-
Chapter 4Expected Shortfall
Expected Shortfall (ES) is a risk measure computed by averating
potentiallosses above a certain level given by the Value at Risk
(VaR). It can beshown that the Expected Shortfall at the confidence
level p coincides withthe Tail Value at Risk (TVaR) defined as the
average of losses suffered inthe worst (1− p)% of events. This
chapter presents the concept of coherentrisk measure, including
Expected Shortfall and Tail Value at Risk (TVaR),together with
experiments based on financial data sets.
4.1 Tail Value at Risk (TVaR) . . . . . . . . . . . . . . . . .
. . . . . 774.2 Conditional Tail Expectation (CTE) . . . . . . . .
. . . . . 784.3 Expected Shortfall (ES) . . . . . . . . . . . . . .
. . . . . . . . . . 814.4 Market Data vs Gaussian Risk Measures . .
. . . . . . . 88Exercises . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 97
4.1 Tail Value at Risk (TVaR)
A natural shortcoming of Value at Risk is failing to provide
information onthe behavior of probability distribution tails beyond
V pX . The next figureillustrates the limitations of Value at Risk,
namely its inability to capturethe properties of a probability
distribution beyond V pX .†
† “Value at Risk is like an airbag that works all the time,
except when you have a caraccident”. - D. Einhorn, hedge fund
manager.
" 77
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
Fig. 4.1: Two distributions having the same Value at Risk V 95%X
= 2.145.
The Tail Value at Risk (or Conditional Value at Risk) aims at
providing asolution to the tail distribution problem observed with
Value at Risk at thelevel p, by averaging over confidence levels
ranging from p to 1.Definition 4.1. The Tail Value at Risk of a
random variable X at the levelp ∈ (0, 1) is defined by the
average
TVpX :=1
1− pw 1pV qXdq. (4.1)
Note that since the function p 7−→ V pX is non-decreasing, we
always have
TVpX =1
1− pw 1pV qXdq >
11− p
w 1pV pXdq = V
pX .
4.2 Conditional Tail Expectation (CTE)
Recall that by Lemma 10.14, given an event A such that P(A) >
0, theconditional expectation of X : Ω −→N given the event A
satisfies
E[X | A] = 1P(A)
E [X1A] ,
see Section 3.2 for an example.Definition 4.2. Consider a random
variable X such that P
(X > V pX
)> 0.
The Conditional Tail Expectation of X at the level p ∈ (0, 1) is
the quantity
CTEpX := E[X | X > V pX
]=
1P(X > V pX
)E[X1{X>V pX}].
The use of the strict inequality “>” in the definition of the
Conditional TailExpectation allows us to avoid any dependence on
P(X = V pX ), and to con-sider risky values strictly beyond V pX .
The Conditional Tail Expectation is
78 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
also called Conditional Value at Risk (CVaR).
Examples of Conditional Tail Expectations can be computed as in
the fol-lowing R code.
1 library(quantmod)2
getSymbols("^HSI",from="2013-06-01",to="2014-10-01",src="yahoo")
returns
V pX with, more precisely,
CTEpX = E[X | X > V pX
]= V pX + E
[(X − V pX
)+ | X > V pX].Proof. We have
E[X | X > V pX
]=
1P(X > V pX
)E[X1{X>V pX}]
=1
P(X > V pX
)(E[(X − V pX)1{X>V pX}]+ V pXE[1{X>V pX}])=
1P(X > V pX
)(E[(X − V pX)+]+ V pXP(X > V pX))= V pX +
1P(X > V pX
)E[(X − V pX)+]= V pX + E
[(X − V pX
)+ ∣∣X > V pX].See Exercise 4.2-(d) for a proof of CTEpX >
E[X ]. �
Next, we check that when P(X = V pX
)= 0, the Conditional Tail Expectation
coincides with the Tail Value at Risk. Note that in this case we
have P(X >
V pX)> 0 by Proposition 3.9-(b).
Proposition 4.4. Assume that P(X = V pX
)= 0. Then we have CTEpX =
TVpX , i.e.
