Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Quantitative Finance
Lecture 8, Value-at-Risk and Expected Shortfall

Adam Farago

Department of Economics, University of Gothenburg

2019

Quantitative Finance, Lecture 8 1 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Introduction

Ways to measure market risk:

Scenario-based risk measures, stress testing

Factor sensitivity analysis

Risk measures based on the loss distribution approach

Value-at-Risk (VaR)
Expected shortfall (ES)

Quantitative Finance, Lecture 8 2 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Introduction

Ways to measure market risk:

Scenario-based risk measures, stress testing

Factor sensitivity analysis

Risk measures based on the loss distribution approach

Value-at-Risk (VaR)
Expected shortfall (ES)

Quantitative Finance, Lecture 8 2 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Introduction

So far we were interested in small changes of the underlying. But how

bad things can get?
What loss level is such that we are X% confident it will not be
exceeded in N business days?
VaR is certainly the risk measurement standard in the industry today.
Regulators base the capital they require banks to keep on VaR.
Mutual funds and hedge funds use it as a risk measure to report to clients.

It captures an important aspect of risk in a single number.

It is easy to understand.
It asks the simple question: How bad can things get?

Quantitative Finance, Lecture 8 3 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Formalization
Consider a fixed time period [t, t + ∆t] , where ∆t can be e.g 1 day, 10
days, one month, one year, etc.
We can define the following objects:
The change in the value of the portfolio during the period [t, t + ∆t] is
∆V ≡ Vt+∆t − Vt
The return on the portfolio during the period [t, t + ∆t] is
∆V
RV ≡
Vt
The loss of the portfolio during the period [t, t + ∆t] is
L ≡ −∆V = − (Vt+∆t − Vt )
The relative loss of the portfolio during the period [t, t + ∆t] is
L
RL ≡ = −RV
Vt
Quantitative Finance, Lecture 8 4 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Formalization

The distribution FL (x) of the random variable L is called the loss

distribution (over the time period [t, t + ∆t])

It is defined as
FL (x) = P (L ≤ x)

FL (x) is the cumulative distribution function (cdf) of L.

Be aware of the sign! L takes a positive value if the value of the

portfolio goes down.

Quantitative Finance, Lecture 8 5 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Defintion
“What loss level is such that we are X% confident it will not be
exceeded in N business days?”
Value-at-Risk (VaR)
Given the loss L and a confidence level α ∈ (0, 1), VaRα (L) is given by the
smallest number x such that the probability that the loss exceeds x is not
larger than 1 − α. That is,

VaRα (L) ≡ inf {x ∈ R : P (L ≥ x) ≤ 1 − α}

Note that
VaRα (L) = inf {x ∈ R : 1 − P (L ≤ x) ≤ 1 − α}
= inf {x ∈ R : P (L ≤ x) ≥ α}
= inf {x ∈ R : FL (x) ≥ α}

Quantitative Finance, Lecture 8 6 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk for discrete random variables

0.5
0.45
If α = 0.99, then
0.4
{x ∈ R : FL (x) ≥ 0.99} = [6000, ∞[

probability
so VaR (L)0.99 = 6000 0.3

0.2 0.2
0.2
If α = 0.97, then
{x ∈ R : FL (x) ≥ 0.97} = [4000, ∞[ 0.1 0.06 0.06
0.03
so VaR (L)0.97 = 4000
0
1
If α = 0.95, then
0.8
{x ∈ R : FL (x) ≥ 0.95} = [4000, ∞[
so VaR (L)0.95 = 4000 0.6
cdf
0.4
If α = 0.9, then
{x ∈ R : FL (x) ≥ 0.9} = [2000, ∞[ 0.2
so VaR (L)0.9 = 2000
0
-6000 -4000 -2000 0 2000 4000 6000
L (loss)

Quantitative Finance, Lecture 8 7 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk for continuous random variables

10-4
For continuous random variables, 2

the definition simplifies.

1.5
Reason: The cdf of a continuous

pdf
random variable is continuous 1

and strictly increasing.

0.5
Therefore: 1-
0
1
VaR (L)α
= {x ∈ R : FL (x) = α} 0.8

= FL−1 (α) 0.6

cdf

where FL−1 is the inverse cdf.

