Brechmann

Statistical inference of vine copulas using the
R-package VineCopula
Eike Christian Brechmann
[email protected]
Technische Universit
at M
unchen
May 23, 2013
Eike Brechmann (TUM)
The R-package VineCopula
May 23, 2013
1 / 26
Available software for vines
In general...
I Uncertainty analysis with Correlations (UNICORN, TU Delft)
includes some functionality for vines.
In R...
I Packages for bivariate and multivariate copulas (copula, fCopulae,
QRMlib,...).
I Daniel Berg (U Oslo/NR): copulaGOF/CopulaLib.
The R-packages VineCopula and CDVine fill this gap.
May 23, 2013
2 / 26
Available software for vines
In general...
I Uncertainty analysis with Correlations (UNICORN, TU Delft)
includes some functionality for vines.
In R...
I Packages for bivariate and multivariate copulas (copula, fCopulae,
QRMlib,...).
I Daniel Berg (U Oslo/NR): copulaGOF/CopulaLib.
The R-packages VineCopula and CDVine fill this gap.
May 23, 2013
2 / 26
Scope of VineCopula
VineCopula provides functions for bivariate analysis...
graphical tools
analytical tools
BiCop...
selection and estimation of bivariate copulas

simulation of bivariate copulas
...and for multivariate analysis using (simplified) regular vine copulas.
sequential and joint maximum likelihood estimation
simulation of vine copulas
RVine...
model selection
illustration of vine trees
General assumption: Data lies in the unit hypercube [0, 1]d .
May 23, 2013
3 / 26
Scope of VineCopula
graphical tools
analytical tools
BiCop...

RVine...
model selection
May 23, 2013
3 / 26
Scope of VineCopula
graphical tools
analytical tools
BiCop...

RVine...
model selection
May 23, 2013
3 / 26
The package CDVine
Functionality for the sub-classes of C- and D-vines.

Links to the package VineCopula: C2RVine and D2RVine.
Vignette:
Brechmann & Schepsmeier (2013).
Modeling dependence with C- and D-vine copulas:
The R-package CDVine.
Journal of Statistical Software 52(3), 127.
May 23, 2013
4 / 26
The building blocks: Bivariate copula families

Each family is denoted by a number to shorten notation (0 = indep.).
Elliptical copulas:
family = 1 Gaussian copula
family = 2 Students t copula
One parameter Archimedean copulas:
Two parameter Archimedean copulas:
Density, distribution & h-functions: BiCopPDF, BiCopCDF & BiCopHfunc.

May 23, 2013
5 / 26

Elliptical copulas:
family = 3 Clayton copula
family = 4 Gumbel copula
family = 5 Frank copula
family = 6 Joe copula

May 23, 2013
5 / 26

Elliptical copulas:
family = 3 Clayton copula
family = 4 Gumbel copula
family = 6 Joe copula
family = 7 Clayton-Gumbel (BB1) copula
family = 8 Joe-Gumbel (BB6) copula
family = 9 Joe-Clayton (BB7) copula
family = 10 Joe-Frank (BB8) copula
May 23, 2013
5 / 26

Elliptical copulas:
family = 13 survival Clayton copula
family = 14 survival Gumbel copula
family = 16 survival Joe copula
family = 17 survival BB1 copula
May 23, 2013
5 / 26

Elliptical copulas:
family = 23 rotated Clayton copula (90 degrees)
family = 24 rotated Gumbel copula (90 degrees)
family = 26 rotated Joe copula (90 degrees)
family = 27 rotated BB1 copula (90 degrees)
May 23, 2013
5 / 26

Elliptical copulas:
May 23, 2013
5 / 26

Elliptical copulas: (parameters: par, par2 (degrees of freedom))
One parameter Archimedean copulas: (parameter: par)
Two parameter Archimedean copulas: (parameters: par, par2)
May 23, 2013
5 / 26

Elliptical copulas: (parameters: par, par2 (degrees of freedom))
One parameter Archimedean copulas: (parameter: par)
Two parameter Archimedean copulas: (parameters: par, par2)
May 23, 2013
5 / 26
Rotation of copulas
I Rotate Archimedean copulas to capture negative dependence:
if (U1 , U2 ) C90 , then (1 U1 , U2 ) C0 .
I Survival copulas correspond to rotation by 180 degrees.
Clayton copulas rotated by 0, 90, 180 and 270 degrees