CTEpX = E[X | X > V pX
]= E
[X | X > V pX
]=
11− p
w 1pV qXdq = TV
pX .(4.2)
Proof. By Lemma 3.13 we construct X as X = V UX where U is
uniformlydistributed on [0, 1], with
" 79
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
U > p =⇒ V UX > VpX =⇒ X > V
pX ,
andX > V pX =⇒ V
UX > V
pX =⇒ U > p.
Since P(X = V pX
)= 0 we find that, with probability 1,
U > p⇐⇒ U > p⇐⇒ V UX > VpX ⇐⇒ X > V
pX ⇐⇒ X > V
pX ,
hence
CTEpX = E[X | X > V pX
]= E
[V UX | V UX > V
pX
]= E
[V UX | U > p
]=
1P(U > p)
E[V UX 1{U>p}
]=
11− p
w 1pV qXdq.
�
The next figure shows the location of Value at Risk and
Conditional TailExpectation on a data set. Note that the sign of
the data has been changedaccording to Proposition 3.10.
Fig. 4.2: Value at Risk and Conditional Tail Expectation.
The Conditional Tail Expectation of a Gaussian N (µ,σ2) random
variableis computed in the next proposition.
Proposition 4.5. Gaussian CTE. Given X ' N (µ,σ2) we have
CTEpX = µ+σ
(1− p)φ(VpZ ) = µ+
σ
(1− p)√
2πe−(V
pZ)2/2, (4.3)
80 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
where V pZ is the Value at Risk of Z ' N (0, 1) at the level p
and
φ(z) =1√2π
e−z2/2, x ∈ R,
is the standard normal probability density function.
Proof. Using the relation P(X > V pX
)= P
(X > V pX
)= 1− p, cf. Proposi-
tion 3.9, we have
CTEpX = TVpX
= E[X | X > V pX
]=
1P(X > V pX
)E[X1{X>V pX}]
=1
1− pw∞V pX
xe−(x−µ)2/(2σ2) dx√2πσ2
=µ
1− pw∞V pX
e−(x−µ)2/(2σ2) dx√2πσ2
+1
1− pw∞V pX
(x− µ)e−(x−µ)2/(2σ2) dx√2πσ2
=µ
1− pP(X > VpX ) +
σ2
(1− p)√
2πσ2[−e−(x−µ)2/(2σ2)
]∞V pX
= µ+σ2
(1− p)√
2πσ2e−((V
pX−µ)/σ)2/2
= µ+σ
(1− p)√
2πe−(V
pZ)2/2
= µ+σ
(1− p)φ(VpZ ),
due to the rescaling relation V pX = µ+ σqpZ , cf. (3.7). �
4.3 Expected Shortfall (ES)
There are several variants for the definition of the Expected
Shortfall ESpX .Next is a frequently used definition.
Definition 4.6. The Expected Shortfall ESpX of a random variable
X at thelevel p ∈ (0, 1) is defined by
ESpX := VpX +
11− pE
[(X − V pX
)1{X>V p
X}]. (4.4)
We also have
ESpX = VpX +
P(X > V pX
)1− p E
[X − V pX
∣∣X > V pX]" 81
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
= V pX +1
1− p(E[X1{X>V p
X}]− V pXE
[1{X>V p
X}])
= V pX +1
1− p(E[X1{X>V p
X}]− V pXP
(X > V pX
))=
11− pE
[X1{X>V p
X}]+
V pX1− p
(1− p−P
(X > V pX
)),
and
ESpX =
1
1− pE[X1{X>V p
X}]= E
[X | X > V pX
]= TVpX if P
(X = V pX
)= 0,
11− pE
[X1{X>V p
X}]+
V pX1− p
(1− p−P
(X > V pX
))if P
(X = V pX
)> 0,
as shown in the next proposition. Note that by Proposition
3.9-(b) we haveP(X > V pX
)> 0 when P
(X = V pX
)= 0.