0.4

0.2
That, is the VaR is a quantile of
the distribution. 0
-6000 -4000 -2000 0 2000 4000 6000
L (loss)

Quantitative Finance, Lecture 8 8 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Comments

VaRα (L) will typically be a positive number, since it is connected to

the loss-distribution.

The VaR can also be reported in terms of returns. Because of the

linearity of VaR,
VaRα (L)
VaRα (RL ) =
Vt
where Vt denotes the current total value of the portfolio.

If we can find an analytical expression for the inverse function FL−1 (y ),

then we can also find an analytical expression for the VaR.

Quantitative Finance, Lecture 8 9 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Comments
VaR at different horizons

The Value-at-Risk is defined for a certain time horizon (as L is defined

over a a certain time period [t, t + ∆t])
If ∆t is one day, then we talk about a one-day VaR.
If ∆t is h days, then we talk about a h-day VaR.

In market risk, the typical horizon is 1 day, 5 days, or 10 days.

Quantitative Finance, Lecture 8 10 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Linearity of the Value-at-Risk

Lineraity
Let a ∈ R and b > 0 constants. Then,

VaRα (a + b · L) = a + b · VaRα (L)

Proof:

VaRα (a + bL) = inf {x ∈ R : P (a + bL ≤ x) ≥ α}

   
x −a
= inf x ∈ R : P L ≤ ≥α
b
= inf {a + bx̃ ∈ R : P (L ≤ x̃) ≥ α}
= a + b inf {x̃ ∈ R : P (L ≤ x̃) ≥ α}
= a + bVaRα (L)

Quantitative Finance, Lecture 8 11 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Analytical VaR when L is normally distributed

Let L be normally distributed with mean µ and variance σ 2 , that is,

2


L ∼ N µ, σ . Then  
x −µ
FL (x) = Φ
σ
where Φ (z) is the cdf for a standard normal random variable, i.e.,
Z z
1 y2
Φ (z) = √ e − 2 dy
−∞ 2π

So FL−1 (y ) is obtained by solving for x in the equation y = FL (x)

 
x −µ x −µ
Φ =y ⇔ = Φ−1 (y ) ⇔ x = µ + σΦ−1 (y )
σ σ

Quantitative Finance, Lecture 8 12 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Analytical VaR when L is normally distributed

So, the Value-at-Risk at confidence level α is

VaRα (L) = µ + σΦ−1 (α)

where Φ−1 (y ) can be obtained from tables or in any software such as

Matlab, Excel, Stata, etc. (in Matlab, write icdf(’norm’,y,0,1)).

Some often used values are:

α Φ−1 (α)
0.999 3.090
0.995 2.576
0.99 2.326
0.95 1.645
0.9 1.282

Quantitative Finance, Lecture 8 13 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Example: portfolio of stocks

Consider a portfolio of US stocks, which invests 10,000$ in each of the

following five stocks (the total value of the portfolio is 50,000$): Apple,
Coca-Cola, Exxon Mobil, General Electric, and Procter & Gamble.

Let us calculate the 5-day VaR and ES with α = 0.95 using the
assumption that the return on the five assets follow a multivariate
normal distribution.

Quantitative Finance, Lecture 8 14 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Example: portfolio of stocks

Multivariate normal returns
Means, standard deviations, and correlations of the weekly (five-day)
returns on these stocks using the sample Jan 2, 1986 to Jun 29, 2018:
Mean Std. Dev. Correlations
1. 2. 3. 4. 5.
1. Coca-Cola 0.30 3.30 1.00 0.40 0.42 0.18 0.54
2. Exxon 0.26 3.00 1.00 0.47 0.18 0.38
3. GE 0.24 4.14 1.00 0.28 0.40
4. Apple 0.60 6.39 1.00 0.12
5. P&G 0.28 3.19 1.00
Means and standard deviations are expressed in % per five trading days.

Let us assume that the five return series follow a multivariate normal
distribution
R ∼ N(µ, Σ)
where µ is the mean vector from the above table and Σ is the
variance-covariance matrix based on the above descriptives.
Quantitative Finance, Lecture 8 15 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Example: portfolio of stocks

Multivariate normal returns
What is the five-day VaR with α = 0.95
The distribution of the portfolio returns is
RV ∼ N µp , σp2

with
µp = w 0 µ = 0.0033
√
σp = w 0 Σw = 0.027
where w 0 = [0.2 0.2 0.2 0.2 0.2].
The loss is L = −RV Vt (with Vt = 50000), so
L ∼ N −µp Vt , σp2 Vt2

Therefore,
VaRα (L)
VaRα (L) = −µp Vt + σp Vt Φ−1 (0.95) = 2057.1 and = 4.11%
Vt
Quantitative Finance, Lecture 8 16 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Approximate-analytical solutions

Unfortunately, if we go outside the multivariate-normal framework, we

quickly lose the possibility of analytical solutions. Numerical techniques
(see the next lecture) come handy in these cases.

