SFB 649 Discussion Paper 2005-004

S

F

B






6

4

9














E

C

O

N

O

M

I

C





R

I

S

K

















B

E

R

L

I

N



Value-at-Risk
Calculations with Time
Varying Copulae

Enzo Giacomini*
Wolfgang Hrdle*






*CASE - Center for Applied Statistics and Economics, Humboldt-
Universitt zu Berlin, Germany

SFB 649, Spandauer Strae 1, D-10178 Berlin
http://sfb649.wiwi.hu-berlin.de

This research was supported by the Deutsche
Forschungsgemeinschaft through the SFB 649 Economic Risk.

Value-at-Risk Calculations with Time Varying Copulae
Enzo Giacomini, Wolfgang Hardle
Center for Applied Statistics and Economics
Institut f ur Statistik und

Okonometrie
Humboldt-Universitat zu Berlin
Spandauer Str. 1, D-10178 Berlin
Germany
E-mail: [email protected]
Value-at-Risk (VaR) of a portfolio is determined by the multivariate distribution of the
risk factors increments. This distribution can be modelled through copulae, where the copulae
parameters are not necessarily constant over time. For an exchange rate portfolio, copulae
with time varying parameters are estimated and the VaR simulated accordingly. Backtesting
underlines the improved performance of time varying copulae.
Value-at-Risk and Copulae
At time t a linear portfolio composed of d positions w = (w
1
, . . . , w
d
)

on assets with
prices S
t
= (S
1,t
, . . . , S
d,t
)

and log prices Z

t
= lnS
t
, has value
(1) V
t
=
d

j=1
w
j
e
Z
j,t
The prot and loss (P&L) function is dened as L
t+1
= (V
t+1
V
t
). Dening X
t+1
= (Z
t+1
Z
t
)
as the time increment in the risk factors from period t to t + 1, the P&L can be expressed as:
(2) L
t+1
=
d

j=1
w
j
S
j,t
(e
X
j,t+1
1)
The Value-at-Risk (V aR) is calculated as the -quantile from F
L
, the distribution of L:
(3) V aR = F
1
L
()
The 1-dimensional distribution F
L
depends on the d-dimensional distribution F
X
. Using copu-
lae, the marginal distributions F
X
j
from each univariate increment can be separately modelled
from their dependence structure and then coupled together to form the multivariate distribu-
tion F
X
.
In the following, the dependence parameter

and the joint distribution

F
X
from a sample
{X
t
}
T
t=1
of log returns from exchange rate positions are estimated with copulae. A Monte Carlo
simulation based on

F
X
generates dierent P&L samples. The quantiles at dierent levels from
these simulation samples are then used as estimators for the Value-at-Risk of dierem portfolio.
Computing Value-at-Risk with Copulae
A copula is a d-dimensional distribution function C : [0, 1]
d
[0, 1] with uniform mar-
ginals on the interval [0, 1]. As in Nelsen (1998), multivariate distributions can be modelled
via:
Theorem 1 (Sklars theorem) Let F be a d-dimensional distribution function with mar-
ginals F
X
1
. . . , F
X
d
. Then there exists a copula C with
(4) F(x
1
, . . . , x
d
) = C{F
X
1
(x
1
), . . . , F
X
d
(x
d
)}
for every x
1
, . . . , x
d
R. If F
X
1
, . . . , F
X
d
are continuous, then C is unique. On the other hand,
if C is a copula and F
X
1
, . . . , F
X
d
are distribution functions, then the function F dened in (4)
is a joint distribution function with marginals F
X
1
, . . . , F
X
d
.
The estimation of the Value-at-Risk, based on an i.i.d. sample {X
t
}
T
t=1
is implemented
in the following procedure:
1. specication of marginal distributions F
X
j
(x
j
)
2. specication of copula C(u
1
, . . . , u
d
; )
3. tting the copula C to obtain

4. generation of Monte Carlo data X

T+1
C(u
1
, . . . , u
d
;

)
5. generation of a sample of portfolio losses L
T+1
(X
T+1
)
6. estimation of

V aR
T+1
, the empirical quantile at level from L
T+1
(X
T+1
).
For copulae belonging to a parametric family C = {C

, } and univariate marginals

F
X
j
(x
j
;
j
), the density of X is given by:
f(x
1
, . . . , x
d
;
1
, . . . ,
d
, ) = c{F
X
1
(x
1
;
1
), . . . , F
X
d
(x
d
;
d
); }
d

j=1
f
j
(x
j
;
j
)
where
c(u
1
, . . . , u
d
) =

d
C(u
1
, . . . , u
d
)
u
1
. . .u
d
In the IFM (inference for margins) method, the log-likelihood function for each of the marginal
distributions
(5)
j
(
j
) =
T

t=1
lnf
i
(x
j,t
;
j
), j = 1, . . . , d
is maximized to obtain estimates (

1
, . . . ,

d
)

. The function
(6) (,

1
, . . . ,

d
) =
T

t=1
[ln c{F
X
1
(x
1,t
;