Proposition 4.7. When P(X = V pX
)= 0 the Expected Shortfall ESpX co-
incides with the Conditional Tail Expectation CTEpX and the Tail
Value atRisk TVpX , i.e., we have
ESpX = E[X | X > V pX
]= E
[X | X > V pX
]= TVpX .
Proof. By Proposition 3.9, when P(X = V pX
)= 0 we have
p = P(X 6 V pX ) and 1− p = P(X > VpX ) = P(X > V
pX ),
hence
ESpX =1
1− pE[X1{X>V p
X}]
=1
1− pE[X1{X>V p
X}]
=1
P(X > V pX
)E[X1{X>V pX}]
= E[X | X > V pX
]= TVpX ,
by Proposition 4.4. �
When P(X = V pX
)= 0, we also have
ESpX =1
1− pE[X1{X>V p
X}]+ V pX −
V pX1− pP
(X > V pX
)=
11− pE
[X1{X>V p
X}]+
V pX1− p
(1− p−P
(X > V pX
)).
82 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
Proposition 4.8. The Expected Shortfall ESpX can be written as
the dis-torted risk measure
ESpX =1
1− pE[XfX (X)] =w 1
0V qXfX (V
qX )dq, (4.5)
where fX is the distortion function defined by
fX (x) :=1
1− p1{x>VpX} + 1{P(X=V p
X)>0}
1− p−P(X > V pX
)(1− p)P
(X = V pX
) 1{x=V pX},
x ∈ R.
Proof. We have
ESpX =1
1− pE[X1{X>V p
X}]+
V pX1− p
(1− p−P
(X > V pX
))=
11− pE
[X1{X>V p
X}]+ 1{P(X=V p
X)>0}
V pX1− p
(1− p−P
(X > V pX
))=
11− pE
[X
(1{X>V p
X} + 1{P(X=V p
X)>0}
1− p−P(X > V pX
)P(X = V pX
) 1{X=V pX}
)].
�
Note that the distortion function fX is a non-decreasing
function that satisfies
E[fX (X)] =1
1− pE[1{X>V p
X} + 1{P(X=V p
X)>0}
1− p−P(X > V pX
)P(X = V pX
) 1{X=V pX}
]
=1
1− p
(E[1{X>V p
X}]+ 1− p−P
(X > V pX
))=
11− p
(P(X > V pX
)+ 1− p−P
(X > V pX
))= 1, x ∈ R. (4.6)
The following proposition, see Acerbi and Tasche (2001), shows
that in gen-eral, the Expected Shortfall at the level p ∈ (0, 1)
coincides with the TailValue at Risk TVpX .
Proposition 4.9. The Expected Shortfall ESpX coincides with the
Tail Valueat Risk TVpX for any p ∈ (0, 1), i.e. we have
ESpX = TVpX =
11− p
w 1pV qXdq.
Proof. Constructing X as X = V UX where U is uniformly
distributed on [0, 1]as in Lemma 3.13, by Proposition 3.8 we
have
" 83
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
U > p =⇒ V UX > VpX =⇒ X > V
pX
and (U < p and X > V pX
)=⇒
(V UX 6 V
pX and X > V
pX
)=⇒
(X 6 V pX and X > V
pX
)=⇒ X = V pX .
Hence by (4.4) and the relations
1− p = E[1{U>p}
]and P
(X > V pX
)= E
[1{X>V p
X}],
we have
V pX(1− p−P
(X > V pX
))= −V pXE
[1{X>V p
X} − 1{U>p}
]= −V pXE
[1{X>V p
X}\{U>p}
]= −V pXE
[1{X>V p
X}∩{UV p
X}∩{UV p
X}]+
V pX1− p
(1− p−P
(X > V pX
))=
11− pE
[X1{X>V p
X}]− 11− pE
[X1{X>V p
X}∩{UV p
X}]− 11− pE
[V UX 1{V U
X>V p
X}∩{UV p
X}∩{U>p}
]=
11− pE
[V UX 1{U>p}
]=
11− p
w 1pV qXdq,
which is the Tail Value at Risk TVpX . �
As a consequence of Propositions 4.4-4.5 and Proposition 4.9,
the GaussianExpected Shortfall at the level p is also given by
ESpX = µ+σ
(1− p)φ(VpZ ).