One important case: when we have derivatives in the portfolio, it is

not easy to derive the distribution of L.

Even if the return on the underlying asset is assumed to be normally

distributed, the return on the derivatives is highly non-normal.

In these cases we can try to come up with approximate-analytical

solutions.

Quantitative Finance, Lecture 8 17 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Example
Continuing from the previous lecture
Consider a portfolio with three assets:
S&P 500 index (you can think of a well-diversified US stock portfolio)
Call option on the S&P 500 (strike price: 2900, maturity: 6 months)
Put option on the S&P 500 (strike price: 2750, maturity: 6 months)

Asset Pi ($) ni ni Pi ($)

S&P 500 2800 100 280,000

Call option 126.8 50 6,339
Put option 120.0 150 18,003

Portfolio value (Vt ) 304,342

Note: The volatility of the S&P 500 is assumed to be 20%, while the risk-free rate is 2%
(both yearly).
Quantitative Finance, Lecture 8 18 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR

Let us take the first order (delta-) approximation of the loss

∂V
L = −∆V = − ∆S = −δV · ∆S
∂S
Assume that stock returns are normally distributed
∆S
∼ N µS , σS2 ⇔ ∆S ∼ N St µS , St2 σS2



St
Then, VaRα (∆S) can be calculated using the formula from the previous
lecture, and
VaRα (L) = −δV VaRα (∆S)
BUT be careful with the sign of δV !!

Quantitative Finance, Lecture 8 19 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
δV < 0 δV > 0
200 200

100 100

0 0
L

L
-100 -100

-200 -200
S

S
pdf of

pdf of

-200 -100 0 100 200 -200 -100 0 100 200

S S

We need Φ−1 (α) We need Φ−1 (1 − α)

Quantitative Finance, Lecture 8 20 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
We need (
St µS + St σS Φ−1 (1 − α) if δV ≥ 0
St µS + St σS Φ−1 (α) if δV < 0
The Delta-Normal VaR is
(
−δV St µS − δV St σS Φ−1 (1 − α) if δV ≥ 0
VaRα (L) =
−δV St µS − δV St σS Φ−1 (α) if δV < 0

But also note that since the normal distribution is symmetric,

Φ−1 (1 − α) = −Φ−1 (α) , so

Delta-Normal VaR
(
−δV St µS + δV St σS Φ−1 (α) if δV ≥ 0
VaRα (L) =
−δV St µS − δV St σS Φ−1 (α) if δV < 0
Quantitative Finance, Lecture 8 21 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
Special cases
∆S
∼ N 0, σS2 , then


If St
(
δV St σS Φ−1 (α) if δV ≥ 0
VaRα (L) = −1
−δV St σS Φ (α) if δV < 0

If ∆S 2


St ∼ N 0, σS and you have only one stock in your portfolio
(implying δV = 1), then
VaRα (L)
VaRα (L) = St σS Φ−1 (α) or = σS Φ−1 (α)
St
If ∆S 2


St ∼ N 0, σS and you have n stocks in your portfolio (implying
δV = n), then
VaRα (L)
VaRα (L) = nSt σS Φ−1 (α) = Vt σS Φ−1 (α) or = σS Φ−1 (α)
Vt
Quantitative Finance, Lecture 8 22 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Horizon

We start from the assumption that

∆S
∼ N µS , σS2


St
We can easily take into account the horizon by choosing the parameters
of the distribution appropriately, so that they correspond to returns over
the horizon [t, t + ∆t].
This will also be true when we use the t-distribution instead (later in
this lecture).

Quantitative Finance, Lecture 8 23 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
Example

Consider our example portfolio with three assets:

S&P 500 index (you can think of a well-diversified US stock portfolio)
Call option on the S&P 500 (strike price: 2900, maturity: 6 months)
Put option on the S&P 500 (strike price: 2750, maturity: 6 months)

The yearly volatility of the S&P 500 is assumed to be 20%.

