Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
GHICA - Risk Analysis with GH Distributions andIndependent Components
Ying Chen
Wolfgang Hardle
Vladimir Spokoiny
Institut fur Statistik and OkonometrieCASE - Center for Applied Statistics and EconomicsHumboldt-Universitat zu Berlin
Weierstraß-Institutfur Angewandte Analysis und Stochastik
http://ise.wiwi.hu-berlin.dehttp://www.case.hu-berlin.dehttp://www.wias-berlin.de
W e ie rstra ß -In stitu t fü r A n g e w a n d te A n a ly s is u n d S to ch a stik
Motivation 1-1
Value at Risk (VaR) calculation
Rt = b>t xt = b>t Σ1/2t εt
Rt the portfolio returnbt the portfolio weightsxt ∈ Rd the individual returns with cov Σt
εt the stochastic term.
VaR at level 100a%: VaRa,t = F−1a,t (Rt),
F−1t equals the inverse of cdf of Rt .
Critical point: pdf of xt .
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-2
Value at Risk (VaR) calculation
Rt = b>t xt = b>t Σ1/2t εt
Rt the portfolio returnbt the portfolio weightsxt ∈ Rd the individual returns with cov Σt
εt the stochastic term.
RiskMetrics
εt ∼ N(0, Id)Σt = $Σt−1 + (1−$)xt−1x
>t−1 (Exponential Moving Average)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-3
Value at Risk (VaR) calculation
Rt = b>t xt = b>t Σ1/2t εt
Rt the portfolio returnbt the portfolio weightsxt ∈ Rd the individual returns with cov Σt
εt the stochastic term.
T-deGARCH
εt ∼ t(df )Σt = $ + α1Σt−1 + β1xt−1x
>t−1 (GARCH(1,1))
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-4
A univariate story
Chen, Hardle and Jeong (2005), GHADA:Generalized Hyperbolic distribution + ADAptive volatilityThe Gaussian density (left) and GH density (right) fits of the devolatilized returns of
DEM/USD rates from 1979-12-01 to 1994-04-01 (black: nonparametric kernel density,
red: normal density, blue: GH density).
Gaussian density
-4 -2 0 2 4
X
00.
10.
20.
30.
40.
5
Y
GH(1,1.74,-0.02,0.78,0.01)
-4 -2 0 2 4
X
00.
10.
20.
30.
4
Y
GARCH(1,1) + t(15) density
-4 -2 0 2 4
X
00.
10.
20.
30.
40.
5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-5
A univariate story
Chen, Hardle and Jeong (2005), GHADA:Generalized Hyperbolic distribution + ADAptive volatilityThe t(15) density (right) and GH density (left) fits of the devolatilized returns of
DEM/USD rates (black: nonparametric kernel density, green: t(15) density, blue: GH
density). GARCH(1,1) process: σ2t = 1.65e − 06 + 0.07x2
t−1 + 0.89σ2t−1.
Gaussian density
-4 -2 0 2 4
X
00.
10.
20.
30.
40.
5
Y
GH(1,1.74,-0.02,0.78,0.01)
-4 -2 0 2 4
X
00.
10.
20.
30.
4
Y
GARCH(1,1) + t(15) density
-4 -2 0 2 4
X
00.
10.
20.
30.
40.
5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-6
Excursion to GHADA
X ∼ GH with density:
fGH(x ;λ, α, β, δ, µ) =(ι/δ)λ
√2πKλ(δι)
Kλ−1/2
{α√
δ2 + (x − µ)2}
{√δ2 + (x − µ)2
/α}1/2−λ
· eβ(x−µ)
Where ι2 = α2 − β2, Kλ(·) is the modified Bessel function of thethird kind with index λ: Kλ(x) = 1
2
∫∞0 yλ−1exp{− x
2 (y + y−1)} dy .
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-7
Subclass of GH distribution
The parameters (µ, δ, β, α)> can be interpreted as trend, riskiness,asymmetry and the likeliness of extreme events.Hyperbolic (HYP) distributions: λ = 1,
fHYP(x ;α, β, δ, µ) =ι
2αδK1(δι)e{−α
√δ2+(x−µ)2+β(x−µ)}, (1)
where x , µ ∈ R, 0 ≤ δ and |β| < α.Normal-inverse Gaussian (NIG) distributions: λ = −1/2,
fNIG (x ;α, β, δ, µ) =αδ
π
K1
{α√
δ2 + (x − µ)2}
√δ2 + (x − µ)2
e{δι+β(x−µ)}, (2)
where x , µ ∈ R, 0 < δ and |β| ≤ α.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-8
Adaptive Volatility Estimation
Volatility (and also θt = Cσγt ) is locally time homogeneous in a
short time interval [τ −m, τ), thus we can estimate:
σ2τ =
1
]{I}∑t∈I
R2t .
Time homogeneous: variables are constant and time independent.
1 2 3 τ −m τ − 1 T
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-9
Homogeneity test
TheoremIf R1, ...,Rτ obey the heteroscedastic model and volatilities σt satisfythe condition b ≤ σ2
t ≤ bB with b,B > 0, given a large value η:
P{|θI − θτ | >∆I (1 +ηsγ√]{I}
) + ηvI}
≤ 4√
eη(1 + logB) exp
{− η2
2aγ(1+ηsγ√]{I}
)2
}.
where ∆I measures the (local) bias: ∆2I = 1
]{I}∑
t∈I (θt − θτ )2 and
vI the volatility of θI .
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-10
Homogeneity test
Under homogeneity |θI − θτ | is bounded by ηvI provided that η issufficiently large.
|θI − θτ | < ηvI
|θI\J − θJ | < η(vI\J + vJ) = η′(
√θ2J/]{J}+
√θ2I\J/]{I\J}).
' $� �τ −m τ − 2/3m τ − 1/3m
J
J
τ − 1
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-11
Extension to the high-dimensional case
Independent component analysis (ICA)
Given xt = {x1t , · · · , xdt}>, find ICs yt :
yt = Wxt ,
where W is a d × d nonsingular matrix:
xt = W−1yt = Ayt
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-12
ICA example
IC1t : generalized hyperbolic variable GH(−0.5, 1.02,−0.11, 1, 0)IC2t : GH(−0.5, 1.29,−0.51, 1, 0)IC3t : GH(−0.5, 1.15,−0.35, 1, 0)
A = W−1 =
9 11 0.5−1 −8 212 −5 1
, xt = A
IC1t
IC2t
IC3t
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-13
ICA example
The Mahalanobis transformation:
cov(x)−1/2 =
0.19 0.22 −0.080.22 0.45 −0.14
−0.08 −0.14 0.12
6= W =
0.01 −0.04 0.080.07 0.01 −0.050.29 0.52 −0.18
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-14
Time plots of 3 independent GH variables (top), the linear transformation (middle)
and the estimated ICs (bottom).IC1: GH(-0.5,1.02,-0.11,1,0)
-4.00
0.00
4.00
8.00
250 500 750 1000
x1 = 9*y1 + 11*y2 +0.5*y3
-15.000.00
15.0030.00
250 500 750 1000
y1
-4.00
0.00
4.00
8.00
250 500 750 1000
IC2: GH(-0.5,1.66,-0.89,1,0)
-4.00
0.00
4.00
8.00
250 500 750 1000
x2 =-1*y1 - 8*y2 +2*y3
-15.000.00
15.0030.00
250 500 750 1000
y2
-4.00
0.00
4.00
8.00
250 500 750 1000
IC3: GH(-0.5,1.72,-0.95,1,0)
-4.00
0.00
4.00
8.00
250 500 750 1000
x3 = 12*y1 -5*y2 + y3
-15.000.00
15.0030.00
250 500 750 1000
y3
-4.00
0.00
4.00
8.00
250 500 750 1000
GHICAsim.xpl
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-15
What will ICA bring us?