84 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
Proposition 4.10. The Expected Shortfall ESpX and the Tail Value
at RiskTVpX are coherent risk measures.
Proof. As ESpX coincides with TVpX for all p ∈ (0, 1) from
Proposition 4.9,
we can use either Relation (4.4) in Definition 4.6 or Relation
(4.1) in Defini-tion 4.1.
(i) Monotonicity. If X 6 Y , since Value at Risk is monotone we
have
ESpX = TVpX
=1
1− pw 1pV qXdq
61
1− pw 1pV qY dq
= TVpY6 ESpY
for all p ∈ (0, 1).
(ii) Homogeneity and translation invariance. Similarly, since
Value at Risk issatisfies the homogeneity and translation
invariance properties, for all µ ∈ Rand λ > 0 we have monotone
we have
ESpµ+λX = TVpµ+λX
=1
1− pw 1pV qµ+λXdq
=1
1− pw 1p
(µ+ λV qX
)dq
= µ+ λ1
1− pw 1pV qXdq
= µ+ λTVpY6 µ+ λESpY
for all p ∈ (0, 1).
(iii) Sub-additivity. We have
(1− p)(ESpX+Y −ES
pX −ES
pY
)= E[(X + Y )fX+Y (X + Y )]−E[XfX (X)]−E[Y fY (Y )]= E[X(fX+Y (X
+ Y )− fX (X))]−E[Y (fX+Y (X + Y )− fY (Y ))]> V pXE[fX+Y (X + Y
)− fX (X)]− V
pY E[fX+Y (X + Y )− fY (Y )]
= V pX (1− p− (1− p))− VpY (1− p− (1− p))
= 0,
where we have used (4.6) and the facts that, for x < V pX
,
" 85
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
(1− p)(fX+Y (x+ y)− fX (x)) = 1{x+y>V pX+Y }
− 1{x>V pX}
+1{P(X+Y =V pX+Y )>0}
1− p−P(X + Y > V pX+Y )P(X + Y = V pX+Y )
1{x+y=V pX+Y }
−1{P(X=V pX)>0}
1− p−P(X > V pX
)P(X = V pX
) 1{x=V pX}
= 1{x+y>V pX+Y }
+ 1{P(X+Y =V pX+Y )>0}
1− p−P(X + Y > V pX+Y )P(X + Y = V pX+Y )
1{x+y=V pX+Y }
> 0, x < V pX ,
and, for x > V pX ,
(1− p)(fX+Y (x+ y)− fX (x)) = 1{x+y>V pX+Y }
− 1{x>V pX}
+1{P(X+Y =V pX+Y )>0}
1− p−P(X + Y > V pX+Y )P(X + Y = V pX+Y )
1{x+y=V pX+Y }
−1{P(X=V pX)>0}
1− p−P(X > V pX
)P(X = V pX
) 1{x=V pX}
= 1{x+y>V pX+Y }
− 1{x>V pX}
+1{P(X+Y =V pX+Y )>0}
1− p−P(X + Y > V pX+Y )P(X + Y = V pX+Y )
1{x+y=V pX+Y }
6 1{x+y>V pX+Y }
− 1{x>V pX}
+1{x+y=V pX+Y }
= 1{x+y>V pX+Y }
− 1{x>V pX}
> 0, x < V pX .
�
Note that in general, the Conditional Tail Expectation is not a
coherent riskmeasure when P
(X = V pX
)> 0.
