − 100 × e 5%× 252 × Φ (−0.1491)
= 98.07 × 1.00 × 0.4764 − 100 × 1.01 × 0.4407
= 3.093
The simulated P&L for the first historical scenario is then equal to:
Πs = 100 × (3.093 − 4.14) = −104.69
Thierry Roncalli Financial Risk Management (Lecture 2) 83 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

Table: Daily P&L of the long position on the call option when the risk factor is
the underlying price

s Rs (in %) St+h Ct+h Πs

1 −1.93 98.07 3.09 −104.69
2 −0.69 99.31 3.72 −42.16
3 −0.71 99.29 3.71 −43.22
4 −0.73 99.27 3.70 −44.28
5 1.22 101.22 4.81 67.46
6 1.01 101.01 4.68 54.64
7 1.04 101.04 4.70 56.46
8 1.08 101.08 4.73 58.89
9 −1.61 98.39 3.25 −89.22

⇒ With the 250 historical scenarios, the 99% value-at-risk is equal to

$154.79, whereas the 97.5% expected shortfall is equal to $150.04

Thierry Roncalli Financial Risk Management (Lecture 2) 84 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The option return RC is not a new risk factor

Figure: Relationship between the asset return RS and the option return RC

Thierry Roncalli Financial Risk Management (Lecture 2) 85 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Adding the risk factor Σt

Σt+h = Σt + ∆Σs

Table: Daily P&L of the long position on the call option when the risk factors
are the underlying price and the implied volatility

s Rs (in %) St+h ∆Σs (in %) Σt+h Ct+h Πs

1 −1.93 98.07 −4.42 15.58 2.32 −182.25
2 −0.69 99.31 −1.32 18.68 3.48 −65.61
3 −0.71 99.29 −3.04 16.96 3.17 −97.23
4 −0.73 99.27 2.88 22.88 4.21 6.87
5 1.22 101.22 −0.13 19.87 4.79 65.20
6 1.01 101.01 −0.08 19.92 4.67 53.24
7 1.04 101.04 1.29 21.29 4.93 79.03
8 1.08 101.08 2.93 22.93 5.24 110.21
9 −1.61 98.39 0.85 20.85 3.40 −74.21

⇒ VaR99% (w ; one day) = $181.70 and ES97.5% (w ; one day) = $172.09

Thierry Roncalli Financial Risk Management (Lecture 2) 86 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The method of sensitivities

The previous approach is called full pricing, because it consists in re-pricing

the option

In the method based on the Greek coefficients, the idea is to approximate

the change in the option price by a Taylor expansion:
Delta approach
Delta-gamma approach
Delta-gamma-theta approach
Delta-gamma-theta-vega approach
Etc.

Thierry Roncalli Financial Risk Management (Lecture 2) 87 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The delta approach

We define the delta approach as follows:

Ct+h − Ct ' ∆t (St+h − St )

where ∆t is the option delta:

∂ CBS (St , Σt , T )
∆t =
∂ St

Thierry Roncalli Financial Risk Management (Lecture 2) 88 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The delta approach applied to delta neutral portfolios

If we consider the introductory example, we have:

Π (w ) = wS (St+h − St ) + wC (Ct+h − Ct )
' (wS + wC ∆t ) (St+h − St )
= (wS + wC ∆t ) St RS,t+h

With the delta approach, we aggregate the risk by netting the different
delta exposures6 . In particular, the portfolio is delta neutral if the net
exposure is zero:

wS + wC ∆t = 0 ⇔ wS = −wC ∆t

With the delta approach, the VaR/ES of delta neutral portfolios is then
equal to zero

6A
long (or short) position on the underlying asset is equivalent to ∆t = 1 (or
∆t = −1).
Thierry Roncalli Financial Risk Management (Lecture 2) 89 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The delta-gamma approach

We can use the second-order approximation or the delta-gamma approach:

1 2
Ct+h − Ct ' ∆t (St+h − St ) + Γt (St+h − St )
2
where Γt is the option gamma:

∂ 2 CBS (St , Σt , T )
Γt =
∂ St2

Thierry Roncalli Financial Risk Management (Lecture 2) 90 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Comparison between delta and delta-gamma approaches