The portfolio return Rt can be calculated as:
Rt = b>t xt (3)
= b>t W−1yt (4)
= b>t W−1D1/2t εt (5)
where covariance of yt : Dt = diag(σ21t , · · · , σ2
dt) and εt i.i.d. shocks.Marginal pdf fyj and volatility σjt of IC yj , j = 1, · · · , d can beestimated univariately.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-16
What will ICA bring us?
The portfolio return Rt can be calculated as:
Rt = b>t xt (3)
= b>t W−1yt (4)
= b>t W−1D1/2t εt (5)
� Equation (4) is ICA that gives us approximately ICs.
� Dt and distributional parameters of εt in (5) can be estimatedapplying GHADA.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Motivation 1-17
VaR estimation procedure: GHICA
1. ICA to get independent components.
2. Apply the GHADA to estimate the local constant vola and fitthe marginal pdf of εt .
3. Determine VaR at level a via MC (d samples, M observationsand N repetitions) .
VaRa,t =1
N
N∑n=1
F−1a,t (R
(n)t )
=1
N
N∑n=1
F−1a,t {b>t W−1D
1/2t ε
(n)t }
4. Perform backtesting.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Outline
1. Motivation: ICA + GHADA = GHICA X
2. ICA: properties and estimation
3. Simulation study
4. Empirical study
5. Conclusion
ICA 2-1
Definition
ICA model: y1t...
ydt
=
w11 · · · w1d
· · · · ·wd1 · · · wdd
x1t
...xdt
yt = Wxt
equivalently xt = Ayt
where xt are d-dimensional observations, yt are ICs and W thenonsingular linear transforming matrix: W−1 = A.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-2
Properties of ICA
Scale identification: the scales of the ICs are not identifiable sinceboth yt and W are unknown.
x1t =d∑
j=1
a1jyjt =d∑
j=1
{ 1
cja1j}{cjyjt}
where cj are constants.Hence: prewhiten xt by the Mahalanobis transformation cov(x)−1/2
and assume that each IC has unit variance: E[y2j ] = 1. The matrix
W is then an orthogonal matrix.From now on xt is prewhitened!
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-3
Properties of ICA
Order identification: the order of the ICs is undetermined.
xt = Ayt = AP−1Pyt
where P is a permutation matrix and Pyt are the original ICs but ina different order.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-4
Properties of ICA
ICs are necessarily non-GaussianConsider two prewhitened Gaussian ICs y1 and y2 with pdf:
f (y1, y2) = |2πI|−12 exp(−y2
1 + y22
2) =
1
2πexp(−||y ||
2
2)
where ||y || is the norm of the vector (y1, y2)>.
The joint density of the observation x1 and x2 is given by:
f (x1, x2) = |2πI|−12 exp(−||Wx ||2
2)|detW | = 1
2πexp(−||x ||
2
2).
Since A is an orthogonal matrix after prewhitening.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-5
How to find the ICs?
CLT: Let (Y1, · · · ,Yd) be a set of d independent random variablesand each has an arbitrary probability distribution, their weightedsum has a limiting cdf that approaches a normal distribution.
yt = W xt = WW−1yt
yjt = w>j xt = w>
j W−1yt = (W−1>wj)>yt
LHS: yjt is expected to be identical to one IC (one variable)RHS: weighted sum of ICs (sum)
Find the maximal nongaussianity of yjt .
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-6
How to find the ICs?
Non-Gaussianity measurement: Negentropy
J(y) = H(ygauss)− H(y)
entropy: H(y) = −∫
fy (η) log fy (η)dη.
Here ygauss is a Gaussian variable with cov(ygauss) = cov(y).Recall that for given variance Gaussian variable has maximal entropy:
H(ygauss) =1
2log |detΣ|+ d
2(1 + log 2π)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-7
How to find the ICs?
Remark: Compared to kurtosis and excess kurtosis, negentropy isless sensitive to outliers.
Negentropy calculation however requires the knowledge of the pdfof yt .
Therefore, Hyvarinen (1998) proposes approximations of negentropythat are computationally feasible:
J(y) ≈ c1[E{G1(y)}]2 + c2{E[G2(y)]− E[G2{N(0, 1)}]}2.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-8
How to find the ICs?
Negentropy approximationApproximation a: c1 = 36/(8
√3− 9) and ca
2 = 1/(2− 6/π)
J(y) ≈ c1[E{y exp(−y2/2)}]2 + ca2 [E{exp(−y2/2)} −
√1/2]2
G a1 (y) = y exp(−y2/2)
G a2 (y) = exp(−y2/2)
Approximation b: c1 = 36/(8√
3− 9) and cb2 = 24/(16
√3− 27/π)
J(y) ≈ c1[E{y exp(−y2/2)}]2 + cb2 [E{|y |} −
√2/π]2
Gb1 (y) = y exp(−y2/2)
Gb2 (y) = |y |
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-9
How to find the ICs?
Comparison of the true negentropy (black) and its approximations (a: red, b:
blue) of simulated Gaussian mixture variable: pN(0, 1) + (1 − p)N(1, 4) for p ∈[0, 1]. GHICAnegentropyapp.xpl
Negentropy comparison
0 0.5 1
p
05
10
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-10
Negentropy approximations and FastICA
Assumption on Negentropy approximation: variable is symmetric.In this sense, G1 as an odd function disappears.
J(y) ≈ C{E[G (y)]− E[G{N(0, 1)}]}2.
G (y) =1
slog cosh(sy), 1 ≤ s ≤ 2 (6)
g(y)def= G ′(y) = tanh(sy) (7)
g ′(y) = s{1− tanh2(sy)} (8)
very often, s = 1 is taken in this approximation.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-11
Negentropy approximations and FastICA
Assumption on Negentropy approximation: variable is symmetric.In this sense, G1 as an odd function disappears.
J(y) ≈ C{E[G (y)]− E[G{N(0, 1)}]}2.
G (y) = − exp(−y2/2) (9)
g(y)def= G ′(y) = y exp(−y2/2) (10)
g ′(y) = (1− y2) exp(−y2/2) (11)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-12
Negentropy approximations and FastICA
Assumption on Negentropy approximation: variable is symmetric.In this sense, G1 as an odd function disappears.
J(y) ≈ C{E[G (y)]− E[G{N(0, 1)}]}2.
G (y) =1
4y4 (12)
g(y)def= G ′(y) = y3 (13)
g ′(y) = 3y2 (14)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-13
FastICA
Objective function
{E{G (WX )} − E[G{N(0, 1)}]}E{Xg(WX )} = 0 (15)
A fast gradient method can be formulated under the constraintW>W = Id :
E{Xg(WX )}+ χW = 0 (16)
The iteration of wj with respect to yj :
w(n+1)j = E[Xg(w
(n)j X )− E{g ′(w (n)
j X )}w (n)j ] (17)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-14
FastICA
Algorithm
1. Choose an initial vector wj of unit norm, W = (w1, · · · ,wd)>.
2. Let w(n)j = E[g(w
(n−1)j x)x ]− E[g ′(w
(n−1)j x)]w
(n−1)j . In
practice, the sample mean is used to calculate E[·].3. Orthogonalization (decorrelated):
w(n)j = w
(n)j −
∑k 6=j(w
(n)>j wk)wk .
4. Normalization: w(n)j = w
(n)j /||w (n)
j ||.
5. If not converged, i.e. ||w (n)j − w
(n−1)j || 6= 0, go back to 2.
6. Set j = j + 1. For j ≤ d , go back to step 1.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
ICA 2-15
VaR estimation procedure: GHICA
1. ICA to get independent components.
2. Apply the GHADA to estimate the local constant vola and fitthe marginal pdf of εt .
3. Determine VaR at level a via MC (d samples, M observationsand N repetitions) .
VaRa,t =1
N
N∑n=1
F−1a,t (R
(n)t )
=1
N
N∑n=1
F−1a,t {b>t W−1D
1/2t ε
(n)t }
4. Perform backtesting.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Simulation study 3-1
Simulation study on 50 NIG RVs
Simulation design
� NIG distributional parameters:λ = −0.5, α ∼ U[1, 2],
δ = 1, β = arg{σ2(y) = 1√α2−β2
α2
α2−β2 = 1},
µ = 0.
� d = 50 samples, each has T = 1000 observations
� An orthogonal matrix W−1 is achieved via the Jordandecomposition of a square matrix (elements ∼ N(0, 1))
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Simulation study 3-2
Simulation study on 50 NIG RVs
Simulation procedure
� Generate 50 independent NIG random variables yt and W−1
� Calculate the input of ICA: xt = W−1yt
� Estimate yt and their distributional parameters (αj , βj , δj , µj)for j = 1, · · · , 50
Simulation goalCompare the pdfs of the generated and the estimated ICs. Thelargest mean absolute error (MAE) is 0.054.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Simulation study 3-3
Simulation study on 50 NIG RVs
Graphical comparisons of the pdfs of the estimated IC and the trueIC1 (dotted line). GHICAsimpdfcomp.xpl
ICest1
-5 0 5
X
00.
10.
20.
30.
40.
50.
6
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Simulation study 3-4
Simulation study on 50 NIG RVs
Graphical comparisons of the pdfs of the estimated ICs and the trueICs (dotted line). GHICAsimpdfcomp.xpl
ICest1
-5 0 5
X
00.
20.
40.
6
Y
ICest11
-5 0 5
X
00.
20.
40.
6
Y
ICest21
-5 0 5
X
00.
20.
40.
6
Y
ICest31
-5 0 5
X
00.
5
Y
ICest41
-5 0 5
X
00.
5
Y
ICest2
-5 0 5
X
00.
20.
40.
6
Y
ICest12
-5 0 5
X
00.
20.
40.
6
Y
ICest22
-5 0 5
X
00.
20.
40.
6
Y
ICest32
-5 0 5
X
00.
20.
40.
6
Y
ICest42
-5 0 5
X
00.
5
Y
ICest3
-5 0 5
X
00.
20.
40.
6
Y
ICest13
-5 0 5
X
00.
20.
40.
60.
8
Y
ICest23
-5 0 5
X
00.
20.
40.
6
Y
ICest33
-5 0 5
X
00.
20.
40.
6
Y
ICest43
-5 0 5
X
00.
5
Y
ICest4
-5 0 5
X0
0.2
0.4
0.6
Y
ICest14
-5 0 5
X
00.
20.
40.
6
Y
ICest24
-5 0 5
X
00.
20.
40.
6
Y
ICest34
-5 0 5
X
00.
20.
40.
6
Y
ICest44
-5 0 5
X
00.
5
Y
ICest5
-5 0 5
X
00.
5
Y
ICest15
-5 0 5
X
00.
20.
40.
6
Y
ICest25
-5 0 5
X
00.
20.
40.
6
Y
ICest35
-5 0 5
X
00.
5
Y
ICest45
-5 0 5
X
00.
5
Y
ICest6
-5 0 5
X
00.
5
Y
ICest16
-5 0 5
X
00.
20.
40.
6
Y
ICest26
-5 0 5
X
00.
20.
40.
6
YICest36
-5 0 5
X
00.
20.
40.
6
Y
ICest46
-5 0 5
X
00.
5
Y
ICest7
-5 0 5
X
00.
20.
40.
6
Y
ICest17
-5 0 5
X
00.
5
Y
ICest27
-5 0 5
X
00.
20.
40.
60.
8
Y
ICest37
-5 0 5
X
00.
20.
40.
60.
8
Y
ICest47
-5 0 5
X
00.
20.
40.
6
Y
ICest8
-5 0 5
X
00.
20.
40.
6
Y
ICest18
-5 0 5
X
00.
20.
40.
6
Y
ICest28
-5 0 5
X
00.
20.
40.
6
Y
ICest38
-5 0 5
X
00.
20.
40.
6
Y
ICest48
-5 0 5
X
00.
5
Y
ICest9
-5 0 5
X
00.
20.
40.
6
Y
ICest19
-5 0 5
X
00.
20.
40.
6
Y
ICest29
-5 0 5
X
00.
20.
40.
6
Y
ICest39
-5 0 5
X
00.
5
Y
ICest49
-5 0 5
X
00.
51
Y
ICest10
-5 0 5
X
00.
20.
40.
6
Y
ICest20
-5 0 5
X
00.
20.
40.
6
Y
ICest30
-5 0 5
X
00.
5
Y
ICest40
-5 0 5
X
00.
20.
40.
6
Y
ICest50
-5 0 5
X
00.
51
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-1
Foreign exchange rate portfolio
Data: (DEM/USD, GBP/USD) daily rates from 1979-12-01 to1994-04-01 (3720 observations), available at FEDC.Trading strategies
b1 = (1, 1)>
b2 = (1, 2)>
b3 = (−1, 2)>
b4 = (−2, 1)>
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-2
Foreign exchange rate portfolio
ICA
W =
(207.93 −213.6377.72 73.29
)A =
(2.30e − 3 6.71e − 3
−2.44e − 3 6.53e − 0
)GH parameters estimation
IC1:HYP( 1.71, −0.17, 0.55, 0.13)NIG ( 1.22, −0.18, 1.11, 0.13)
IC2:HYP( 1.78, 0.03, 0.72, −0.02)NIG ( 1.37, 0.03, 1.28, −0.03)
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-3
Foreign exchange rate portfolio
Temporal dependence: ACF plots of the log returns: DEM/USDon the left and GBP/USD on the right.GHICAfxdescriptive.xpl
ACF plot of DEM/USD
0 5 10 15 20 25 30
lag
00.5
1
acf
ACF plot of GBP/USD
0 5 10 15 20 25 30
lag
00.5
1
acf
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-4
Foreign exchange rate portfolio
Temporal dependence: ACF plots of the 2 independent series: IC1left and IC2 right. GHICAfxdescriptive.xpl
ACF plot of IC1
0 5 10 15 20 25 30
lag
00.5
1
acf
ACF plot of IC2
0 5 10 15 20 25 30
lag
00.5
1
acf
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-5
Foreign exchange rate portfolio
Linear dependenceScatterplots of the FX returns (left) and ICs (right) of the FXportfolio. GHICAfxdescriptive.xpl
FX DEM/USD & GBP/USD
-4 -2 0 2
DEM/USD*E-2
-4-2
02
4
GB
P/U
SD*E
-2
FX IC1 & IC2
-5 0 5
IC1
-50
IC2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-6
Foreign exchange rate portfolio
Joint density: Comparison of the nonparametric joint density(black) of the returns of the exchange rates and the product (blue)of the HYP marginal densities of two ICs. The red surface is theGaussian fitting with the same covariance as the returns of the ex-change rates.GHICAfxjointdensitycomp.xpl
Joint pdf of FX returns
-0.03
-0.01 0.00
0.02 0.04-0.04
-0.03-0.01
0.00 0.02
1610.16
3220.31
4830.47
6440.63
8050.79
Joint pdf of ICs
-3.99
-1.86 0.27
2.39 4.52-8.61
-5.71-2.82
0.07 2.97
0.06
0.12
0.18
0.24
0.30
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-7
Foreign exchange rate portfolio
Adaptive vola: GHICAfxdescriptive.xpl
FX Adaptive Vola of IC1
5 10 15 20 25 30 35X*E2
12
3
Y
FX Adaptive Vola of IC2
5 10 15 20 25 30 35X*E2
0.5
11.