Performance analytics in R - Expected Shortfall (ES)
1 library(PerformanceAnalytics)2 ES(returns, p=.95,
method="historical")
The 95% Expected Shortfall is ES95%X = −0.02087832. The
historical Ex-pected Shortfall can be exactly recovered by the
empirical Condtional TailExpectation (CTE) as
86 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
1 mean(returns[returns source("comparison.R")Number of samples=
265VaR 95 = -0.03420879 , Threshold= 0.9433962CTE 95 =
-0.04646176ES 95 = -0.04623058Historical VaR 95 0=
-0.03316604Gaussian VaR 95 = -0.03209374Historical ES 95 =
-0.04552403Gaussian ES 95 = -0.04043227
Fig. 4.3: Value at Risk and Expected Shortfall.
" 87
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
library(quantmod)
getSymbols("0700.HK",from="2015-01-03",to="2016-02-01",src="yahoo")stock=Ad(`0700.HK`)stock.rtn=as.numeric((stock-lag(stock))/stock);stock.rtn
-
N. Privault
Value at Risk vs Expected Shortfall
1
chart.VaRSensitivity(ts(returns),methods=c("HistoricalVaR","HistoricalES"),colorset=bluefocus,
lwd=2)
0.89 0.9 0.91 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99
Confidence Level
−0.12
−0.10
−0.08
−0.06
−0.04
Historical VaRHistorical ES
Value at
Risk
Risk Confidence Sensitivity of 1800.HK.Adjusted
Fig. 4.4: Value at Risk vs Expected Shortfall.
4.4 Market Data vs Gaussian Risk Measures
Market returns vs Gaussian and power tails
Consider for example the market returns data obtained from
fetching DJI andSTI index data using the R package Quantmod and the
following scripts.1 library(quantmod)
getSymbols("^STI",from="1990-01-03",to="2015-01-03",src="yahoo");stock=Ad(`STI`);3
getSymbols("^DJI",from="1990-01-03",to=Sys.Date(),src="yahoo");stock=Ad(`DJI`);
stock.rtn=diff(log(stock));returns 3*s])/length(stock.rtn)13
length(y[abs(y-m)>3*s])/length(y);2*(1-pnorm(3*s,0,s))
Figure 4.5 shows the mismatch between the distributional
properties of mar-ket log-returns vs standardized Gaussian returns,
which tend to underesti-mate the probabilities of extreme events.
Note that when X ' N (0,σ2),
88 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
99.73% of samples of X are falling within the interval
[−3σ,+3σ], i.e.P(|X| 6 3σ) = 0.9973002.
1995 2000 2005 2010 2015
−0.05
0.00
0.05
Fig. 4.5: Market returns vs normalized Gaussian returns.
1 stock.ecdf=ecdf(as.vector(stock.rtn));x
-
N. Privault
1 dev.new(width=16,height=8)qqnorm(returns, col = "blue",
xaxs="i", yaxs="i", las=1, cex.lab=1.4, cex.axis=1, lwd=3)
3 grid(lwd = 2)
−3 −2 −1 0 1 2 3
−0.10
−0.05
0.00
0.05
0.10
Normal Q−Q Plot
Theoretical Quantiles
Sam
ple
Qua
ntile
s
Fig. 4.7: Quantile-Quantile plot.
1
ks.test(y,"pnorm",mean=m,sd=s)ks.test(returns,"pnorm",mean=m,sd=s)
The Kolmogorov-Smirnov test clearly rejects the null (normality)
hypothesis.
One-sample Kolmogorov-Smirnov test
data: returnsD = 0.075577, p-value < 2.2e-16alternative
hypothesis: two-sided
This mismatch can be further illustrated by the empirical
probability densityplot in Figure 4.8, which is obtained from the
following R code.
1 dev.new(width=16,height=8)2 x
-
Notes on Financial Risk and Analytics
−0.10 −0.05 0.00 0.05 0.100
10
20
30
40
50
60
x
Empirical densityGaussian density
Fig. 4.8: Empirical density vs normalized Gaussian density.
Power tail distributions
We note that the empirical density has significantly higher
kurtosis and nonzero skewness in comparison with the Gaussian
probability density. On theother hand, power tail probability
densities of the form ϕ(x) ' Cα/xα,x → ∞, can provide a better fit
of empirical probability density functions,as shown in Figure
4.9.