Figure: Approximation of the option price with the Greek coefficients

Thierry Roncalli Financial Risk Management (Lecture 2) 91 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Extension to other risk factors

The Taylor expansion can be generalized to a set of risk factors
Ft = (F1,t , . . . , Fm,t ):
m
X ∂ Ct
Ct+h − Ct ' (Fj,t+h − Fj,t ) +
∂ Fj,t
j=1
m X
m
1 X ∂ 2 Ct
(Fj,t+h − Fj,t ) (Fk,t+h − Fk,t )
2 ∂ Fj,t ∂ Fk,t
j=1 k=1

The delta-gamma-theta-vega approach is defined as follows:

1 2
Ct+h − Ct ' ∆t (St+h − St ) + Γt (St+h − St ) + Θt h + υ t (Σt+h − Σt )
2
where Θt = ∂t CBS (St , Σt , T ) is the option theta and
υ t = ∂Σt CBS (St , Σt , T ) is the option vega

⇒ We can also include vanna and volga effects

Thierry Roncalli Financial Risk Management (Lecture 2) 92 / 117
 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The Black-Scholes Greek coefficients

∆t = e (bt −rt )τ Φ (d1 )

e (bt −rt )τ φ (d1 )
Γt = √
St Σt τ
1
Θt = −rt Ke −rt τ Φ (d2 ) − √ St Σt e (bt −rt )τ φ (d1 ) −
2 τ
(bt − rt ) St e (bt −rt )τ Φ (d1 )
(bt −rt )τ
√
υt = e St τ φ (d1 )

(HFRM, Exercise 2.4.7 page 121)

Thierry Roncalli Financial Risk Management (Lecture 2) 93 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

In the case of our previous example (Slide 82), we obtain ∆t = 0.5632,

Γt = 0.0434, Θt = −11.2808 and υ t = 17.8946

We have:
Π∆
1 (w ) = 100 × 0.5632 × (98.07 − 100) = −108.69
2
Π∆+Γ
1 (w ) = −108.69 + 100 × 1
2 × 0.0434 × (98.07 − 100) = −100.61
Π∆+Γ+Θ
1 (w ) = −100.61 − 11.2808 × 1/252 = −105.09
Πυ
1 (w ) = 100 × 17.8946 × (15.58% − 20%) = −79.09
Π∆+Γ+Θ+υ
1 (w ) = −105.90 − 79.09 = −184.99

Thierry Roncalli Financial Risk Management (Lecture 2) 94 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

Table: Calculation of the P&L based on the Greek sensitivities

s Rs (in %) St+h Πs Π∆
s Π∆+Γ
s Π∆+Γ+Θ
s
1 −1.93 98.07 −104.69 −108.69 −100.61 −105.09
2 −0.69 99.31 −42.16 −38.86 −37.83 42.30
3 −0.71 99.29 −43.22 −39.98 −38.89 −43.37
4 −0.73 99.27 −44.28 −41.11 −39.96 −44.43
5 1.22 101.22 67.46 68.71 71.93 67.46
6 1.01 101.01 54.64 56.88 59.09 54.61
7 1.04 101.04 56.46 58.57 60.91 56.44
8 1.08 101.08 58.89 60.82 63.35 58.87
9 −1.61 98.39 −89.22 −90.67 −85.05 −89.53
VaR99% (w ; one day) 154.79 171.20 151.16 155.64
ES97.5% (w ; one day) 150.04 165.10 146.37 150.84

Thierry Roncalli Financial Risk Management (Lecture 2) 95 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Application to the VaR and ES

Table: Calculation of the P&L using the vega coefficient

s St+h Σt+h Πs Πυs Π∆+υ

s Πs∆+Γ+υ Πs∆+Γ+Θ+υ
1 98.07 15.58 −182.25 −79.09 −187.78 −179.71 −184.19
2 99.31 18.68 −65.61 −23.62 −62.48 −61.45 −65.92
3 99.29 16.96 −97.23 −54.40 −94.38 −93.29 −97.77
4 99.27 22.88 6.87 51.54 10.43 11.58 7.10
5 101.22 19.87 65.20 −2.33 66.38 69.61 65.13
6 101.01 19.92 53.24 −1.43 55.45 57.66 53.18
7 101.04 21.29 79.03 23.08 81.65 84.00 79.52
8 101.08 22.93 110.21 52.43 113.25 115.78 111.30
9 98.39 20.85 −74.21 15.21 −75.46 −69.84 −74.32
VaR99% (w ; one day) 181.70 77.57 190.77 179.29 183.76
ES97.5% (w ; one day) 172.09 73.90 184.90 169.34 173.81