>
>
>
>
dat0 = BiCopSim(N=500, family=3, par=2)

dat90 = BiCopSim(N=500, family=23, par=-2)
May 23, 2013
6 / 26
Rotation of copulas
I Rotate Archimedean copulas to capture negative dependence:
if (U1 , U2 ) C90 , then (1 U1 , U2 ) C0 .
I Survival copulas correspond to rotation by 180 degrees.
Clayton copulas rotated by 0, 90, 180 and 270 degrees

0.0
0.2
0.4
0.6
u1
0.8
1.0
0.0
0.2
0.4
0.6
u1
0.8
1.0
1.0
270 degrees
0.0
0.2
0.4
0.6
u1
0.8
0.8
1.0
0.6
u2
0.4
0.2
0.0
0.6
0.8
u2
0.8
180 degrees
1.0
0.4
1.0
0.6
0.2
u2
0.0
0.2
0.4
0.0
0.8
0.6
u2
90 degrees
0.4
0.2
1.0
0 degrees
0.0
>
>
>
>
0.0
0.2
0.4
0.6
0.8
1.0
u1
May 23, 2013
6 / 26
Two parameter Archimedean copulas

I The BB1 and the BB7 copula can model tail-asymmetric dependence
with different non-zero lower and upper tail dependence.
I Density expressions and derivatives are however numerically involved.
Clayton-Gumbel (BB1) copula

> BiCopPar2Tau(family=7,
+
par=0.25, par2=2.5)
[1] 0.64
> BiCopPar2TailDep(family=7,
+
par=0.25, par2=2.5)
$lower
[1] 0.33
$upper
[1] 0.68
May 23, 2013
7 / 26


+
par=0.25, par2=2.5)
[1] 0.64
+
par=0.25, par2=2.5)
$lower
[1] 0.33
$upper
[1] 0.68
May 23, 2013
7 / 26


+
par=0.25, par2=2.5)
> BiCopMetaContour(family=7,
+
par=0.25, par2=2.5)
0.2
0.15
0.1
0.05
$lower
[1] 0.33
+
par=0.25, par2=2.5)
[1] 0.64
0.01
$upper
[1] 0.68
May 23, 2013
7 / 26
Reminder: R-vine copulas

Three components: vine trees, pair copulas, copula parameters
2
5
4
3
T1
Density
c = c14 c15 c24 c34
c12;4 c13;4 c45;1

c23;14 c35;14
c25;134
May 23, 2013
8 / 26

1,5
5
2,4
3,4
1,4
1
4
T1
Density
c = c14 c15 c24 c34
c12;4 c13;4 c45;1
c23;14 c35;14
c25;134
May 23, 2013
8 / 26

1,5
5
2,4
3,4
1,4
1
4
T1
Density
2,4
1,5
c12;4 c13;4 c45;1
1,4
3,4
c = c14 c15 c24 c34

T2
c23;14 c35;14
c25;134
May 23, 2013
8 / 26

1,5
5
2,4
3,4
1,4
1
4
T1
Density
2,4
1,5
c12;4 c13;4 c45;1
1,4
3,4
c = c14 c15 c24 c34

T2
c23;14 c35;14
c25;134
May 23, 2013
8 / 26

1,5
5
2,4
3,4
1,4
1
4
T1
Density
1,2|4
2,4
c = c14 c15 c24 c34
4,5|1
1,5
c12;4 c13;4 c45;1
1,4
1,3|4
3,4
T2
c23;14 c35;14
c25;134
May 23, 2013
8 / 26

1,5
5
2,4
3,4
1,4
1
4
T1
Density
1,2|4
2,4
c = c14 c15 c24 c34
4,5|1
1,5
c12;4 c13;4 c45;1
1,4
1,3|4
4,5|1
1,3|4
3,4
1,2|4
T2
c23;14 c35;14
c25;134
T3
May 23, 2013
8 / 26

1,5
2,4
3,4
1,4
4
T1
Density
2,4
1,2|4
c = c14 c15 c24 c34
4,5|1
1,5
c12;4 c13;4 c45;1
1,4
1,3|4
3,5|14
4,5|1
3,4
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
T2
c23;14 c35;14
c25;134
T3
T4
May 23, 2013
8 / 26
Storing R-vine copulas in matrix notation