−0.10 −0.05 0.00 0.05 0.100
10
20
30
40
50
60Empirical densityPower density
Fig. 4.9: Empirical density vs power density.
The above fitting of empirical probability density function is
using a powerprobability density function defined by a rational
fraction obtained by thefollowing R script.
" 91
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
1 install.packages("pracma")2 library(pracma); x
-
Notes on Financial Risk and Analytics
Hn(x) :=(−1)nϕ(x)
∂nϕ
∂xn(x), x ∈ R,
denote the Hermite polynomial of degree n, with H0(x) = 1.
Given X a random variable, the sequence (κXn )n>1 of
cumulants of Xhas been introduced in Thiele (1899). In the sequel
we will use the MomentGenerating Function (MGF) of the random
variable X, defined as
MX (t) := E[etX]= 1 +
∑n>1
tn
n!E[Xn], t ∈ R. (4.7)
Definition 4.11. The cumulants of a random variable X are
defined to bethe coefficients (κXn )n>1 appearing in the series
expansion
log(E[etX])
= log
1 +∑n>1
tn
n!E[Xn]
=∑n>1
κXntn
n!, t ∈ R, (4.8)
of the logarithmic moment generating function (log-MGF) of
X.
The cumulants of X were originally called “semi-invariants” due
to the prop-erty κX+Yn = κXn + κYn , n > 1, when X and Y are
independent randomvariables. Indeed, in this case we have∑
n>1κX+Yn
tn
n!= log
(E[et(X+Y )
])= log
(E[etX]E[etY])
= log E[etX]+ log E
[etY]
=∑n>1
κXntn
n!+∑n>1
κYntn
n!
=∑n>1
(κXn + κ
Yn
) tnn!
, t ∈ R,
showing that κX+Yn = κXn + κYn , n > 1.
a) First moment and cumulant. Taking n = 1 and π = {1}, we find
κX1 =E[X ].
b) Variance and second cumulant. We have
κX2 = E[X2]− (E[X ])2 = E
[(X −E[X ])2
],
and√κX2 is the standard deviation of X.
c) The third cumulant of X is given as the third central
moment
" 93
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
κX3 = E[(X −E[X ])3],
and the coefficient
SkX :=κX3
(κX2 )3/2 =
E[(X −E[X ])3
](E[(X −E[X ])2])3/2
is the skewness of X.d) Similarly, we have
κX4 = E[(X −E[X ])4]− 3(κX2 )2
= E[(X −E[X ])4]− 3(E[(X −E[X ])2
])2,and the excess kurtosis of X is defined as
EKX :=κX4
(κX2 )2 =
E[(X −E[X ])4](E[(X −E[X ])2])2
− 3.
The next proposition summarizes the Gram-Charlier expansion
method toobtain series expansion of a probability density function,
see Gram (1883),Charlier (1914) and § 17.6 of Cramér (1946).
Proposition 4.12. (Proposition 2.1 in Tanaka et al. (2010)) The
Gram-Charlier expansion of the continuous probability density
function φX (x) of arandom variable X is given by
φX (x) =1√κX2
ϕ
x− κX1√κX2
+ 1√κX2
∞∑n=3
cnHn
x− κX1√κX2
ϕx− κX1√
κX2
,where c0 = 1, c1 = c2 = 0, and the sequence (cn)n>3 is given
from thecumulants (κXn )n>1 of X as
cn =1
(κX2 )n/2
[n/3]∑m=1
∑l1+···+lm=nl1,...,lm>3
κXl1 · · ·κXlm
m!l1! · · · lm!, n > 3.