52
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-8
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (1, 1)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.08 -0.02 0.14 1 -0.01 0.02 -0.02 0.05 -0.02 0.06 -0.02 0.09 -0.03 0.14
2 -0.01 0.01 -0.01 0.04 -0.02 0.05 -0.02 0.08 -0.02 0.13 2 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.02 0.15
· · · · · · · · · · · · · · · · · · · · · ·424 -0.03 0.07 -0.06 0.15 -0.07 0.21 -0.08 0.34 -0.09 0.50 423 -0.04 0.06 -0.07 0.15 -0.07 0.23 -0.08 0.33 -0.10 0.45
425 -0.03 0.06 -0.06 0.14 -0.07 0.22 -0.08 0.27 -0.09 0.44 425 -0.04 0.06 -0.07 0.14 -0.07 0.24 -0.08 0.28 -0.10 0.46
426 -0.03 0.04 -0.05 0.11 -0.06 0.18 -0.06 0.25 -0.08 0.38 426 -0.03 0.05 -0.06 0.12 -0.06 0.18 -0.07 0.27 -0.08 0.53
597 -0.03 0.06 -0.06 0.16 -0.06 0.22 -0.07 0.30 -0.09 0.54 607 -0.04 0.06 -0.06 0.15 -0.07 0.21 -0.08 0.32 -0.10 0.46
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.02 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.03 0.14 622 -0.03 0.07 -0.06 0.12 -0.07 0.18 -0.08 0.26 -0.09 0.48
1000 -0.01 0.02 -0.02 0.04 -0.02 0.07 -0.02 0.10 -0.03 0.16 1000 -0.01 0.02 -0.02 0.05 -0.02 0.07 -0.03 0.11 -0.03 0.18
Tab. 4: Descriptive statistics of quantile estimates: b = (1, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.13 -0.04 0.23 1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.04 0.14 -0.04 0.25
2 -0.01 0.02 -0.02 0.05 -0.03 0.08 -0.03 0.11 -0.04 0.18 2 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.12 -0.04 0.19
· · · · · · · · · · · · · · · · · · · · · ·424 -0.05 0.10 -0.09 0.23 -0.10 0.31 -0.12 0.49 -0.14 0.74 325 -0.06 0.11 -0.10 0.22 -0.11 0.30 -0.13 0.39 -0.15 0.71
425 -0.05 0.10 -0.10 0.21 -0.10 0.33 -0.12 0.40 -0.14 0.66 423 -0.06 0.11 -0.10 0.25 -0.11 0.32 -0.12 0.46 -0.15 0.75
597 -0.05 0.09 -0.09 0.24 -0.10 0.33 -0.11 0.45 -0.13 0.81 424 -0.06 0.10 -0.10 0.22 -0.11 0.32 -0.13 0.45 -0.15 0.88
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.03 -0.03 0.07 -0.03 0.08 -0.04 0.13 -0.04 0.20 597 -0.05 0.09 -0.10 0.22 -0.10 0.34 -0.12 0.46 -0.14 0.71
1000 -0.02 0.03 -0.03 0.06 -0.03 0.10 -0.04 0.14 -0.05 0.26 1000 -0.02 0.03 -0.03 0.08 -0.04 0.11 -0.04 0.16 -0.05 0.27
Tab. 5: Descriptive statistics of quantile estimates: b = (1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
19
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-9
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (1, 1)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.08 -0.02 0.14 1 -0.01 0.02 -0.02 0.05 -0.02 0.06 -0.02 0.09 -0.03 0.14
2 -0.01 0.01 -0.01 0.04 -0.02 0.05 -0.02 0.08 -0.02 0.13 2 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.02 0.15
· · · · · · · · · · · · · · · · · · · · · ·424 -0.03 0.07 -0.06 0.15 -0.07 0.21 -0.08 0.34 -0.09 0.50 423 -0.04 0.06 -0.07 0.15 -0.07 0.23 -0.08 0.33 -0.10 0.45
425 -0.03 0.06 -0.06 0.14 -0.07 0.22 -0.08 0.27 -0.09 0.44 425 -0.04 0.06 -0.07 0.14 -0.07 0.24 -0.08 0.28 -0.10 0.46
426 -0.03 0.04 -0.05 0.11 -0.06 0.18 -0.06 0.25 -0.08 0.38 426 -0.03 0.05 -0.06 0.12 -0.06 0.18 -0.07 0.27 -0.08 0.53
597 -0.03 0.06 -0.06 0.16 -0.06 0.22 -0.07 0.30 -0.09 0.54 607 -0.04 0.06 -0.06 0.15 -0.07 0.21 -0.08 0.32 -0.10 0.46
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.02 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.03 0.14 622 -0.03 0.07 -0.06 0.12 -0.07 0.18 -0.08 0.26 -0.09 0.48
1000 -0.01 0.02 -0.02 0.04 -0.02 0.07 -0.02 0.10 -0.03 0.16 1000 -0.01 0.02 -0.02 0.05 -0.02 0.07 -0.03 0.11 -0.03 0.18
Tab. 4: Descriptive statistics of quantile estimates: b = (1, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.13 -0.04 0.23 1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.04 0.14 -0.04 0.25
2 -0.01 0.02 -0.02 0.05 -0.03 0.08 -0.03 0.11 -0.04 0.18 2 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.12 -0.04 0.19
· · · · · · · · · · · · · · · · · · · · · ·424 -0.05 0.10 -0.09 0.23 -0.10 0.31 -0.12 0.49 -0.14 0.74 325 -0.06 0.11 -0.10 0.22 -0.11 0.30 -0.13 0.39 -0.15 0.71
425 -0.05 0.10 -0.10 0.21 -0.10 0.33 -0.12 0.40 -0.14 0.66 423 -0.06 0.11 -0.10 0.25 -0.11 0.32 -0.12 0.46 -0.15 0.75
597 -0.05 0.09 -0.09 0.24 -0.10 0.33 -0.11 0.45 -0.13 0.81 424 -0.06 0.10 -0.10 0.22 -0.11 0.32 -0.13 0.45 -0.15 0.88
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.03 -0.03 0.07 -0.03 0.08 -0.04 0.13 -0.04 0.20 597 -0.05 0.09 -0.10 0.22 -0.10 0.34 -0.12 0.46 -0.14 0.71
1000 -0.02 0.03 -0.03 0.06 -0.03 0.10 -0.04 0.14 -0.05 0.26 1000 -0.02 0.03 -0.03 0.08 -0.04 0.11 -0.04 0.16 -0.05 0.27
Tab. 5: Descriptive statistics of quantile estimates: b = (1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
19
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-10
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (1, 2)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.08 -0.02 0.14 1 -0.01 0.02 -0.02 0.05 -0.02 0.06 -0.02 0.09 -0.03 0.14
2 -0.01 0.01 -0.01 0.04 -0.02 0.05 -0.02 0.08 -0.02 0.13 2 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.02 0.15
· · · · · · · · · · · · · · · · · · · · · ·424 -0.03 0.07 -0.06 0.15 -0.07 0.21 -0.08 0.34 -0.09 0.50 423 -0.04 0.06 -0.07 0.15 -0.07 0.23 -0.08 0.33 -0.10 0.45
425 -0.03 0.06 -0.06 0.14 -0.07 0.22 -0.08 0.27 -0.09 0.44 425 -0.04 0.06 -0.07 0.14 -0.07 0.24 -0.08 0.28 -0.10 0.46
426 -0.03 0.04 -0.05 0.11 -0.06 0.18 -0.06 0.25 -0.08 0.38 426 -0.03 0.05 -0.06 0.12 -0.06 0.18 -0.07 0.27 -0.08 0.53
597 -0.03 0.06 -0.06 0.16 -0.06 0.22 -0.07 0.30 -0.09 0.54 607 -0.04 0.06 -0.06 0.15 -0.07 0.21 -0.08 0.32 -0.10 0.46
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.02 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.03 0.14 622 -0.03 0.07 -0.06 0.12 -0.07 0.18 -0.08 0.26 -0.09 0.48
1000 -0.01 0.02 -0.02 0.04 -0.02 0.07 -0.02 0.10 -0.03 0.16 1000 -0.01 0.02 -0.02 0.05 -0.02 0.07 -0.03 0.11 -0.03 0.18
Tab. 4: Descriptive statistics of quantile estimates: b = (1, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.13 -0.04 0.23 1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.04 0.14 -0.04 0.25
2 -0.01 0.02 -0.02 0.05 -0.03 0.08 -0.03 0.11 -0.04 0.18 2 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.12 -0.04 0.19