Efficient enconding of R-vine models needed for statistical inference.
I Matrix notation by Morales-Napoles et al. (2010) and Dimann et al.
(2013).
2
5
3
1
4
3
5 4
1 5 5
4 1 1 1
1,5
5
2,4
3,4
1,4
1
1,2|4
2,4
1,3|4
3,4
T1
4,5|1
1,5
1
2
3
4
1,4
{2, 4}, {3, 4}, {4, 1}, {5, 1}

{2, 1|4}, {3, 1|4}, {4, 5|1}
{2, 3|14}, {3, 5|14}
3,5|14
4,5|1
{2, 5|314}
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
May 23, 2013
T2
T3
T4
9 / 26

(2013).
2
5
3
1
4
3
5 4
1 5 5
4 1 1 1
1,5
5
2,4
3,4
1,4
1
1,2|4
2,4
1,3|4
3,4
T1
4,5|1
1,5
1
2
3
4
1,4
{2, 4}, {3, 4}, {4, 1}, {5, 1}

{2, 1|4}, {3, 1|4}, {4, 5|1}
{2, 3|14}, {3, 5|14}
3,5|14
4,5|1
{2, 5|314}
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
May 23, 2013
T2
T3
T4
9 / 26

(2013).
2
5
3
1
4
3
5 4
1 5 5
4 1 1 1
1,5
5
2,4
3,4
1,4
1
1,2|4
2,4
1,3|4
3,4
T1
4,5|1
1,5
1
2
3
4
1,4
{2, 4}, {3, 4}, {4, 1}, {5, 1}

{2, 1|4}, {3, 1|4}, {4, 5|1}
{2, 3|14}, {3, 5|14}
3,5|14
4,5|1
{2, 5|314}
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
May 23, 2013
T2
T3
T4
9 / 26

(2013).
2
5
3
1
4
3
5 4
1 5 5
4 1 1 1
1,5
5
2,4
3,4
1,4
1
1,2|4
2,4
1,3|4
3,4
T1
4,5|1
1,5
1
2
3
4
1,4
{2, 4}, {3, 4}, {4, 1}, {5, 1}

{2, 1|4}, {3, 1|4}, {4, 5|1}
{2, 3|14}, {3, 5|14}
3,5|14
4,5|1
{2, 5|314}
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
May 23, 2013
T2
T3
T4
9 / 26

(2013).
2
5
3
1
4
3
5 4
1 5 5
4 1 1 1
1,5
5
2,4
3,4
1,4
1
1,2|4
2,4
1,3|4
3,4
T1
4,5|1
1,5
1
2
3
4
1,4
{2, 4}, {3, 4}, {4, 1}, {5, 1}

{2, 1|4}, {3, 1|4}, {4, 5|1}
{2, 3|14}, {3, 5|14}
3,5|14
4,5|1
{2, 5|314}
2,3|14
1,3|4
1,2|4
2,5|134
3,5|14
2,3|14
May 23, 2013
T2
T3
T4
9 / 26
R-vine copula and parameter matrices
Copula families
2
5 3
3 5
1 1
4 4
and parameters can be stored in associated matrices.
C25;314
C23;14 C35;14
C21;4
C31;4 C45;1
5 5
C24
C34
C41 C51
1 1 1
May 23, 2013
10 / 26
R-vine matrix objects

An RVineMatrix object contains all required matrices:
>
+
>
>
>
+
>
>
>
+
>
>
>
>
>
+
Matrix = c(2,5,3,1,4,0,3,5,1,4,0,0,4,5,1,
0,0,0,5,1,0,0,0,0,1)
Matrix = matrix(Matrix,5,5)
family = c(0,1,3,4,4,0,0,3,4,1,0,0,0,4,1,
0,0,0,0,3,0,0,0,0,0)
family = matrix(family,5,5)
par = c(0,0.2,0.9,1.5,3.9,0,0,1.1,1.6,0.9,0,0,0,1.9,0.5,
0,0,0,0,4.8,0,0,0,0,0)
par = matrix(par,5,5)
par2 = matrix(0,5,5)
RVM = RVineMatrix(Matrix=Matrix, family=family, par=par,
par2=par2, names=c("V1","V2","V3","V4","V5"))
May 23, 2013
11 / 26
Simulation
> simdat = RVineSim(500, RVM)