The coefficients c3 and c4 can be expressed from the skweness
κX3 /(κX2 )3/2and the excess kurtosis κX4 /(κX2 )2 as
c3 =κX3
3!(κX2 )3/2and c4 =
κX44!(κX2 )2
.
a) The first-order expansion
94 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
φ(1)X (x) =
1√κX2
ϕ
x− κX1√κX2
corresponds to normal moment matching approximation.
b) The third-order expansion is given by
φ(3)X (x) =
1√κX2
1 + c3H3x− κX1√
κX2
ϕx− κX1√
κX2
.c) The fourth-order expansion is given by
φ(4)X (x) =
1√κX2
1 + c3H3x− κX1√
κX2
+ c4H4x− κX1√
κX2
ϕx− κX1√
κX2
.1
install.packages("SimMultiCorrData");install.packages("PDQutils")2
library(quantmod);library(SimMultiCorrData);library(PDQutils)
dev.new(width=16,height=8)4 plot(stock.dens$x,stock.dens$y, xlab
= 'x', type = 'l', lwd=4, col="red",ylab = '', main = '',
xlim =c(-0.1,0.1), ylim=c(0,65), xaxs="i", yaxs="i", las=1,
cex.lab=1.8, cex.axis=1.8)lines(x, qx, type="l", lty=2, lwd=4,
col="blue")
6 m
-
N. Privault
Historical and Gaussian risk measures
1
dev.new(width=16,height=8)chart.VaRSensitivity(stock.rtn[,1,drop=FALSE],methods=c("HistoricalVaR","GaussianVaR"),
colorset=bluefocus, lwd=4)
The next Figure 4.11 uses the above R code to compare the
historical andGaussian values at risk.
0.89 0.9 0.905 0.915 0.925 0.935 0.945 0.955 0.965 0.975
0.985
Confidence Level
−0.
030
−0.
025
−0.
020
−0.
015
−0.
010
Historical VaRGaussian VaR
Val
ue a
t Ris
k
Risk Confidence Sensitivity of DJI.Adjusted
Fig. 4.11: Historical vs Gaussian estimates of Value at
Risk.
1 dev.new(width=16,height=8)2
chart.VaRSensitivity(stock.rtn[,1,drop=FALSE],methods=c("HistoricalES","GaussianES"),
colorset=bluefocus, lwd=4)
In the next Figure 4.12 we compare the Gaussian and historical
estimates ofExpected Shortfall.
0.89 0.9 0.905 0.915 0.925 0.935 0.945 0.955 0.965 0.975
0.985
Confidence Level
−0.
045
−0.
040
−0.
035
−0.
030
−0.
025
−0.
020
Historical ESGaussian ES
Val
ue a
t Ris
k
Risk Confidence Sensitivity of DJI.Adjusted
Fig. 4.12: Quantile function.
96 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
In Table 4.1 we summarize some properties of risk measures.
Risk Measure Additivity Homogeneity Subadditivity Coherence
V pX X X 7 7
CTEpX X X 7 7
TVpX X X X X
ESpX X X X X
Table 4.1: Summary of Risk Measures.
Note that Value at Risk V pX is coherent on Gaussian random
variables ac-cording to Remark 3.12. Similarly, the Conditional
Tail Expectation CTEpXis coherent on random variables having a
continuous CDF by Propositions 4.4and 4.10.
Exercises
Exercise 4.1 LetX denote an exponentially distributed random
variable withparameter λ > 0, i.e. the distribution of X has the
cumulative distributionfunction (CDF)
FX (x) = P(X 6 x) = 1− e−λx, x > 0,
and the probability density function (PDF)
fX (x) = F′X (x) = λe−λx, x > 0.
" 97
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
0 1 2 4 5 6
qX0
.95
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
p=0.95
(a) Exponential quantile and PDF.
q0.95
X
0 1 2 3 4 5 60
0.10.20.30.40.50.60.70.80.91
Probabilitydensity
(b) Exponential PDF.
a) Compute the conditional tail expectation
E[X | X > VaRpX ] =1
P(X > VaRpX )
w∞VaRp
X
xfX (x)dx.
b) Compute the tail value at risk
TVpX =1
1− pw 1pV qXdq.