The VaR parameters:

We look at a 5-day horizon.
We use the confidence level of 95% (α = 0.95).
∆S


The 5-day return on the S&P 500 is normally distributed: St ∼ N 0, σS2
q
5
σS = 0.2 · 250

Quantitative Finance, Lecture 8 24 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
Example
In our example, the 5-day VaR with α = 0.95 can be calculated as (note
that δV > 0)
VaR0.95 (∆S) = St · σS · Φ−1 (0.05)
= 2800 · 0.0283 · (−1.645) = −130.3
and
VaR0.95 (L) = −δV VaR0.95 (∆S)
= (−63.8) · (−130.3) = 8307.2
Equivalently
VaR0.95 (L) = δV St σS Φ−1 (0.95)
= 63.8 · 2800 · 0.0283 · 1.645 = 8307.2
VaR0.95 (L)
This also means Vt = 2.73%
Quantitative Finance, Lecture 8 25 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
Example
10000
8307.2

V 5000

-5000
S
pdf of

-200 -100 0 100 200

Quantitative Finance, Lecture 8 S 26 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
When is it a good approximation?

As you can see, in our example, the delta-approximation is not too

accurate. In what cases does this approximation work well?

Short period VaR (e.g., 1-day).

Because for a shorter the period, smaller movements in ∆S are expected.

Quantitative Finance, Lecture 8 27 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
When is it a good approximation?
10000

5000
3715.1
V

-5000
S
pdf of

-200 -100 0 100 200

Quantitative Finance, Lecture 8 S 28 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal VaR
When is it a good approximation?

As you can see, in our example, the delta-approximation is not too

accurate. In what cases does this approximation work well?

Short period VaR (e.g., 1-day).

Because for a shorter the period, smaller movements in ∆S are expected.

Linear assets (e.g., the portfolio contains only the S&P 500).
Then the delta-approximation is exact.

Quantitative Finance, Lecture 8 29 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Gamma-Normal VaR

We can improve the accuracy if we use a Delta-Gamma approximation

instead
1 ∂2V
 
∂V 2
L = −∆V = − ∆S + (∆S)
∂S 2 ∂S 2
 
1 2
= − δV · ∆S + γV (∆S)
2

But, (∆S)2 is a squared normally distributed (chi-squared) variable

⇒ NOT normally distributed.
Thus, we cannot use the analytical formula we used previously.
But we can calculate the VaR in two steps.

Quantitative Finance, Lecture 8 30 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Gamma-Normal VaR

First, calculate the VaR for the stock price changes, but again you have
to be careful, which tail of the distribution you look at:

VaR0.95 (∆S) = St · σS · Φ−1 (1 − α) = −130.3

Then use this value to calculate the portfolio VaR:

 
1 2
VaR0.95 (L) = − δV · VaR0.95 (∆S) + γV (VaR0.95 (∆S))
2
= 6645.5

VaR0.95 (L)
= 2.18%
Vt

Quantitative Finance, Lecture 8 31 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Gamma-Normal VaR
10000

6645.5
5000
V

-5000
S
pdf of

-200 -100 0 100 200

S
Quantitative Finance, Lecture 8 32 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Full-Valuation-Normal VaR

We can take the previous idea one step further:

First, calculate VaR0.95 (∆S) as in the case of the Delta-Gamma
approximation.
Second, calculate the price of all assets in the portfolio assuming that the
stock price drops to this value.
Third, calculate the value of the portfolio in this scenario, and the associated
loss.
This is called the Full-Valuation approach.
In our example, the VaR with this approach is

VaR0.95 (L) = 6601.8

VaR0.95 (L)
= 2.17%
Vt

Quantitative Finance, Lecture 8 33 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Full-Valuation-Normal VaR
10000

6601.8
5000
V

-5000
S
pdf of

-200 -100 0 100 200

S
Quantitative Finance, Lecture 8 34 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution
How reasonable is the assumption of normally distributed returns?
Many financial data series have excess kurtosis or fat tails:
The empirical distribution is more peaked than the normal distribution.
The empirical distribution has heavier tails than the normal distribution.
This means that small changes and large changes are more likely than the
normal distribution would suggest.