> head(simdat)
V1
V2
[1,] 0.51 0.24
[2,] 0.23 0.14
[3,] 0.65 0.38
[4,] 0.43 0.18
[5,] 0.86 0.86
[6,] 0.71 0.71
V3
0.42
0.16
0.46
0.08
0.85
0.80
V4
0.33
0.12
0.29
0.08
0.86
0.68
V5
0.45
0.20
0.70
0.26
0.87
0.88
May 23, 2013
12 / 26
Simulation
V2
V3
V4
0.47
V5
0.8
0.4
0.0
0.44
0.74 0.70
0.31
0.8
0.4
0.0
0.72 0.60 0.62

0.0
0.78
0.8
0.43
0.4
0.0
V1
0.8
V5
0.45
0.20
0.70
0.26
0.87
0.88
0.4
V4
0.33
0.12
0.29
0.08
0.86
0.68
0.8
0.0
V3
0.42
0.16
0.46
0.08
0.85
0.80
0.4
> simdat = RVineSim(500, RVM)

> head(simdat)
V1
V2
[1,] 0.51 0.24
[2,] 0.23 0.14
[3,] 0.65 0.38
[4,] 0.43 0.18
[5,] 0.86 0.86
[6,] 0.71 0.71
0.0
0.8
0.8
0.4
0.4
0.0
0.0
0.4
0.8
0.0
0.4
0.8
0.0
May 23, 2013
0.4
0.8
12 / 26
Example
Daily log returns of 15 major German stocks.

Observed from January 2005 to August 2009 (1158 observations).
Time series are filtered using GARCH(1,1) with Students t
innovations.
Data set of standardized residuals transformed to [0,1].
Load into workspace:
> data(daxreturns)
Now: selection of trees, pair copulas and parameters in inverse order.
May 23, 2013
13 / 26
A first look at the data
0.32
BAYN.DE
0.42
0.35
0.29
BMW.DE
0.43
0.38
0.30
0.49
DAI.DE
0.0
0.4
0.8
0.0
0.4
0.8
0.40
0.4
BAS.DE
0.0
0.8
0.8
0.8
0.4
0.0
0.4
0.40
0.0
0.4
0.8
0.8
ALV.DE
0.4
0.0
0.0
0.0
0.4
0.8
0.0
0.4
0.8
0.0
0.4
0.8
May 23, 2013
14 / 26
Parameter estimation I
Sequential estimation (based on BiCopEst)

either using bivariate inversion of Kendalls :
> RVineSeqEst(data, RVM, method="itau")
or bivariate maximum likelihood estimation:
> RVineSeqEst(data, RVM, method="mle")
I Very fast, since only bivariate estimation.

I Provides good starting values for joint maximum likelihood estimation.
May 23, 2013
15 / 26
Parameter estimation II
Maximum likelihood estimation of all parameters jointly

(log-likelihood computation: RVineLogLik).
> RVineMLE(data, RVM, start, start2, maxit,
+
grad, hessian, se)
I Starting values can be calculated using sequential estimation.
I Analytical gradient can be used for numerical optimization
(see RVineGrad).
I Standard errors can be computed based on the analytical Hessian
(see RVineStdError and RVineHessian).
May 23, 2013
16 / 26

> RVineMLE(data, RVM, start=0, start2=0, maxit,
+
grad, hessian, se)
(see RVineGrad).
May 23, 2013
16 / 26

+
grad=TRUE, hessian, se)
(see RVineGrad).
May 23, 2013
16 / 26

+
grad, hessian=TRUE, se=TRUE)
(see RVineGrad).
May 23, 2013
16 / 26
Parameter estimation III

> mle = RVineMLE(data=daxreturns[,1:5], RVM, start=0, start2=0,
+
grad=TRUE, hessian=TRUE, se=TRUE)
> mle$RVM$par
[,1] [,2]
[1,] 0.00 0.000
[2,] 0.14 0.000
[3,] 0.70 0.079
[4,] 1.30 1.206
[5,] 1.42 0.366
> mle$se
[,1]
[1,] 0.000
[2,] 0.029
[3,] 0.059
[4,] 0.031
[5,] 0.032
[,3]
0.00
0.00
0.00
1.42
0.52
[,2]
0.000
0.000
0.037
0.028
0.024
[,4] [,5]
0.00
0
0.00
0
0.00
0
0.00
0
0.99
0
[,3]
0.000
0.000
0.000
0.036
0.020
[,4] [,5]
0.000
0
0.000
0
0.000
0
0.000
0
0.059
0
May 23, 2013
17 / 26

+
> mle$RVM$par
[,1] [,2]
[1,] 0.00 0.000
[2,] 0.14 0.000
[3,] 0.70 0.079
[4,] 1.30 1.206
[5,] 1.42 0.366
> mle$se
[,1]
[1,] 0.000
[2,] 0.029
[3,] 0.059
[4,] 0.031
[5,] 0.032
[,3]
0.00
0.00
0.00
1.42
0.52
[,2]
0.000
0.000
0.037
0.028
0.024
[,4] [,5]
0.00
0
0.00
0
0.00
0
0.00
0
0.99
0
[,3]
0.000
0.000
0.000
0.036
0.020
[,4] [,5]
0.000
0
0.000
0
0.000
0
0.000
0
0.059
0
May 23, 2013
17 / 26

+
> mle$RVM$par
[,1] [,2]
[1,] 0.00 0.000
[2,] 0.14 0.000
[3,] 0.70 0.079
[4,] 1.30 1.206
[5,] 1.42 0.366
> mle$se
[,1]
[1,] 0.000
[2,] 0.029
[3,] 0.059
[4,] 0.031
[5,] 0.032
[,3]
0.00
0.00
0.00
1.42
0.52
[,2]
0.000
0.000
0.037
0.028
0.024
[,4] [,5]
0.00
0
0.00
0
0.00
0
0.00
0
0.99
0
[,3]
0.000
0.000
0.000
0.036
0.020
[,4] [,5]
0.000
0
0.000
0
0.000
0
0.000
0
0.059
0
May 23, 2013
17 / 26
Pair copula selection

Manually using tools for bivariate analysis (e.g., contour plots or
goodness-of-fit tests: BiCopMetaContour, BiCopGofTest,...).
Automatically using AIC or BIC from a set of copula families:
BiCopSelect (bivariate) or RVineCopSelect (multivariate).
> cops = RVineCopSelect(data=daxreturns[,1:5], familyset=NA,
+
Matrix=Matrix, selectioncrit="AIC",
+
indeptest=FALSE, level=0.05)
> cops$family
[,1] [,2] [,3] [,4] [,5]
[1,]
0
0
0
0
0
[2,]
5
0
0
0
0
[3,]
2
5
0
0
0
[4,]
2
20
2
0
0
[5,]
17
2
14
20
0
May 23, 2013
18 / 26

+
+
> cops$family
[,1] [,2] [,3] [,4] [,5]
[1,]
0
0
0
0
0
[2,]
5
0
0
0
0
[3,]
2
5
0
0
0
[4,]
2
20
2
0
0
[5,]
17
2
14
20
0
May 23, 2013
18 / 26

+
+
> cops$family
[,1] [,2] [,3] [,4] [,5]
[1,]
0
0
0
0
0
[2,]
5
0
0
0
0
[3,]
2
5
0
0
0
[4,]
2
20
2
0
0
[5,]
17
2
14
20
0
May 23, 2013
18 / 26

+
+
> cops$family
[,1] [,2] [,3] [,4] [,5]
[1,]
0
0
0
0
0
[2,]
5
0
0
0
0
[3,]
2
5
0
0
0
[4,]
2
20
2
0
0
[5,]
17
2
14
20
0
May 23, 2013
18 / 26
Vine tree selection I

Sequential tree-by-tree selection according to Dimann et al. (2013):
Select maximum spanning tree (respecting the proximity condition) in
terms of the absolute empirical pairwise Kendalls values.
2 Select and estimate pair copulas of the tree.
3 Compute transformed observations using h-functions and go back to
Step 1.
1
I R- and C-vine copulas can be selected.

I The vine copula can be truncated to reduce the model complexity.
May 23, 2013
19 / 26

Step 1.
1
> RVineStructureSelect(data, familyset, type,

+
selectioncrit, indeptest,
+
level, trunclevel)
May 23, 2013
19 / 26

Step 1.
1
> RVineStructureSelect(data, familyset, type="RVine",

+
+
level, trunclevel)
May 23, 2013
19 / 26

Step 1.
1
> RVineStructureSelect(data, familyset, type,

+
+
level, trunclevel=2)
May 23, 2013
19 / 26
Vine tree selection II
> rvm = RVineStructureSelect(data=daxreturns)