Exercise 4.2 Consider X an (integrable) random variable and z ∈
R suchthat P(X > z) > 0.a) Show that E[X | X > z] >
z.b) Show that E[X | X > z] > E[X ].c) Show that E[X | X >
z] > E[X ] if P(X 6 z) > 0.d) Show that CTEpX > E[X ].
Exercise 4.3 Consider the following data set.
Find the Value at Risk VaRpX and the Conditional Tail
Expectation CTEpX =
E[X | X > VaRpX
]and mark their values on the graph in the following cases.
a) p = 0.9.
98 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
b) p = 0.8.
Exercise 4.4
Let p = 0.9. For the above data set represented by the random
variable X,compute the numerical values of the following
quantities.
a) VaR90X ,b) E
[X1{X>V 90
X}],
c) P(X > V 90X
),
d) CTE90X = E[X | X > V 90X
]= E
[X1{X>V 90
X}]/P(X > V 90X
),
e) E[X1{X>V 90
X}],
f) P(X > V 90X
),
g) ES90X =1
1− p(E[X1{X>V 90
X}]+ V 90X
(1− p−P
(X > V 90X
))),
h) TV90X =1
1− pw 1pV qXdq,
and mark the values of VaR90%X , CTE90%X , ES
90%X , TV
90%X on the above
graph.
Exercise 4.5 Consider a random variable X ∈ {10, 100, 150} with
the distri-bution
P(X = 10) = 96%, P(X = 100) = 3%, P(X = 150) = 1%.
Compute
a) the Value at Risk V 98%X ,b) the Tail Value at Risk TV98%X
,c) the Conditional Tail Expectation E
[X | X > V 98%X
], and
d) the Expected Shortfall E98%X .
" 99
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
N. Privault
Exercise 4.6 Consider two independent random variables X and Y
withsame distribution given by
P(X = 0) = P(Y = 0) = 90% and P(X = 100) = P(Y = 100) = 10%.
a) Plot the cumulative distribution function of X on the
following graph:
0.88
0.90
0.92
0.94
0.96
0.98
1.00
1.02
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
190 200 210−10−20 x
FX(x)
0
Fig. 4.14: Cumulative distribution function of X.
b) Plot the cumulative distribution function ofX+Y on the
following graph:
0.800.820.840.860.880.900.920.940.960.981.00
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
190 200 210−10−20 x
FX+Y (x)
0
Fig. 4.15: Cumulative distribution function of X + Y .
c) Give the values at risk V 99%X+Y , V95%X+Y , V
90%X+Y .
d) Compute the Tail Value at Risk
TV90%X :=1
1− pw 1pV qXdq
at the level p = 90%.e) Compute the Tail Value at Risk
TVpX+Y :=1
1− pw 1pV qX+Y dq
at the levels p = 90% and p = 80%.
Exercise 4.7 (Exercise 3.2 continued).
a) Compute the Tail Value at Risk
TVpX :=1
1− pw 1pV qXdq
100 "
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
-
Notes on Financial Risk and Analytics
for all p in the interval [0.99, 1], and give the value of
TV99%X .b) Taking p = 0.98, compute the Conditional Tail
Expectation
CTE98%X = E[X | X > V 98%X
]=
1P(X > V pX
)E [X1{X>V pX}
].
" 101
This version: June 13,
2021https://personal.ntu.edu.sg/nprivault/indext.html
https://personal.ntu.edu.sg/nprivault/indext.html
pbs@ARFix@91: pbs@ARFix@92: pbs@ARFix@93: pbs@ARFix@94:
pbs@ARFix@95: pbs@ARFix@96: pbs@ARFix@97: pbs@ARFix@98:
pbs@ARFix@99: pbs@ARFix@100: pbs@ARFix@101: pbs@ARFix@102:
pbs@ARFix@103: pbs@ARFix@104: pbs@ARFix@105: pbs@ARFix@106:
pbs@ARFix@107: pbs@ARFix@108: pbs@ARFix@109: pbs@ARFix@110:
pbs@ARFix@111: pbs@ARFix@112: pbs@ARFix@113: pbs@ARFix@114:
pbs@ARFix@115: