ACTL3301/5301 Quantitative Risk Management

Week 1b: Risk Measures

▶ Risk Measures
▶ Value at Risk
▶ Variance
▶ Expected Shortfall
▶ Coherent Risk Measures
▶ Convex Risk Measures

Version 2024. Copyright UNSW School of Risk and Actuarial Studies

Reading: McNeil et al. (2015), Section 2.3
1 / 24
Risk Measures

▶ Consider a risk (random loss) X.

▶ We are often interested to evaluate the riskiness of X.
▶ Typically this involves determining ρ(X), the risk measure of X.
▶ This can be viewed as a representation of the riskiness of the
position X.
▶ Depending on the context, X can be one transaction, one business
unit, or even representative of the corporation as a whole.

2 / 24
Applications of Risk Measures

▶ In practice, risk measures are used for a number of problems

including in particular
▶ Determination of Economic (Risk) Capital
▶ As a Management Tool
▶ Insurance Premium setting.

3 / 24
Examples of Risk Measures

▶ Value at Risk
▶ Variance
▶ Moment based risk measures
▶ Expected shortfall

4 / 24
Value at Risk

▶ Let FL (l) = P (L ≤ l) be the distribution function of the loss on

a portfolio of risky assets
▶ VaR is defined as follows: Given a confidence level α ∈ (0, 1)
the VaR of the portfolio at confidence level α is the smallest
number l such that with probability at least α the loss L does not
exceed l:

VaRα (L) = inf {l : P (L ≤ l) ≥ α}

= inf {l : P (L > l) ≤ 1 − α}

▶ It is a quantile of the loss distribution

▶ Typical values of α: 0.95, 0.99, 0.995

5 / 24
6 / 24
Value at Risk: Normal distribution
▶ For normal distribution of FL (l) with mean µ and variance σ 2
we have
VaRα (L) = µ + σΦ−1 (α) ,
where Φ is the standard normal distribution function and
Φ−1 (α) is the α quantile of Φ.

7 / 24
Value at Risk: Some Potential Issues

▶ VaR ignores model risk - you may be using the wrong loss model
(and parameters are estimated from data).
▶ VaR ignores liquidity which impacts on time taken to trade
securities.
▶ Horizon for VaR need to reflect the time period the portfolio is
expected to be held
▶ VaR is not sub-additive - the VaR of a portfolio could be greater
than the sum of the VaR’s of the individual risks (to be discussed
in the next lecture).

8 / 24
Variance
For a risk L with mean µ and finite second moment,
var (L) = E (L − µ)2 .

▶ commonly used in finance and portfolio theory

▶ need to assume second moments exist
▶ usually only useful for symmetric risks

9 / 24
Expected Shortfall

▶ Expected shortfall (ES) at level α is defined as1

Z 1
1
ESα = qu (FL ) du
1−α α

where qu (FL ) is the quantile function of FL

▶ Related to VaR:
Z 1
1
ESα = VaRu (L) du
1−α α

▶ The average over VaRu for all u ≥ α

1
Note that different authors use different names with subtle differences.
10 / 24
Exercise
Show that the expected shortfall at level α is always greater than or
equal to the value at risk at the same level α:
ESα ≥ VaRα

11 / 24
Expected Shortfall
▶ For a continuous integrable loss L with continuous df FL ,

E (L; L ≥ qα (L))
ESα = = E (L|L ≥ VaRα ) , α ∈ (0, 1).
1−α

▶ More complicated for discontinuous loss distribution

E (L; L > qα (L)) + qα (L) × (P (L ≤ qα ) − α)

ESα =
1−α

12 / 24
If FL is normal with mean µ and variance σ 2 then

ϕ Φ−1 (α)


ESα = µ + σ
1−α
where ϕ is the density of the standard normal distribution.

13 / 24
Desired Properties of a Risk Measure

▶ One way to determine whether a risk measure is “good” is to

look at its properties.

▶ A risk measure is called a coherent risk measure if it satisfies the

following four properties.
▶ Monotonicity
▶ Translation Invariance
▶ Subadditivity
▶ Positive Homogeneity

14 / 24
Coherence: Monotonicity
Axiom (monotonicity). For two risks L1 and L2 such that L1 ≤ L2
almost surely, we have ρ(L1 ) ≤ ρ(L2 ).

15 / 24
Coherence: Translation Invariance
Axiom (translation invariance). For all risks L and every l ∈ R we
have ρ(L + l) = ρ(L) + l.

16 / 24
Coherence: Subadditivity
Axiom (subadditivity). For all risks L1 and L2 we have
ρ(L1 + L2 ) ≤ ρ(L1 ) + ρ(L2 ).

17 / 24
Coherence: Positive Homogeneity
Axiom (positive homogeneity). For all risks L and every λ > 0 we
have ρ(λL) = λρ(L).

18 / 24
Non-Subadditivity of VaR
Example (non-subadditivity of VaR). Consider a portfolio of two
zero-coupon bonds with a maturity of one year that default
independently. Assume:
▶ Each bond has a default probability p = 0.9%;
▶ The current price of the bonds and the face value of the bonds is
equal to 100;
▶ The bonds pay an interest rate of 5%.
Denote by Li the loss incurred by holding one unit of bond i, and
write L = L1 + L2 .

Show: VaR0.99 (L) > VaR0.99 (L1 ) + VaR0.99 (L2 ).

19 / 24
20 / 24
21 / 24
Coherence of ES
Example 2.26 (coherence of expected shortfall). Translation
invariance, monotonicity and positive homogeneity are immediate
from the corresponding properties of the quantile.

A general proof of subadditivity is given in Theorem 8.14 of the

textbook.

22 / 24
Convex Risk Measure

Axiom (Convexity). For all risks L1 and L2 and all λ ∈ [0, 1], we have

ρ (λL1 + (1 − λ)L2 ) ≤ λρ(L1 ) + (1 − λ)ρ(L2 ).

Definition (Convex risk measure). A risk measure is called convex if

it satisfies the following properties:
▶ monotonicity
▶ translation invariance
▶ convexity

23 / 24
Exercise
Show that a coherent risk measure is a convex risk measure.

24 / 24