> rvm$Matrix
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13] [,14] [,15]
[1,]
15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
[2,]
11
4
0
0
0
0
0
0
0
0
0
0
0
0
0
[3,]
9
11
3
0
0
0
0
0
0
0
0
0
0
0
0
[4,]
7
9
11
11
0
0
0
0
0
0
0
0
0
0
0
[5,]
13
7
9
12
9
0
0
0
0
0
0
0
0
0
0
[6,]
3
13
7
8
12
7
0
0
0
0
0
0
0
0
0
[7,]
12
3
13
10
8
12
13
0
0
0
0
0
0
0
0
[8,]
8
12
5
5
10
8
12
5
0
0
0
0
0
0
0
[9,]
10
8
1
2
5
10
8
12
1
0
0
0
0
0
0
[10,]
2
10
12
13
2
5
10
8
12
12
0
0
0
0
0
[11,]
1
2
8
9
13
2
5
10
8
10
8
0
0
0
0
[12,]
14
1
6
7
1
13
2
2
10
6
10
10
0
0
0
[13,]
6
14
14
14
14
1
14
1
2
14
6
6
2
0
0
[14,]
4
6
10
6
6
14
6
14
14
2
14
14
6
14
0
[15,]
5
5
2
1
7
6
1
6
6
8
2
2
14
6
6
May 23, 2013
20 / 26
Vine tree selection II
> rvm = RVineStructureSelect(data=daxreturns)

> rvm$Matrix
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13] [,14] [,15]
[1,]
15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
[2,]
11
4
0
0
0
0
0
0
0
0
0
0
0
0
0
[3,]
9
11
3
0
0
0
0
0
0
0
0
0
0
0
0
[4,]
7
9
11
11
0
0
0
0
0
0
0
0
0
0
0
[5,]
13
7
9
12
9
0
0
0
0
0
0
0
0
0
0
[6,]
3
13
7
8
12
7
0
0
0
0
0
0
0
0
0
[7,]
12
3
13
10
8
12
13
0
0
0
0
0
0
0
0
[8,]
8
12
5
5
10
8
12
5
0
0
0
0
0
0
0
[9,]
10
8
1
2
5
10
8
12
1
0
0
0
0
0
0
[10,]
2
10
12
13
2
5
10
8
12
12
0
0
0
0
0
[11,]
1
2
8
9
13
2
5
10
8
10
8
0
0
0
0
[12,]
14
1
6
7
1
13
2
2
10
6
10
10
0
0
0
[13,]
6
14
14
14
14
1
14
1
2
14
6
6
2
0
0
[14,]
4
6
10
6
6
14
6
14
14
2
14
14
6
14
0
[15,]
5
5
2
1
7
6
1
6
6
8
2
2
14
6
6
May 23, 2013
20 / 26
Illustrating R-vine copula models

Selected R-vine trees:
> RVineTreePlot(data=NULL, RVM=rvm, tree=1,
+
edge.labels=c("family","theotau"))
May 23, 2013
21 / 26

+
edge.labels=c("family","theotau"))
Tree 1
FME.DE
SBB1,0.22
VOW3.DE
RWE.DE
t,0.35
DTE.DE
t,0.55
SBB8,0.34
DAI.DE
t,0.44
t,0.48
EOAN.DE
BMW.DE
DBK.DE
t,0.45
t,0.37
SIE.DE
t,0.43
BAS.DE
t,0.4
t,0.52
LIN.DE
t,0.42
BAYN.DE
ALV.DE
t,0.37
t,0.52
SAP.DE
MUV2.DE
May 23, 2013
21 / 26

+
edge.labels=FALSE)
Tree 2
DTE.DE,FME.DE
ALV.DE,MUV2.DE
DBK.DE,DTE.DE
ALV.DE,DBK.DE
BAS.DE,BAYN.DE
ALV.DE,SAP.DE
DBK.DE,SIE.DE
BAS.DE,LIN.DE
BAS.DE,SIE.DE
DAI.DE,DBK.DE
BAS.DE,EOAN.DE
BMW.DE,DAI.DE
DAI.DE,VOW3.DE
EOAN.DE,RWE.DE
May 23, 2013
21 / 26
Discrimination among vine copula models I
AIC and BIC: RVineAIC and RVineBIC.