1
), . . . , F
X
d
(x
d,t
;

d
); }]
is then maximized over to get the dependence parameter estimate

. The estimates

IFM
=
(

1
, . . . ,

d
,

)

solve
(
1
/
1
, . . . ,
d
/
d
, /) = 0
Backtesting
This procedure is applied to a daily exchange rate portfolio (DEM/USD and GBP/USD
from 01.12.1979 to 01.04.1994) with T = 250. The univariate risk factor increments (log
returns) are assumed to be Gaussian distributed with parameters estimated from the data.
The selected copulae belong to the bivariate one-parametric Gumbel family:
(7) C(u, v) = exp([(ln u)

+ (ln v)

1
), 1
To evaluate the performance of the copula in the VaR calculations, dierent portfolio com-
positions are used to generate P&L samples. The quantiles from the samples at four levels

1
= 0.05,
2
= 0.01,
3
= 0.005 and
4
= 0.001 are used as estimators for VaR.
The estimated VaR is compared with the realization of the P&L function, an exceedance occur-
ing for each P&L value smaller than the estimated VaR. The ratio of the number of exceedances
to the number of observations gives the empirical level .
VaR - Gumbel Copula
1983 1986 1989 1992
time
-
0
.
2
-
0
.
1
0
0
.
1
P
&
L
Figure 1: Value-at-Risk at levels
1
= 0.05 (yellow),
2
= 0.01 (green),
3
= 0.005 (red), and

4
= 0.001 (blue ), P&L (black), estimated at each time from a Monte Carlo sample of 10.000
P&L values generated with Gumbel copula, w = (2, 1)

.
VaR - Gumbel Copula
1983 1986 1989 1992
time
-
0
.
2
-
0
.
1
0
0
.
1
P
&
L
Figure 2: Value-at-Risk at level = 0.05 (yellow), P&L (black) and exceedances (red), =
0.0573, w = (2, 1)

. P&L samples generated with Gumbel copula.

Table 1: Gumbel copula, empirical levels for dierent portfolios.
level (10
2
)
5 1 0.5 0.1
Portfolio w

empirical level (10

2
)
(1, 1) 6.05 2.45 1.75 0.83
(1, 2) 6.34 2.74 1.75 1.00
(2, 1) 5.73 2.24 1.58 0.69
(2, 3) 6.22 2.56 1.75 0.92
(3, 2) 5.99 2.30 1.55 0.74
(1, 2) 1.64 0.37 0.20 0.11
(1, 2) 2.01 0.51 0.43 0.11
(2, 1) 4.44 1.49 0.95 0.40
(2, 1) 4.09 1.35 1.09 0.49
REFERENCES
Durrelman, V., Nikeghbali, A. and Roncalli, T. (2000). Which Copula is the Right One
?, Groupe de Recherche Operationnelle Credit Lyonnais.
Embrechts, P., McNeil, A. and Straumann, D. (1999). Correlation and Dependence in
Risk Management: Properties and Pitfalls Correlation, Risk Management: Value at Risk and
Beyond, Cambridge University Press, Cambridge.
Franke, J., Hardle, W. and Hafner, C. (2004). Statistics of Financial Markets, Springer-
Verlag, Heidelberg.
Giacomini, E. and Hardle, W. (2005). Nonparametric Risk Management with Adaptive
Copulae, Discussion Paper, SFB 649.
Hardle, W., Herwartz, H. and Spokoiny, V. (2003). Time Inhomogeneous Multiple Volatil-
ity Modeling, Journal of Financial Econometrics, 2003 1 (1): 55-95.
Hardle, W., Kleinow, T. and Stahl, G. (2002). Applied Quantitative Finance, Springer-
Verlag, Heidelberg.
Joe, H. (1997). Multivariate Models and Dependence Concepts, Chapman & Hall, London.
Nelsen, R. (1998). An Introduction to Copulas, Springer-Verlag, New York.
ACKNOWLEDGEMENT
Financial support of Deutsche Forschungsgemeinschaft via SFB 649

Okonomisches Risiko,
Humboldt-Universitat zu Berlin, is gratefully acknowledged.
R

ESUM

E
La Value at Risk (VaR) dun portefeuille est determinee par la distribution multivariee des
increments des facteurs de risques. Cette distribution peut etre modelisee par des copules dont
les param`etres ne sont pas necessairement constants par rapport au temps. Pour un portefeuille
de taux de change, des copules dependant du temps sont estimes et la VaR est ainsi simulee.
Le backtesting conrme lamelioration apportee par les copules dependant du temps.
SFB 649 Discussion Paper Series

For a complete list of Discussion Papers published by the SFB 649,
please visit http://sfb649.wiwi.hu-berlin.de.


001 Nonparametric Risk Management with Generalized
Hyperbolic Distributions by Ying Chen, Wolfgang Hrdle
and Seok-Oh Jeong, January 2005.
002 Selecting Comparables for the Valuation of the European
Firms by Ingolf Dittmann and Christian Weiner, February
2005.
003 Competitive Risk Sharing Contracts with One-sided
Commitment by Dirk Krueger and Harald Uhlig, February
2005.
004 Value-at-Risk Calculations with Time Varying Copulae by
Enzo Giacomini and Wolfgang Hrdle, February 2005.
SFB 649, Spandauer Strae 1, D-10178 Berlin
http://sfb649.wiwi.hu-berlin.de

This research was supported by the Deutsche
Forschungsgemeinschaft through the SFB 649 Economic Risk.