Quantitative Finance, Lecture 8 35 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution
How reasonable is the assumption of normally distributed returns?
Many financial data series have excess kurtosis or fat tails:
The empirical distribution is more peaked than the normal distribution.
The empirical distribution has heavier tails than the normal distribution.
This means that small changes and large changes are more likely than the
normal distribution would suggest.
Example: S&P 500 from 2010 to 2017

0.6% 30.2% 30.2% 0.6%

38.4%

µ − 2.5σ µ − .5σ µ + .5σ µ + 2.5σ

Quantitative Finance, Lecture 8 35 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution
How reasonable is the assumption of normally distributed returns?
Many financial data series have excess kurtosis or fat tails:
The empirical distribution is more peaked than the normal distribution.
The empirical distribution has heavier tails than the normal distribution.
This means that small changes and large changes are more likely than the
normal distribution would suggest.
Example: S&P 500 from 2010 to 2017

0.6% 30.2% 30.2% 0.6%

1.8% 20.5% 23.7% 1.2%
38.4%
52.8%

µ − 2.5σ µ − .5σ µ + .5σ µ + 2.5σ

Quantitative Finance, Lecture 8 35 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution

The t-distribution has lots of applications in statistics.

Characteristics:
Symmetric and bell-shaped (like the normal distribution).
Has heavier tails: i.e., it is more prone to producing values that fall far from
its mean.
ν
Has a mean of zero and ν−2 variance (the variance is only defined if ν > 2).
Described by one parameter: ν, the degrees of freedom:
Fat tails increase when degrees of freedom decreases.
When degrees of freedom go to infinity, t-distribution converges to normal
distribution.

Quantitative Finance, Lecture 8 36 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution

Let us assume that the loss distribution can be described by a linear

transformation of a t-distributed random variable, i.e.,

L = µ + ωT , with T ∼ t (ν)

The expected value and variance of the loss distribution are

ν
E (L) = µ , and Var (L) = ω 2
ν−2
Let gν (x) denote the density function (pdf) and tν (x) denote the
distribution function (cdf) of T .

Note: This distribution is referred to as “t Location-Scale Distribution” in

Matlab

Quantitative Finance, Lecture 8 37 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution
Have a look at the S&P 500 example again.
Consider a t-distributed
q variable, L, with
ν = 3, µ, and ω = σ 2 ν−2
ν
Note that this implies E (L) = µ, and Var (L) = σ 2

0.6% 30.2% 30.2% 0.6%

1.8% 20.5% 23.7% 1.2%
38.4%
52.8%

µ − 2.5σ µ − .5σ µ + .5σ µ + 2.5σ

Quantitative Finance, Lecture 8 38 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

t-distribution
Have a look at the S&P 500 example again.
Consider a t-distributed
q variable, L, with
ν = 3, µ, and ω = σ 2 ν−2
ν
Note that this implies E (L) = µ, and Var (L) = σ 2

0.6% 30.2% 30.2% 0.6%

1.8% 20.5% 23.7% 1.2%
1.2% 21.3% 38.4% 21.3% 1.2%
52.8%
55.0%

µ − 2.5σ µ − .5σ µ + .5σ µ + 2.5σ

Quantitative Finance, Lecture 8 38 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Value-at-Risk
Let us assume that the Loss distribution can be described as

L = µ + ωT , with T ∼ t (ν)
The Value-at-Risk at confidence level α is
VaRα (L) = µ + ωtν−1 (α)

where tν−1 (x) denotes the inverse of the distribution function of T . Values
can be obtained from using, e.g., Matlab (write icdf(’t’,x,ν)).
Some values with ν = 3:
α tν−1 (α)
0.999 10.215
0.995 5.841
0.99 4.541
0.95 2.353
0.9 1.638
Quantitative Finance, Lecture 8 39 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-t VaR

Let us take the delta-approximation of the loss L = −δV · ∆S

Assume that stock return can be described by
∆S
= µS + ωS T , with T ∼ t (ν)
St
Then,
∆S = St µS + St ωS T
We will use the linearity of VaR. Again, be careful with the sign of δV .
The Delta-t VaR is
(
−δV St µS − δV St ωS tν−1 (1 − α) if δV ≥ 0
VaRα (L) =
−δV St µS − δV St ωS tν−1 (α) if δV < 0

Quantitative Finance, Lecture 8 40 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-t VaR

The t distribution is symmetric, tν−1 (1 − α) = −tν−1 (α) , so

Delta-t VaR
(
−δV St µS + δV St ωS tν−1 (α) if δV ≥ 0
VaRα (L) =
−δV St µS − δV St ωS tν−1 (α) if δV < 0

The special cases we considered for the Delta-Normal VaR apply here
similarly.