Tests by Vuong (1989) and by Clarke (2007) for non-nested
comparisons of two R-vine models RVM1 and RVM2:
> RVineVuongTest(data, RVM1, RVM2)
> RVineClarkeTest(data, RVM1, RVM2)
May 23, 2013
22 / 26
Discrimination among vine copula models II

Select a C-vine copula for comparison:
> cvm = RVineStructureSelect(daxreturns, type="CVine")
Compare the models in terms of AIC values and the Vuong test:
> c(RVineAIC(daxreturns,rvm)$AIC, RVineAIC(daxreturns,cvm)$AIC)
[1] -9808.44 -9804.42
> RVineVuongTest(daxreturns, rvm, cvm)
$statistic
[1] 0.27
$statistic.Akaike
[1] 0.068
$statistic.Schwarz
[1] -0.44
...
The models are essentially indistinguishable.

May 23, 2013
23 / 26
Discrimination among vine copula models II

Select a C-vine copula for comparison:
> cvm = RVineStructureSelect(daxreturns, type="CVine")
Compare the models in terms of AIC values and the Vuong test:
> c(RVineAIC(daxreturns,rvm)$AIC, RVineAIC(daxreturns,cvm)$AIC)
[1] -9808.44 -9804.42
> RVineVuongTest(daxreturns, rvm, cvm)
$statistic
[1] 0.27
$statistic.Akaike
[1] 0.068
$statistic.Schwarz
[1] -0.44
...
The models are essentially indistinguishable.

May 23, 2013
23 / 26
Outlook
Current projects and plans:
Move code from C to C++.
Parallelize numerical maximum likelihood estimation.
Implement asymmetric copulas (Tawn).
Bayesian estimation and model selection.
Thank you to my package coauthors: Ulf Schepsmeier & Jakob Stober
May 23, 2013
24 / 26
Outlook
Current projects and plans:
Move code from C to C++.
Parallelize numerical maximum likelihood estimation.
Implement asymmetric copulas (Tawn).
Bayesian estimation and model selection.
Thank you to my package coauthors: Ulf Schepsmeier & Jakob Stober
May 23, 2013
24 / 26
Bibliography
Brechmann, E. C. and C. Czado (2013).
Risk management with high-dimensional vine copulas: An analysis of the Euro Stoxx 50.
Statistics & Risk Modeling , forthcoming.
Brechmann, E. C. and U. Schepsmeier (2013).
Modeling dependence with C- and D-vine copulas: The R-package CDVine.
Journal of Statistical Software 52 (3), 127.
Clarke, K. A. (2007).
A simple distribution-free test for nonnested model selection.
Political Analysis 15 (3), 347363.
Dimann, J., E. C. Brechmann, C. Czado, and D. Kurowicka (2013).
Selecting and estimating regular vine copulae and application to financial returns.
Computational Statistics & Data Analysis 59 (1), 5269.
Morales-N
apoles, O., R. M. Cooke, and D. Kurowicka (2010).
About the number of vines and regular vines on n nodes.
Working paper .
Vuong, Q. H. (1989).
Ratio tests for model selection and non-nested hypotheses.
Econometrica 57 (2), 307333.
May 23, 2013
25 / 26
Thank you very much for your attention!
Visit: http://cran.r-project.org/web/packages/VineCopula/
May 23, 2013
26 / 26

Brechmann

Uploaded by

Copyright:

Available Formats

Brechmann

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brechmann

Uploaded by

Copyright:

Available Formats

Statistical inference of vine copulas using the

May 23, 2013

Eike Brechmann (TUM)

The R-package VineCopula

May 23, 2013

Available software for vines

Eike Brechmann (TUM)

The R-package VineCopula

May 23, 2013

Available software for vines

Eike Brechmann (TUM)

The R-package VineCopula

May 23, 2013

selection and estimation of bivariate copulas

The R-package VineCopula

May 23, 2013

selection and estimation of bivariate copulas

The R-package VineCopula

May 23, 2013

selection and estimation of bivariate copulas

The R-package VineCopula

May 23, 2013

The package CDVine

Functionality for the sub-classes of C- and D-vines.

Eike Brechmann (TUM)

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

Two parameter Archimedean copulas:

Density, distribution & h-functions: BiCopPDF, BiCopCDF & BiCopHfunc.

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

Density, distribution & h-functions: BiCopPDF, BiCopCDF & BiCopHfunc.

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

The building blocks: Bivariate copula families

The R-package VineCopula

May 23, 2013

I Survival copulas correspond to rotation by 180 degrees.

Clayton copulas rotated by 0, 90, 180 and 270 degrees

dat0 = BiCopSim(N=500, family=3, par=2)

Eike Brechmann (TUM)

The R-package VineCopula