Thierry Roncalli Financial Risk Management (Lecture 2) 96 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The hybrid method

The hybrid method consists of combining the two approaches:

1 we first calculate the P&L for each (historical or simulated) scenario
with the method based on the sensitivities;
2 we then identify the worst scenarios;
3 we finally revalue these worst scenarios by using the full pricing
method.

⇒ The underlying idea is to consider the faster approach to locate the

value-at-risk, and then to use the most accurate approach to calculate the
right value

Thierry Roncalli Financial Risk Management (Lecture 2) 97 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

The hybrid method

Table: The 10 worst scenarios identified by the hybrid method

Full pricing Greeks

i ∆−Γ−Θ−υ ∆−Θ ∆−Θ−υ
s Πs s Πs s Πs s Πs
1 100 −183.86 100 −186.15 182 −187.50 134 −202.08
2 1 −182.25 1 −184.19 169 −176.80 100 −198.22
3 134 −181.15 134 −183.34 27 −174.55 1 −192.26
4 27 −163.01 27 −164.26 134 −170.05 169 −184.32
5 169 −162.82 169 −164.02 69 −157.66 27 −184.04
6 194 −159.46 194 −160.93 108 −150.90 194 −175.36
7 49 −150.25 49 −151.43 194 −149.77 49 −165.41
8 245 −145.43 245 −146.57 49 −147.52 182 −164.96
9 182 −142.21 182 −142.06 186 −145.27 245 −153.37
10 79 −135.55 79 −136.52 100 −137.38 69 −150.68

Thierry Roncalli Financial Risk Management (Lecture 2) 98 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Backtesting

mark-to-model 6= mark-to-market

For on-exchange products, the simulated P&L is equal to:

Πs (w ) = Pt+1 (w ) − Pt (w )
| {z } | {z }
mark-to-model mark-to-market

whereas the realized P&L is equal to:

Π (w ) = Pt+1 (w ) − Pt (w )
| {z } | {z }
mark-to-market mark-to-market

Thierry Roncalli Financial Risk Management (Lecture 2) 99 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Backtesting

For exotic options and OTC derivatives, the simulated P&L is the
difference between two mark-to-model values:

Πs (w ) = Pt+1 (w ) − Pt (w )
| {z } | {z }
mark-to-model mark-to-model

and the realized P&L is also the difference between two mark-to-model
values:
Π (w ) = Pt+1 (w ) − Pt (w )
| {z } | {z }
mark-to-model mark-to-model

⇒ Model risk

Thierry Roncalli Financial Risk Management (Lecture 2) 100 / 117

 Capital requirements Definition
Statistical estimation of risk measures Computation
Risk allocation Options and derivatives

Model risk

4 types of model risk:

1 Operational risk
2 Parameter risk
3 Mis-specification risk
4 Hedging risk

(HFRM, Chapter 9, Page 491)

Thierry Roncalli Financial Risk Management (Lecture 2) 101 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

On the importance of risk allocation

Let us consider two trading desks A and B, whose risk measure is

respectively R (wA ) and R (wB ). At the global level, the risk measure is
equal to R (wA+B ). The question is then how to allocate R (wA+B ) to the
trading desks A and B:

R (wA+B ) = RC A (wA+B ) + RC B (wA+B )

Remark
This question is an important issue for the bank because risk allocation
means capital allocation:

K (wA+B ) = KA (wA+B ) + KB (wA+B )

Capital allocation is not neutral, because it will impact the profitability

of business units that compose the bank

Thierry Roncalli Financial Risk Management (Lecture 2) 102 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Euler allocation principle

We decompose the P&L as follows:

n
X
Π= Πi
i=1

where Πi is the P&L of the i th sub-portfolio

We note R (Π) the risk measure associated with the P&L
We consider the risk-adjusted performance measure (RAPM) defined
by:
E [Π]
RAPM (Π) =
R (Π)
We consider the portfolio-related RAPM of the i th sub-portfolio
defined by:
E [Πi ]
RAPM (Πi | Π) =
R (Πi | Π)

