Worst-Case Value-at-Risk of Non-Linear Portfolios

Worst-Case Value-at-Risk of Non-Linear
Portfolios
Steve Zymler Daniel Kuhn Ber Rustem
Department of Computing
Imperial College London
Zymler, Kuhn and Rustem Worst-Case Value-at-Risk of Non-Linear Portfolios
Portfolio Optimization
Consider a market consisting of m assets.

Optimal Asset Allocation Problem
Choose the weights vector w R
m
to make the portfolio return
high, whilst keeping the associated risk (w) low.
Portfolio optimization problem:

minimize
wR
m
(w)
subject to w W.
Popular risk measures :
Variance Markowitz model
Value-at-Risk Focus of this talk

Value-at-Risk: Denition
Let
r denote the random returns of the m assets.
The portfolio return is therefore w

T
r.
Value-at-Risk (VaR)
The minimal level R such that the probability of w
T
r
exceeding is smaller than .
VaR
(w) = min
_
: P
_
w
T
r
_

_
Theoretical and Practical Problems of VaR
VaR lacks some desirable theoretical properties:
Not a coherent risk measure.
Needs precise knowledge of the distribution of
r.
Non-convex function of w
VaR minimization intractable.
To optimize VaR: resort to VaR approximations.
Example: assume
r N(
r
,
r
), then
VaR
(w) =
T
r
w
1
()
_
w
T
r
w,
Normality assumption unrealistic

may underestimate the actual VaR.
Worst-Case Value-at-Risk
Only know means

r
and covariance matrix
r
0 of
r.
Let P
r
be the set of all distributions of
r with mean
r
and
covariance matrix
r
.
Worst-Case Value-at-Risk (WCVaR)
WCVaR
(w) = min
_
: sup
PP
r
P
_
w
T
r
_

_
WCVaR is immunized against uncertainty in P:

distributionally robust.
Unless the most pessimistic distribution in P

r
is the true
distribution, actual VaR will be lower than WCVaR.
Robust Optimization Perspective on WCVaR
El Ghaoui et al. have shown that

WCVaR
(w) =
T
w + ()
w
T
w,
where () =
_
(1 )/.
Connection to robust optimization:

WCVaR
(w) = max
rU
w
T
r,
where the ellipsoidal uncertainty set U
is dened as
U
=
_
r : (r
r
)
T
1
r
(r
r
) ()
2
_
.
Therefore,
min
wW
WCVaR
(w) min
wW
max
rU
w
T
r.
Worst-Case VaR for Derivative Portfolios
Assume that the market consists of:
n m basic assets with returns

, and
mn derivatives with returns .

are only risk factors.
We partition asset returns as
r = (
, ).
Derivative returns are uniquely determined by basic

asset returns

. There exists f : R
n
R
m
with
r = f (
).
f is highly non-linear and can be inferred from:
Contractual specications (option payoffs)
Derivative pricing models

Worst-Case VaR for Derivative Portfolios
WCVaR is applicable but not suitable for portfolios

containing derivatives:
Moments of are difcult to estimate accurately.
Disregards perfect dependencies between and

.
WCVaR severly overestimates the actual VaR, because:
r
only accounts for linear dependencies
is symmetric but derivative returns are skewed

Generalized Worst-Case VaR Framework
We develop two new Worst-Case VaR models that:
Use rst- and second-order moments of

but not .
Incorporate the non-linear dependencies f

Generalized Worst-Case VaR
Let P denote set of all distributions of

with mean and
covariance matrix .
min
_
: sup
PP
P
_
w
T
f (
)
_

_
When f (
) is:
convex polyhedral Worst-Case Polyhedral VaR (SOCP)
nonconvex quadratic Worst-Case Quadratic VaR (SDP)

Piecewise Linear Portfolio Model
Assume that the mn derivatives are European put/call options

maturing at the end of the investment horizon T.
Basic asset returns:

r
j
= f
j
(
) =

j
for j = 1, . . . , n.
Assume option j is a call with strike k

j
and premium c
j
on basic
asset i with initial price s
i
, then

r
j
is
f
j
(
) =
1
c
j
max
_
0, s
i
(1 +

i
) k
j
_
1
= max
_
1, a
j
+ b
j

i
1
_
, where a
j
=
s
i
k
j
c
j
, b
j
=
s
i
c
j
.
Likewise, if option j is a put with premium p

j
, then

r
j
is
f
j
(
) = max
_
1, a
j
+ b
j

i
1
_
, where a
j
=
k
j
s
i
p
j
, b
j
=
s
i
p
j
.
Piecewise Linear Portfolio Model
In compact notation, we can write
r as
r = f (
) =
_

max
_
e, a +B
e
_
_
.
Partition weights vector as w = (w
, w
).
No derivative short-sales: w W = w
0.
Portfolio return of w W can be expressed as

w
T
r = w
T
f (
)
= (w
)
T
+ (w
)
T
max
_
e, a +B
e
_
.
Worst-Case Polyhedral VaR
Use the piecewise linear portfolio model:

w
T
f (
) = (w
)
T
+ (w
)
T
max
_
e, a +B
e
_
.
Worst-Case Polyhedral VaR (WCPVaR)
For any w W, we dene WCPVaR
(w) as
WCPVaR
(w) = min
_
: sup
PP
P
_
w
T
f (
)
_