Quantitative Finance, Lecture 8 41 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Example
The Delta-Gamma and Full Valuation approaches can be done similarly
as in case of the normal distribution.
In our example q
5
Normal distribution: µS = 0 and σS = 0.2 250
q
t distribution: µS = 0, ωS = σS ν−2
ν , and ν = 3

Normal t
α = 0.95
Delta 8307.2 2.73% 6862.1 2.25%
Delta-Gamma 6645.5 2.18% 5728.2 1.88%
Full Valuation 6601.8 2.17% 5702.2 1.87%
α = 0.99
Delta 11749.0 3.86% 13240.0 4.35%
Delta-Gamma 8425.1 2.77% 9019.0 2.96%
Full Valuation 8319.1 2.73% 8878.5 2.92%
Quantitative Finance, Lecture 8 42 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Coherent Risk Measures

Properties of VaR

Monotonicty
If hthere are
i two portfolios with loss distributions L and L̃ such that
P L ≤ L̃ = 1, then
 
VaRα (L) ≤ VaRα L̃

If one portfolio always produces a bigger loss than another its risk measure
should be greater. In other words, if one portfolio always performs worse
than another portfolio, it clearly should be viewed as more risky.

Quantitative Finance, Lecture 8 43 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Coherent Risk Measures

Properties of VaR

Lineraity
Let a ∈ R and b > 0 constants. Then,

VaRα (a + b · L) = a + b · VaRα (L)

Linearity has two important implications:

Translation invariance: If we add an amount of cash K to a portfolio its
risk measure should go down by K . The cash provides a buffer against
losses and should reduce the capital requirement by K .
Homogeneity: Changing the size of a portfolio by λ should result in the
risk measure being multiplied by λ. If we double the size of the
portfolio, we should require twice as much capital.

Quantitative Finance, Lecture 8 44 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Coherent Risk Measures

Properties of VaR

Subadditivity
Consider two portfolios with loss distributions L and L̃. The risk measure
RM satisfies subadditivity if
   
RM L + L̃ ≤ RM (L) + RM L̃

This captures a diversification effect, saying that the riskiness should not
increase when merging two portfolios.

VaR as a risk measure does not fulfill subadditivity, i.e., we can find
examples when
   
VaRα L + L̃ > VaRα (L) + VaRα L̃

Quantitative Finance, Lecture 8 45 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Coherent Risk Measures

Coherent risk measures

A risk measure is coherent if it satisfies the following properties:
1 Monotonicity
2 Translation invariance
3 Homogeneity
4 Subadditivity

Quantitative Finance, Lecture 8 46 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Coherent Risk Measures

Coherent risk measures

A risk measure is coherent if it satisfies the following properties:
1 Monotonicity X
2 Translation invariance X
3 Homogeneity X
4 Subadditivity ×

VaR satisfies the first three conditions but not the fourth one, i.e.,
it is not a coherent risk measure.

Quantitative Finance, Lecture 8 46 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Expected Shortfall
Formalization

Recall the following

L is the loss of a portfolio during the period [t, t + ∆t].
FL (x) = P (L ≤ x) is the loss distribution.

Expected Shortfall (ES)

Let L be a loss which is a continuous random variable with distribution
function FL (x). Then, for any confidence level α ∈ (0, 1), the expected
shortfall is
ESα (L) = E [L | L ≥ VaRα (L)]

ES asks: “If things do get bad, what is the expected loss?”

ES is also sometimes referred to as Conditional Value-at-Risk (CVaR),
or Expected Tail Loss

Quantitative Finance, Lecture 8 47 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

ES is a coherent risk measures

Coherent risk measures

A risk measure is coherent if it satisfies the following properties:
1 Monotonicity
2 Translation invariance
3 Homogeneity
4 Subadditivity

Quantitative Finance, Lecture 8 48 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

ES is a coherent risk measures

Coherent risk measures

A risk measure is coherent if it satisfies the following properties:
1 Monotonicity X
2 Translation invariance X
3 Homogeneity X
4 Subadditivity X

ES satisfies all four conditions, i.e., it is a coherent risk measure.

Quantitative Finance, Lecture 8 48 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

ES when L is normally distributed

Let L be normally distributed with mean µ and variance σ 2 , that is,
L ∼ N µ, σ 2 . It can be shown that

φ x−µ


σ
E [L | L > x] = µ + σ
1 − Φ x−µ


σ
We also know that
ESα (L) = E [L | L ≥ VaRα (L)]
So, x = VaRα (L) = µ + σΦ−1 (α), implying
φ Φ−1 (α)


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


=µ+σ
1−α
where φ (x) is the probability density function of the standard normal
distribution.
Quantitative Finance, Lecture 8 49 / 56
Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

ES when L is normally distributed

The φ (x) values can be obtained from tables or in any software such as
Matlab, Excel, Stata, etc. (in Matlab, write normpdf(x)).
Some helpful values:
φ(Φ−1 (α))
α 1−α
0.999 3.367
0.995 2.892
0.99 2.665
0.95 2.063
0.9 1.755

Quantitative Finance, Lecture 8 50 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal ES

Let us take the first order (delta-) approximation of the loss

∂V
L = −∆V = − ∆S = −δV · ∆S
∂S
Assume that stock returns are normally distributed
∆S
∼ N µS , σS2 ⇔ ∆S ∼ N St µS , St2 σS2



St
Then, ESα (∆S) can be calculated analytically and then

ESα (L) = −δV ESα (∆S)

BUT be careful with the sign of δV !!

Quantitative Finance, Lecture 8 51 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal ES
δV < 0 δV > 0
200 200

100 100

0 0
L

L
-100 -100

-200 -200
S

S
pdf of

pdf of

-200 -100 0 100 200 -200 -100 0 100 200

S S

We need E [∆S | ∆S > x] We need E [∆S | ∆S < x]

Quantitative Finance, Lecture 8 52 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Delta-Normal ES
Without going into the details of the derivation here

Delta-Normal ES

−δ S µ + δ S σ φ(Φ−1 (α)) if δV ≥ 0
V t S V t S 1−α
ESα (L) =
−δ S µ − δ S σ φ(Φ−1 (α)) if δV < 0
V t S V t S 1−α

You can easily derive the special cases we considered for the
Delta-Normal VaR.
In our example:

α = 0.95 α = 0.99
VaR 8307.2 11749.0
ES 10417.5 13460.4

Quantitative Finance, Lecture 8 53 / 56

Value-at-Risk Normal VaR Approximate - Normal VaR t VaR Coherent risk measures Expected shortfall

Comments

Important: the Delta-Gamma and Full Valuation approaches do not

work for the ES.

If you assume ∆S 2


St ∼ N 0, σS , as we did today, then the delta-normal
method gives
VaRα (L) = δV St σS Φ−1 (α)
φ(Φ−1 (α))
ESα (L) = δV St σS 1−α
The VaR and the ES are tightly linked, so the ES does not provide too
much additional information. This is due to the normality assumption. If
you have more general models, the ES can provide useful extra
information.

Similarly, you can derive an analytical formula for ES, when returns
follow a t-distribution (we will not do it now).

Quantitative Finance, Lecture 8 54 / 56

Slides that are important for the exam

Introducing VaR
Basic concepts and notation: Slides 4-5
Know and understand the definition of VaR: Slides 6, 8-10
Linearity of VaR (be able to provide the proof also): Slide 11

Analytical VaR calculations

When L is normally distributed: Slides 12-13
Delta-Normal VaR (be able to derive the formula and apply it): Slides 19-25
Delta-Gamma-Normal VaR (be able to calculate it): Slides 30-31
Full-Valuation-Normal VaR (understand the logic and be able to implement):
Slide 33
t-distribution (understand the formulation): Slides 36-39
Delta-t VaR (be able to derive the formula and apply it): Slides 40-41
Delta-Gamma-t and Full-Valuation-t VaR (be able to calculate it): not on
slides but the procedure is exactly the same as for the normal case

Quantitative Finance, Lecture 8 55 / 56

Slides that are important for the exam

Coherent risk measures (know the properties that define a coherent RM,
be able to give an economic interpretation of them): Slides 43-46, 48

Expected Shortfall
Definition of ES (know and understand): Slide 47
ES when L is normally distributed (know the formula and be able to apply
it): Slide 49-50
Delta-Normal ES (be able to derive the formula and apply it): Slides 51-54

Quantitative Finance, Lecture 8 56 / 56