Thierry Roncalli Financial Risk Management (Lecture 2) 103 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Euler allocation principle

Based on the notion of RAPM, Tasche (2008) states two properties of risk
contributions that are desirable from an economic point of view:
1 Risk contributions R (Πi | Π) to portfolio-wide risk R (Π) satisfy the
full allocation property if:
n
X
R (Πi | Π) = R (Π)
i=1

2 Risk contributions R (Πi | Π) are RAPM compatible if there are some

εi > 0 such that:

RAPM (Πi | Π) > RAPM (Π) ⇒ RAPM (Π + hΠi ) > RAPM (Π)

for all 0 < h < εi

⇒ This property means that assets with a better risk-adjusted
performance than the portfolio continue to have a better RAPM if their
allocation increases in a small proportion
Thierry Roncalli Financial Risk Management (Lecture 2) 104 / 117
 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Euler allocation principle

Tasche (2008) shows that if there are risk contributions that are RAPM
compatible, then R (Πi | Π) is uniquely determined as:
d
R (Πi | Π) = R (Π + hΠi )
dh h=0

and the risk measure is homogeneous of degree 1

If we consider the risk measure R (w ) defined in terms of weights, the risk

contribution of sub-portfolio i is uniquely defined as:
∂ R (w )
RC i = wi
∂ wi
and the risk measure satisfies the Euler decomposition (or the Euler
allocation principle):
n n
X ∂ R (w ) X
R (w ) = wi = RC i
∂ wi
i=1 i=1

Thierry Roncalli Financial Risk Management (Lecture 2) 105 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to Gaussian risk measures

If we assume that the portfolio return R (w ) is a linear function of the
weights w , the expression of the standard deviation-based risk measure
becomes:
√
>
R (w ) = −µ (w ) + c · σ (w ) = −w µ + c · w > Σw

where µ and Σ are the mean vector and the covariance matrix of
sub-portfolios

We have:
∂ R (w ) 1 >

−1/2 Σw
= −µ + c · w Σw (2Σw ) = −µ + c · √
∂w 2 w > Σw

The risk contribution of the i th sub-portfolio is then:

 
(Σw )i
RC i = wi · −µi + c · √
w > Σw
Thierry Roncalli Financial Risk Management (Lecture 2) 106 / 117
 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to Gaussian risk measures

We verify that the standard deviation-based risk measure satisfies the full
allocation property:
n n  
X X (Σw )i
RC i = wi · −µi + c · √
i=1 i=1 w > Σw
 
Σw
= w > −µ + c · √
w > Σw
√
>
= −w µ + c · w > Σw
= R (w )

Thierry Roncalli Financial Risk Management (Lecture 2) 107 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to Gaussian risk measures

Gaussian VaR risk contribution:

 
(Σw )i
RC i = wi · −µi + Φ−1 (α) · √
w > Σw
Gaussian ES risk contribution:
!
−1


φ Φ (α) (Σw )i
RC i = wi · −µi + ·√
(1 − α) w > Σw

Thierry Roncalli Financial Risk Management (Lecture 2) 108 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to Gaussian risk measures

block
We consider the Apple/Coca-Cola portfolio that has been used for
calculating the Gaussian VaR. We recall that the nominal exposures were
$1 093.3 (Apple) and $842.8 (Coca-Cola), the estimated standard
deviation of daily returns was equal to 1.3611% for Apple and 0.9468% for
Coca-Cola and the cross-correlation of stock returns was equal to
12.0787%.