· · · · · · · · · · · · · · · · · · · · · ·424 -0.05 0.10 -0.09 0.23 -0.10 0.31 -0.12 0.49 -0.14 0.74 325 -0.06 0.11 -0.10 0.22 -0.11 0.30 -0.13 0.39 -0.15 0.71
425 -0.05 0.10 -0.10 0.21 -0.10 0.33 -0.12 0.40 -0.14 0.66 423 -0.06 0.11 -0.10 0.25 -0.11 0.32 -0.12 0.46 -0.15 0.75
597 -0.05 0.09 -0.09 0.24 -0.10 0.33 -0.11 0.45 -0.13 0.81 424 -0.06 0.10 -0.10 0.22 -0.11 0.32 -0.13 0.45 -0.15 0.88
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.03 -0.03 0.07 -0.03 0.08 -0.04 0.13 -0.04 0.20 597 -0.05 0.09 -0.10 0.22 -0.10 0.34 -0.12 0.46 -0.14 0.71
1000 -0.02 0.03 -0.03 0.06 -0.03 0.10 -0.04 0.14 -0.05 0.26 1000 -0.02 0.03 -0.03 0.08 -0.04 0.11 -0.04 0.16 -0.05 0.27
Tab. 5: Descriptive statistics of quantile estimates: b = (1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
19
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-11
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (1, 2)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.08 -0.02 0.14 1 -0.01 0.02 -0.02 0.05 -0.02 0.06 -0.02 0.09 -0.03 0.14
2 -0.01 0.01 -0.01 0.04 -0.02 0.05 -0.02 0.08 -0.02 0.13 2 -0.01 0.01 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.02 0.15
· · · · · · · · · · · · · · · · · · · · · ·424 -0.03 0.07 -0.06 0.15 -0.07 0.21 -0.08 0.34 -0.09 0.50 423 -0.04 0.06 -0.07 0.15 -0.07 0.23 -0.08 0.33 -0.10 0.45
425 -0.03 0.06 -0.06 0.14 -0.07 0.22 -0.08 0.27 -0.09 0.44 425 -0.04 0.06 -0.07 0.14 -0.07 0.24 -0.08 0.28 -0.10 0.46
426 -0.03 0.04 -0.05 0.11 -0.06 0.18 -0.06 0.25 -0.08 0.38 426 -0.03 0.05 -0.06 0.12 -0.06 0.18 -0.07 0.27 -0.08 0.53
597 -0.03 0.06 -0.06 0.16 -0.06 0.22 -0.07 0.30 -0.09 0.54 607 -0.04 0.06 -0.06 0.15 -0.07 0.21 -0.08 0.32 -0.10 0.46
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.02 -0.02 0.04 -0.02 0.06 -0.02 0.09 -0.03 0.14 622 -0.03 0.07 -0.06 0.12 -0.07 0.18 -0.08 0.26 -0.09 0.48
1000 -0.01 0.02 -0.02 0.04 -0.02 0.07 -0.02 0.10 -0.03 0.16 1000 -0.01 0.02 -0.02 0.05 -0.02 0.07 -0.03 0.11 -0.03 0.18
Tab. 4: Descriptive statistics of quantile estimates: b = (1, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.13 -0.04 0.23 1 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.04 0.14 -0.04 0.25
2 -0.01 0.02 -0.02 0.05 -0.03 0.08 -0.03 0.11 -0.04 0.18 2 -0.01 0.02 -0.03 0.06 -0.03 0.09 -0.03 0.12 -0.04 0.19
· · · · · · · · · · · · · · · · · · · · · ·424 -0.05 0.10 -0.09 0.23 -0.10 0.31 -0.12 0.49 -0.14 0.74 325 -0.06 0.11 -0.10 0.22 -0.11 0.30 -0.13 0.39 -0.15 0.71
425 -0.05 0.10 -0.10 0.21 -0.10 0.33 -0.12 0.40 -0.14 0.66 423 -0.06 0.11 -0.10 0.25 -0.11 0.32 -0.12 0.46 -0.15 0.75
597 -0.05 0.09 -0.09 0.24 -0.10 0.33 -0.11 0.45 -0.13 0.81 424 -0.06 0.10 -0.10 0.22 -0.11 0.32 -0.13 0.45 -0.15 0.88
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.03 -0.03 0.07 -0.03 0.08 -0.04 0.13 -0.04 0.20 597 -0.05 0.09 -0.10 0.22 -0.10 0.34 -0.12 0.46 -0.14 0.71
1000 -0.02 0.03 -0.03 0.06 -0.03 0.10 -0.04 0.14 -0.05 0.26 1000 -0.02 0.03 -0.03 0.08 -0.04 0.11 -0.04 0.16 -0.05 0.27
Tab. 5: Descriptive statistics of quantile estimates: b = (1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
19
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-12
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (−1, 2)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.07 -0.02 0.10
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.12 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.13
· · · · · · · · · · · · · · · · · · · · · ·601 -0.03 0.05 -0.06 0.15 -0.06 0.22 -0.07 0.28 -0.08 0.41 601 -0.03 0.05 -0.06 0.15 -0.07 0.19 -0.08 0.30 -0.09 0.48
607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.26 -0.09 0.47 606 -0.03 0.06 -0.06 0.15 -0.07 0.23 -0.08 0.30 -0.09 0.55
611 -0.03 0.06 -0.06 0.13 -0.06 0.20 -0.07 0.28 -0.08 0.43 607 -0.04 0.07 -0.07 0.13 -0.07 0.19 -0.08 0.26 -0.10 0.48
613 -0.03 0.05 -0.06 0.15 -0.07 0.20 -0.08 0.27 -0.09 0.39 613 -0.04 0.06 -0.06 0.12 -0.07 0.18 -0.08 0.27 -0.09 0.49
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 1000 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.10
Tab. 6: Descriptive statistics of quantile estimates: b = (−1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11 1 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.07 -0.02 0.11
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.06 -0.02 0.10 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.12
· · · · · · · · · · · · · · · · · · · · · ·607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.25 -0.09 0.41 601 -0.03 0.06 -0.06 0.15 -0.07 0.22 -0.08 0.30 -0.09 0.55
608 -0.03 0.05 -0.06 0.13 -0.06 0.19 -0.07 0.28 -0.09 0.40 605 -0.04 0.06 -0.06 0.14 -0.07 0.19 -0.08 0.30 -0.09 0.48
612 -0.03 0.05 -0.06 0.14 -0.07 0.21 -0.08 0.27 -0.09 0.47 612 -0.04 0.07 -0.07 0.14 -0.07 0.19 -0.08 0.27 -0.09 0.38
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 999 -0.01 0.01 -0.01 0.02 -0.01 0.04 -0.02 0.05 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1000 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.06 -0.02 0.10
Tab. 7: Descriptive statistics of quantile estimates: b = (−2, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
20
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-13
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (−1, 2)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.07 -0.02 0.10
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.12 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.13
· · · · · · · · · · · · · · · · · · · · · ·601 -0.03 0.05 -0.06 0.15 -0.06 0.22 -0.07 0.28 -0.08 0.41 601 -0.03 0.05 -0.06 0.15 -0.07 0.19 -0.08 0.30 -0.09 0.48
607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.26 -0.09 0.47 606 -0.03 0.06 -0.06 0.15 -0.07 0.23 -0.08 0.30 -0.09 0.55
611 -0.03 0.06 -0.06 0.13 -0.06 0.20 -0.07 0.28 -0.08 0.43 607 -0.04 0.07 -0.07 0.13 -0.07 0.19 -0.08 0.26 -0.10 0.48
613 -0.03 0.05 -0.06 0.15 -0.07 0.20 -0.08 0.27 -0.09 0.39 613 -0.04 0.06 -0.06 0.12 -0.07 0.18 -0.08 0.27 -0.09 0.49
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 1000 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.10
Tab. 6: Descriptive statistics of quantile estimates: b = (−1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11 1 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.07 -0.02 0.11
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.06 -0.02 0.10 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.12
· · · · · · · · · · · · · · · · · · · · · ·607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.25 -0.09 0.41 601 -0.03 0.06 -0.06 0.15 -0.07 0.22 -0.08 0.30 -0.09 0.55
608 -0.03 0.05 -0.06 0.13 -0.06 0.19 -0.07 0.28 -0.09 0.40 605 -0.04 0.06 -0.06 0.14 -0.07 0.19 -0.08 0.30 -0.09 0.48
612 -0.03 0.05 -0.06 0.14 -0.07 0.21 -0.08 0.27 -0.09 0.47 612 -0.04 0.07 -0.07 0.14 -0.07 0.19 -0.08 0.27 -0.09 0.38
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 999 -0.01 0.01 -0.01 0.02 -0.01 0.04 -0.02 0.05 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1000 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.06 -0.02 0.10
Tab. 7: Descriptive statistics of quantile estimates: b = (−2, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
20
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-14
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (−2, 1)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.07 -0.02 0.10
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.12 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.13
· · · · · · · · · · · · · · · · · · · · · ·601 -0.03 0.05 -0.06 0.15 -0.06 0.22 -0.07 0.28 -0.08 0.41 601 -0.03 0.05 -0.06 0.15 -0.07 0.19 -0.08 0.30 -0.09 0.48
607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.26 -0.09 0.47 606 -0.03 0.06 -0.06 0.15 -0.07 0.23 -0.08 0.30 -0.09 0.55
611 -0.03 0.06 -0.06 0.13 -0.06 0.20 -0.07 0.28 -0.08 0.43 607 -0.04 0.07 -0.07 0.13 -0.07 0.19 -0.08 0.26 -0.10 0.48
613 -0.03 0.05 -0.06 0.15 -0.07 0.20 -0.08 0.27 -0.09 0.39 613 -0.04 0.06 -0.06 0.12 -0.07 0.18 -0.08 0.27 -0.09 0.49
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 1000 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.10
Tab. 6: Descriptive statistics of quantile estimates: b = (−1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11 1 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.07 -0.02 0.11
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.06 -0.02 0.10 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.12
· · · · · · · · · · · · · · · · · · · · · ·607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.25 -0.09 0.41 601 -0.03 0.06 -0.06 0.15 -0.07 0.22 -0.08 0.30 -0.09 0.55
608 -0.03 0.05 -0.06 0.13 -0.06 0.19 -0.07 0.28 -0.09 0.40 605 -0.04 0.06 -0.06 0.14 -0.07 0.19 -0.08 0.30 -0.09 0.48
612 -0.03 0.05 -0.06 0.14 -0.07 0.21 -0.08 0.27 -0.09 0.47 612 -0.04 0.07 -0.07 0.14 -0.07 0.19 -0.08 0.27 -0.09 0.38
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 999 -0.01 0.01 -0.01 0.02 -0.01 0.04 -0.02 0.05 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1000 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.06 -0.02 0.10
Tab. 7: Descriptive statistics of quantile estimates: b = (−2, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
20
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-15
Foreign exchange rate portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: b = (−2, 1)>, MC simulation with M = 10000, N = 100,T = 1000. GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.07 -0.02 0.10
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.12 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.13
· · · · · · · · · · · · · · · · · · · · · ·601 -0.03 0.05 -0.06 0.15 -0.06 0.22 -0.07 0.28 -0.08 0.41 601 -0.03 0.05 -0.06 0.15 -0.07 0.19 -0.08 0.30 -0.09 0.48
607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.26 -0.09 0.47 606 -0.03 0.06 -0.06 0.15 -0.07 0.23 -0.08 0.30 -0.09 0.55
611 -0.03 0.06 -0.06 0.13 -0.06 0.20 -0.07 0.28 -0.08 0.43 607 -0.04 0.07 -0.07 0.13 -0.07 0.19 -0.08 0.26 -0.10 0.48
613 -0.03 0.05 -0.06 0.15 -0.07 0.20 -0.08 0.27 -0.09 0.39 613 -0.04 0.06 -0.06 0.12 -0.07 0.18 -0.08 0.27 -0.09 0.49
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.09 1000 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.07 -0.02 0.10
Tab. 6: Descriptive statistics of quantile estimates: b = (−1, 2)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.07 -0.02 0.11 1 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.07 -0.02 0.11
2 -0.01 0.01 -0.01 0.03 -0.01 0.05 -0.02 0.06 -0.02 0.10 2 -0.01 0.01 -0.01 0.03 -0.02 0.05 -0.02 0.08 -0.02 0.12
· · · · · · · · · · · · · · · · · · · · · ·607 -0.04 0.06 -0.06 0.13 -0.07 0.18 -0.08 0.25 -0.09 0.41 601 -0.03 0.06 -0.06 0.15 -0.07 0.22 -0.08 0.30 -0.09 0.55
608 -0.03 0.05 -0.06 0.13 -0.06 0.19 -0.07 0.28 -0.09 0.40 605 -0.04 0.06 -0.06 0.14 -0.07 0.19 -0.08 0.30 -0.09 0.48
612 -0.03 0.05 -0.06 0.14 -0.07 0.21 -0.08 0.27 -0.09 0.47 612 -0.04 0.07 -0.07 0.14 -0.07 0.19 -0.08 0.27 -0.09 0.38
· · · · · · · · · · · · · · · · · · · · · ·999 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 999 -0.01 0.01 -0.01 0.02 -0.01 0.04 -0.02 0.05 -0.02 0.11
1000 -0.01 0.01 -0.01 0.03 -0.01 0.04 -0.02 0.06 -0.02 0.10 1000 -0.01 0.01 -0.01 0.03 -0.02 0.04 -0.02 0.06 -0.02 0.10
Tab. 7: Descriptive statistics of quantile estimates: b = (−2, 1)>, MC simulation with M = 10000, N = 100, T = 1000.GHICAmcqstat.xpl
20
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-16
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0500)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-17
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0100)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-18
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0050)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-19
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0025)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-20
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0010)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-21
Backtesting
LR1 is the likelihood ratio w.r.t. the risk level test and LR2 w.r.t. the independence
test. * indicates the rejection at 99% level.
GHICA (HYP fit) GHICA (NIG fit) RiskMetrics Method T-deGARCH Methodb> a(10−2) n/T% LR1 LR2 n/T% LR1 LR2 n/T% LR1 LR2 n/T% LR1 LR2
(1,1) 5 6.7 5.52 *27.60 5.8 1.28 *28.82 10.8 *53.95 *19.84 11.0 *57.33 *26.05(1,1) 1 1.9 6.47 *16.29 1.4 1.43 *11.14 5.5 *99.59 *21.70 5.3 *92.67 *22.37(1,1) 0.5 0.9 2.59 5.58 0.7 0.71 0.00 5.0 *142.33 *23.44 4.2 *106.16 *18.96(1,1) 0.25 0.4 0.76 0.00 0.3 0.09 0.00 3.8 *137.10 *22.04 2.5 *70.64 *26.70(1,1) 0.1 0.2 0.77 0.00 0.1 0.00 0.00 2.3 *100.72 *27.74 1.2 *37.75 0.00
(1,2) 5 6.8 6.16 *27.23 5.4 0.32 *30.50 10.9 *55.63 *24.78 11.0 *57.33 *24.48(1,2) 1 1.6 3.07 *10.51 1.4 1.43 *11.14 5.3 *92.67 *22.37 4.9 *79.30 *23.81(1,2) 0.5 0.9 2.59 0.00 0.8 1.52 0.00 4.7 *128.43 *21.77 4.4 *114.93 *22.83(1,2) 0.25 0.5 1.93 0.00 0.5 1.93 0.00 4.0 *148.23 *24.40 2.8 *84.94 *22.67(1,2) 0.1 0.3 2.59 0.00 0.2 0.77 0.00 2.5 *113.53 *18.05 1.5 *53.43 *7.11
(-1,2) 5 6.4 3.80 *38.56 5.5 0.51 *28.17 12.0 *75.40 *35.66 12.0 *75.40 *35.66(-1,2) 1 1.4 1.43 *11.14 1.2 0.37 *11.86 6.0 *117.58 *25.86 5.5 *99.59 *23.40(-1,2) 0.5 1.2 *7.06 *11.86 1.1 5.38 *15.84 4.4 *114.93 *18.33 3.2 *65.54 *20.90(-1,2) 0.25 1.0 *12.78 *7.97 0.7 5.43 *8.76 3.1 *99.91 *21.26 2.5 *70.64 *23.81(-1,2) 0.1 0.6 *11.52 *9.11 0.3 2.59 *10.80 2.5 *113.53 *23.81 1.7 *64.58 *28.58
(-2,1) 5 6.5 4.34 *22.35 5.0 0.00 *20.34 12.7 *89.18 *18.41 12.7 *89.18 *18.41(-2,1) 1 1.6 3.07 0.00 1.1 0.09 0.00 5.6 *103.12 *14.79 4.7 *72.87 4.49(-2,1) 0.5 0.9 2.59 0.00 0.8 1.52 0.00 4.2 *106.16 2.40 3.2 *65.54 5.54(-2,1) 0.25 0.5 1.93 0.00 0.3 0.09 0.00 3.2 *105.05 5.54 2.3 *61.50 6.22(-2,1) 0.1 0.1 0.00 0.00 0.1 0.00 0.00 2.3 *100.72 6.22 1.6 *58.94 0.00
Tab. 8: Backtesting of the VaR forecast of the exchange portfolios: MC simulation with M = 10000, N = 100, T = 1000. * indicates themodel is rejected at 99% level.
GHICAfxVaRplot.xpl
22
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-22
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0500)
5 10
X*E2
-20
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-23
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0100)
5 10
X*E2
-20
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-24
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0050)
5 10
X*E2
-20
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-25
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0025)
5 10
X*E2
-20
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-26
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(16)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0010)
5 10
X*E2
-20
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-27
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(13)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0500)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-28
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(13)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0100)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-29
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(13)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0050)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-30
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(13)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0025)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-31
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−1, 2)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(13)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0010)
5 10
X*E2
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-32
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−2, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(14)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0500)
5 10
X*E2
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-33
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−2, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(14)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0100)
5 10
X*E2
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-34
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−2, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(14)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0050)
5 10
X*E2
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-35
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−2, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(14)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0025)
5 10
X*E2
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-36
VaR forecasts of exchange rate portfolio (DEM/USD and GBP/USD) for 1000 days to
1994-04-01, b = (−2, 1)>. VaR forecasts and exceedances of the GHICA methodology
are marked in blue while those based on the RiskMetrics and T(14)-deGARCH are
marked in red and green respectively. GHICAfxVaRplot.xplVaR plot (hyp-alpha = 0.0010)
5 10
X*E2
-15
-10
-50
5
Y*E
-2
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-37
German stock portfolio
Data: 20 components (ALLIANZ, BASF, BAYER, BMW,COBANK, DAIMLER, DEUTSCHE BANK, DEGUSSA, DRESD-NER, HOECHST, KARSTADT, LINDE, MAN, MANNESMANN,PREUSSAG, RWE, SCHERING, SIEMENS, THYSSEN, VOLKSW)5748 observations from 1974-01-02 to 1996-12-30. Downloadedfrom MD*Base.Trading strategy: b = (1, · · · , 1)> ∈ IR20
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-38
German stock portfolio
Descriptive statistics: GHICASFMdescriptive.xpl
Returns 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Mean 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Variance 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Skewness 0.07 -0.17 0.04 -0.01 -0.24 0.03 -0.30 -0.39 0.12 -0.12 -0.40 -0.24 -0.59 -0.27 0.13 -0.19 -0.03 -0.54 -0.05 -0.32
Kurtosis 32.40 8.65 9.60 17.02 10.03 26.67 13.77 19.12 8.82 9.98 20.43 14.56 18.03 13.69 10.34 16.72 9.57 10.30 6.10 10.23