_
.
Worst-Case Polyhedral VaR: Convex Reformulations
Theorem: SDP Reformulation of WCPVaR
WCPVaR of w can be computed as an SDP:
WCPVaR
(w) =min
s. t. M S
n+1
, y R
mn
, R, R
, M , M 0, 0, 0 y w
M+
0 w
+B
T
y
(w
+B
T
y)
T
+2( + y
T
a e
T
w
0
Where we use the second-order moment matrix :
=
+
T
T
1

Worst-Case Polyhedral VaR: Convex Reformulations
Theorem: SOCP Reformulation of WCPVaR
WCPVaR of w can be computed as an SOCP:
WCPVaR
(w) = min
0gw
T
(w
+B
T
g) + ()
1/2
(w
+B
T
g)
2
. . .
. . . a
T
g + e
T
w
SOCP has better scalability properties than SDP.

Robust Optimization Perspective on WCPVaR
WCPVaR minimization is equivalent to:

min
wW
max
rU
p
w
T
r.
where the uncertainty set U
p
R
m
is dened as
U
p
=
_
_
_
r R
m
:
R
n
such that
( )
T
1
( ) ()
2
and
r = f ()
_
_
_
Unlike U
, the set U
p
is not symmetric!
Robust Optimization Perspective on WCPVaR
Example: WCPVaR vs WCVaR
Consider Black-Scholes Economy containing:
Stocks A and B, a call on stock A, and a put on stock B.
Stocks have drifts of 12% and 8%, and volatilities of 30%

and 20%, with instantaneous correlation of 20%.
Stocks are both $100.
Options mature in 21 days and have strike prices $100.
Assume we hold equally weighted portfolio.
Goal: calculate VaR of portfolio in 21 days.
Generate 5,000,000 end-of-period stock and option prices.
Calculate rst- and second-order moments from returns.
Estimate VaR using: Monte-Carlo VaR, WCVaR, and

WCPVaR.
Example: WCPVaR vs WCVaR
0
2
4
6
8
10
12
14
16
-4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1
p
r
o
b
a
b
i
l
i
t
y

(
%
)
portfolio loss
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
80 82 84 86 88 90 92 94 96 98 100
V
a
R
Confidence level (15)%
MonteCarlo VaR
WorstCase VaR
WorstCase Polyhedral VaR
At condence level = 1%:
WCVaR unrealistically high: 497%.
WCVaR is 7 times larger than WCPVaR.
WCPVaR is much closer to actual VaR.

Delta-Gamma Portfolio Model
mn derivatives can be exotic with arbitrary maturity time.

Value of asset i = 1 . . . m is representable as v
i
(
, t ).
For short horizon time T, second-order Taylor expansion is

accurate approximation of

r
i
:
r
i
= f
i
(
)
i
+
T
i

+
1
2
i = 1, . . . , m.
Portfolio return approximated by (possibly non-convex):

w
T
r = w
T
f () (w) +(w)
T
+
1
2
T
(w)
,
where we use the auxiliary functions
(w) =
m
i =1
w
i
i
, (w) =
m
i =1
w
i
i
, (w) =
m
i =1
w
i
i
.
We now allow short-sales of options in w

Worst-Case Quadratic VaR
Worst-Case Quadratic VaR (WCQVaR)
For any w W, we dene WCQVaR as
min
_
: sup
PP
P
_
(w) (w)
T

1
2
T
(w)
_

_
Theorem: SDP Reformulation of WCQVaR
WCQVaR can be found by solving an SDP:
WCQVaR
(w) =min
s. t. M S
n+1
, R, R
, M , M 0, 0,
M+
_
(w) (w)
(w)
T
+2( + (w))
_
0
There seems to be no SOCP reformulation of WCQVaR.

Robust Optimization Perspect on WCQVaR
WCQVaR minimization is equivalent to:

min
wW
max
ZU
q
Q(w), Z
where
Q(w) =
_
1
2
(w)
1
2
(w)
1
2
(w)
T
(w)
_
,
and the uncertainty set U
q
S
n+1
is dened as
U
q
=
_
Z =
_
X
T
1
_
S
n+1
: Z 0, Z 0
_
U
q
is lifted into S
n+1
to compensate for non-convexity.
Robust Optimization Perspect on WCQVaR
There is a connection between U
R
m
and U
q
S
n+1
.
If we impose: w W = (w) 0 then robust

optimization problem reduces to:
min
wW
max
rU
q
w
T
r
where the uncertainty set U
q
R
m
is dened as
U
q
=
_
_
_
r R
m
:
R
n
such that
( )
T
1
( ) ()
2
and
r
i
=
i
+
T
i
+
1
2
i
i = 1, . . . , m
_
_
_
Unlike U
, the set U
q
is not symmetric!
Robust Optimization Perspective on WCQVaR
Example: WCQVaR vs WCVaR
Now we want to estimate VaR after 2 days (not 21 days).
VaR not evaluated at option maturity times

use WCQVaR (not WCPVaR).
Use Black-Scholes to calculate prices and greeks.

0
1
2
3
4
5
6
7
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
p
r
o
b
a
b
i
l
i
t
y

(
%
)
portfolio loss
0
0.2
0.4
0.6
0.8
1
1.2
1.4
80 82 84 86 88 90 92 94 96 98 100
V
a
R
Confidence level (1-)%
Monte-Carlo VaR
Worst-Case VaR
Worst-Case Quadratic VaR
At = 1%: WCVaR still 3 times larger than WCQVaR.

Index Tracking using Worst-Case Quadratic VaR
Total test period: Jan. 2nd, 2004 Oct. 10th, 2008.
Estimation Window: 600 days. Out-of-sample returns: 581.

0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
0 100 200 300 400 500 600
R
e
l
a
t
i
v
e

W
e
a
l
t
h
Period
robust strategy with options
robust strategy without options
S&P 500 (benchmark)
Outperformance: option strat 56%, stock-only strat 12%.
Sharpe Ratio: option strat 0.97, stock-only strat 0.13.
Allocation option strategy: 89% stocks, 11% options.

Questions?
Paper available on optimization-online.

c 2007 Salvador Dal, Gala-Salvador Dal Foundation/Artists Rights Society (ARS), New York

Worst-Case Value-at-Risk of Non-Linear Portfolios

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Worst-Case Value-at-Risk of Non-Linear Portfolios

Încărcat de

Drepturi de autor:

Formate disponibile

Worst-Case Value-at-Risk of Non-Linear

Consider a market consisting of m assets.

Portfolio optimization problem:

Popular risk measures :

Variance Markowitz model

Value-at-Risk Focus of this talk

r denote the random returns of the m assets.

The portfolio return is therefore w

VaR lacks some desirable theoretical properties:

Not a coherent risk measure.

Needs precise knowledge of the distribution of

To optimize VaR: resort to VaR approximations.

Normality assumption unrealistic

Only know means

WCVaR is immunized against uncertainty in P:

Unless the most pessimistic distribution in P

El Ghaoui et al. have shown that

Connection to robust optimization:

Assume that the market consists of:

n m basic assets with returns

mn derivatives with returns .

Derivative returns are uniquely determined by basic

f is highly non-linear and can be inferred from:

Contractual specications (option payoffs)

Derivative pricing models

WCVaR is applicable but not suitable for portfolios

Moments of are difcult to estimate accurately.

Disregards perfect dependencies between and

WCVaR severly overestimates the actual VaR, because:

is symmetric but derivative returns are skewed

We develop two new Worst-Case VaR models that:

Use rst- and second-order moments of

Incorporate the non-linear dependencies f

convex polyhedral Worst-Case Polyhedral VaR (SOCP)

nonconvex quadratic Worst-Case Quadratic VaR (SDP)

Assume that the mn derivatives are European put/call options

Basic asset returns:

Assume option j is a call with strike k

Likewise, if option j is a put with premium p

In compact notation, we can write

Partition weights vector as w = (w

Portfolio return of w W can be expressed as

Use the piecewise linear portfolio model:

Zymler, Kuhn and Rustem Worst-Case Value-at-Risk of Non-Linear Portfolios

SOCP has better scalability properties than SDP.

WCPVaR minimization is equivalent to:

Consider Black-Scholes Economy containing:

Stocks A and B, a call on stock A, and a put on stock B.

Stocks have drifts of 12% and 8%, and volatilities of 30%

Stocks are both $100.

Options mature in 21 days and have strike prices $100.

Assume we hold equally weighted portfolio.

Goal: calculate VaR of portfolio in 21 days.

Generate 5,000,000 end-of-period stock and option prices.

Calculate rst- and second-order moments from returns.

Estimate VaR using: Monte-Carlo VaR, WCVaR, and

At condence level = 1%:

WCVaR unrealistically high: 497%.

WCVaR is 7 times larger than WCPVaR.

WCPVaR is much closer to actual VaR.

mn derivatives can be exotic with arbitrary maturity time.

For short horizon time T, second-order Taylor expansion is

Portfolio return approximated by (possibly non-convex):

We now allow short-sales of options in w