Thierry Roncalli Financial Risk Management (Lecture 2) 109 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to Gaussian risk measures

Table: Risk decomposition of the 99% Gaussian value-at-risk

Asset wi MRi RC i RC ?i
Apple 1093.3 2.83% 30.96 75.14%
Coca-Cola 842.8 1.22% 10.25 24.86%
R (w ) 41.21

Table: Risk decomposition of the 99% Gaussian expected shortfall

Asset wi MRi RC i RC ?i
Apple 1093.3 3.24% 35.47 75.14%
Coca-Cola 842.8 1.39% 11.74 24.86%
R (w ) 47.21

Thierry Roncalli Financial Risk Management (Lecture 2) 110 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Application to non-normal risk measures

Generalized formulas
The risk contribution for the value-at-risk is equal to:

RC i = E [Li | L (w ) = VaRα (L)]

The risk contribution for the expected shortfall is equal to:

RC i = E [Li | L (w ) ≥ VaRα (L)]

⇒ These formulas can easily be applied to historical and Monte Carlo risk
measures (HFRM, pages 109-116)

Thierry Roncalli Financial Risk Management (Lecture 2) 111 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Calculating the Gaussian VaR risk contribution

Asset returns are assumed to be Gaussian:

R ∼ N (µ, Σ)

The portfolio’s loss is equal to:

n
X
L (w ) = −R (w ) = − wi Ri = −w > R
i=1

We notice that:
Li = −wi Ri
and:

E [Li | L (w ) = VaRα (w ; h)] = −wi E [Ri | L (w ) = VaRα (w ; h)]

Thierry Roncalli Financial Risk Management (Lecture 2) 112 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Calculating the Gaussian VaR risk contribution

We have:    
R In
= R
L (w ) −w >
and:
     
R µ Σ −Σw
∼N ,
L (w ) −w > µ −w > Σ w > Σw

We would like to calculate:

RC i = −wi E [Ri | L (w ) = VaRα (w ; h)]

Thierry Roncalli Financial Risk Management (Lecture 2) 113 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Conditional distribution in the case of the normal

distribution

Let us consider a Gaussian random vector defined as follows:

     
X µx Σx,x Σx,y
∼N ,
Y µy Σy ,x Σy ,y

We have:

Y | X = x ∼ N µy |x , Σy ,y |x
where:
µy |x = E [Y | X = x] = µy + Σy ,x Σ−1
x,x (x − µx )

and:
Σy ,y |x = cov (Y | X = x) = Σy ,y − Σy ,x Σ−1
x,x Σx,y

Thierry Roncalli Financial Risk Management (Lecture 2) 114 / 117

 Capital requirements Definition
Statistical estimation of risk measures Application to Gaussian risk measures
Risk allocation Application to non-normal risk measures

Calculating the Gaussian VaR risk contribution

√
> −1
Since VaRα (w ; h) = −w µ + Φ (α) w > Σw , we have:
h √ i
E [R | L (w ) = VaRα (w ; h)] = E R | L (w ) = −w > µ + Φ−1 (α) w > Σw
>

−1
= µ − Σw w Σw ·
 √ 
−w > µ + Φ−1 (α) w > Σw − −w > µ


√
−1 w > Σw
= µ − Φ (α) Σw −1
(w > Σw )
−1 Σw
= µ − Φ (α) √
w > Σw
and:
 
Σw wi · (Σw )i
RC i = −wi µ − Φ−1 (α) √ = −wi µi + Φ−1 (α) √
w > Σw i w > Σw

Thierry Roncalli Financial Risk Management (Lecture 2) 115 / 117

 Exercises
Appendix References

Exercises

Value-at-risk
Exercise 2.4.2 – Covariance matrix
Exercise 2.4.4 – Value-at-risk of a long/short portfolio
Exercise 2.4.4 – Value-at-risk of an equity portfolio hedged with put
options

Expected shortfall
Exercise 2.4.10 – Expected shortfall of an equity portfolio
Exercise 2.4.11 – Risk measure of a long/short portfolio

Options and derivatives

Exercise 2.4.6 – Risk management of exotic options
Exercise 2.4.7 – P&L approximation with Greek sensitivities

Thierry Roncalli Financial Risk Management (Lecture 2) 116 / 117

 Exercises
Appendix References

References

Basel Committee on Banking Supervision (1996)

Amendment to the Capital Accord to Incorporate Market Risks,
January 1996
Basel Committee on Banking Supervision (2009
Revisions to the Basel II Market Risk Framework, July 2009
Basel Committee on Banking Supervision (2019)
Minimum Capital Requirements for Market Risk, January 2019.
Roncalli, T. (2020)
Handbook of Financial Risk Management, Chapman and Hall/CRC
Financial Mathematics Series, Chapter 2.

Thierry Roncalli Financial Risk Management (Lecture 2) 117 / 117