ρ1 -0.06 -0.02 -0.02 -0.01 0.04 -0.02 0.03 0.02 0.06 0.01 0.00 -0.03 0.02 0.05 0.08 0.04 0.06 0.06 0.04 0.05
ρ2 -0.02 -0.04 -0.04 -0.01 -0.04 -0.01 -0.03 -0.01 -0.03 -0.04 0.00 -0.01 -0.01 -0.01 -0.05 -0.02 -0.01 -0.01 -0.05 -0.01
ICs 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Mean 0.01 -0.00 0.01 0.01 0.00 0.00 0.00 0.01 -0.00 0.01 0.01 -0.00 -0.01 -0.03 -0.00 -0.01 -0.00 -0.02 -0.00 0.00
Variance 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Skewness -1.44 -0.81 0.71 0.8 1.01 -0.57 0.68 0.38 -0.84 0.48 0.52 -0.04 -0.04 -0.24 -0.17 -0.51 -0.05 -0.13 0.06 -0.32
Kurtosis 59.71 46.51 57.51 43.72 81.20 13.92 25.31 27.28 14.84 14.30 14.40 9.12 11.89 6.26 6.94 9.34 7.08 6.48 5.62 6.15
ρ1 0.00 -0.26 -0.26 -0.11 -0.12 0.10 -0.02 -0.13 -0.01 0.06 -0.12 0.08 0.04 0.04 0.08 0.08 0.08 0.05 -0.01 0.07
ρ2 -0.01 0.02 0.02 0.00 0.00 0.01 0.00 -0.01 -0.01 0.00 -0.01 0.01 0.02 -0.01 0.00 -0.03 0.02 0.00 0.00 0.00
Tab. 9: Descriptive statistics of the 20-dimensional German stock portfolio.GHICAsfmdescriptive.xpl
ICs 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
HYP α 2.01 1.54 1.63 1.63 1.59 2.19 1.64 1.57 1.59 1.83 1.78 1.89 1.96 1.76 1.70 1.95 1.76 1.85 1.91 1.79
HYP β 0.07 -0.04 0.07 0.11 0.13 0.09 0.14 0.02 -0.10 0.04 0.07 0.09 -0.00 -0.12 -0.11 -0.21 -0.03 -0.07 -0.00 -0.20
HYP δ 1.12 0.42 0.38 0.35 0.30 1.34 0.38 0.30 0.41 0.77 0.61 0.91 0.96 0.72 0.53 0.91 0.65 0.81 0.90 0.72
HYP µ -0.05 0.02 -0.04 -0.10 -0.12 -0.08 -0.12 -0.01 0.10 -0.03 -0.06 -0.09 -0.01 0.09 0.11 0.17 0.02 0.04 -0.00 0.20
NIG α 1.63 0.97 0.98 0.95 0.90 1.83 0.95 0.90 1.00 1.34 1.21 1.45 1.54 1.29 1.15 1.55 1.25 1.35 1.44 1.34
NIG β 0.05 -0.03 0.06 0.10 0.11 0.06 0.12 0.02 -0.09 0.04 0.07 0.06 0.00 -0.10 -0.10 -0.22 -0.02 -0.06 -0.00 -0.19
NIG δ 1.56 1.00 0.90 0.86 0.83 1.73 0.86 0.85 0.97 1.24 1.07 1.37 1.42 1.23 1.05 1.38 1.15 1.26 1.34 1.23
NIG µ -0.03 0.02 -0.04 -0.09 -0.10 -0.04 -0.11 -0.01 0.09 -0.03 -0.06 -0.06 -0.02 0.07 0.10 0.18 0.01 0.02 -0.00 0.19
Tab. 10: HYP and NIG parameters estimates of the German stock portfolio.GHICAsfmdescriptive.xpl
25
Linear dependence: the smallest correlation (ALLIANZ -PREUSSAG) is 0.37
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-39
German stock portfolio
ICA and GH: HYP and NIG parameters estimated of the ICs onthe basis of the German stock portfolio.GHICASFMdescriptive.xpl
ICs 1 2 3 4 5 6 7 8 9 10 · · ·HYP α 2.01 1.54 1.63 1.63 1.59 2.19 1.64 1.57 1.59 1.83 · · ·HYP β 0.07 -0.04 0.07 0.11 0.13 0.09 0.14 0.02 -0.10 0.04 · · ·HYP δ 1.12 0.42 0.38 0.35 0.30 1.34 0.38 0.30 0.41 0.77 · · ·HYP µ -0.05 0.02 -0.04 -0.10 -0.12 -0.08 -0.12 -0.01 0.10 -0.03 · · ·NIG α 1.63 0.97 0.98 0.95 0.90 1.83 0.95 0.90 1.00 1.34 · · ·NIG β 0.05 -0.03 0.06 0.10 0.11 0.06 0.12 0.02 -0.09 0.04 · · ·NIG δ 1.56 1.00 0.90 0.86 0.83 1.73 0.86 0.85 0.97 1.24 · · ·NIG µ -0.03 0.02 -0.04 -0.09 -0.10 -0.04 -0.11 -0.01 0.09 -0.03 · · ·
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-40
German stock portfolio
Density estimations: pdfs of the 20 ICs on the basis of the Germanstock portfolio. The HYP fit is displayed as a straight curve and thenonparametric density estimation is plotted by a dotted curve.GHICASFMdescriptive.xpl
Empirical density & hyp fitting - IC1
-5 0
X
00.1
0.20.3
0.4
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-41
German stock portfolio
Density estimation: pdfs of the 20 ICs on the basis of the Germanstock portfolio. The NIG fit is displayed as a straight curve and thenonparametric density estimation is plotted by a dotted curve.GHICASFMdescriptive.xpl
Empirical density & nig fitting - IC1
-5 0
X
00.1
0.20.3
0.4
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-42
German stock portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: MC simulation with M = 10000, N = 100, T = 1000.GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.28 0.36 -0.43 0.66 -0.45 0.92 -0.50 1.33 -0.56 1.91 1 -0.30 0.42 -0.47 0.82 -0.49 1.13 -0.55 1.55 -0.62 2.15
2 -0.27 0.35 -0.41 0.76 -0.43 0.93 -0.48 1.25 -0.54 2.04 2 -0.29 0.40 -0.44 0.79 -0.47 0.93 -0.52 1.35 -0.59 1.96
· · · · · · · · · · · · · · · · · · · · · ·993 -0.38 0.48 -0.59 1.11 -0.63 1.41 -0.70 1.98 -0.80 3.34 993 -0.40 0.58 -0.64 1.17 -0.68 1.55 -0.76 2.15 -0.87 3.75
· · · · · · · · · · · · · · · · · · · · · ·997 -0.4 0.54 -0.64 1.08 -0.67 1.58 -0.75 2.15 -0.84 3.18 997 -0.44 0.71 -0.69 1.55 -0.73 2.14 -0.81 2.61 -0.92 3.42
998 -0.38 0.55 -0.59 0.96 -0.63 1.56 -0.70 2.08 -0.79 3.22 998 -0.41 0.59 -0.64 1.20 -0.68 1.58 -0.76 1.91 -0.86 3.49
999 -0.32 0.46 -0.49 0.89 -0.52 1.39 -0.58 1.90 -0.65 2.75 999 -0.34 0.50 -0.53 0.96 -0.56 1.18 -0.62 1.71 -0.70 2.42
1000 -0.35 0.49 -0.54 0.94 -0.57 1.33 -0.63 1.75 -0.72 2.89 1000 -0.38 0.53 -0.59 0.99 -0.62 1.38 -0.69 1.90 -0.78 2.87
Tab. 11: Descriptive statistics of quantile estimates: MC simulation with M = 10000, N = 100, T = 1000.GHICAstatMCq.xpl
28
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-43
German stock portfolio
MC quantile estimation: Descriptive statistics of quantile esti-mates: MC simulation with M = 10000, N = 100, T = 1000.GHICAstatMCq.xpl
HYP a 5% a 1% a 0.5% a 0.25% a 0.1% NIG a 5% a 1% a 0.5% a 0.25% a 0.1%
Day mean STD% mean STD% mean STD% mean STD% mean STD% Day mean STD% mean STD% mean STD% mean STD% mean STD%
1 -0.28 0.36 -0.43 0.66 -0.45 0.92 -0.50 1.33 -0.56 1.91 1 -0.30 0.42 -0.47 0.82 -0.49 1.13 -0.55 1.55 -0.62 2.15
2 -0.27 0.35 -0.41 0.76 -0.43 0.93 -0.48 1.25 -0.54 2.04 2 -0.29 0.40 -0.44 0.79 -0.47 0.93 -0.52 1.35 -0.59 1.96
· · · · · · · · · · · · · · · · · · · · · ·993 -0.38 0.48 -0.59 1.11 -0.63 1.41 -0.70 1.98 -0.80 3.34 993 -0.40 0.58 -0.64 1.17 -0.68 1.55 -0.76 2.15 -0.87 3.75
· · · · · · · · · · · · · · · · · · · · · ·997 -0.4 0.54 -0.64 1.08 -0.67 1.58 -0.75 2.15 -0.84 3.18 997 -0.44 0.71 -0.69 1.55 -0.73 2.14 -0.81 2.61 -0.92 3.42
998 -0.38 0.55 -0.59 0.96 -0.63 1.56 -0.70 2.08 -0.79 3.22 998 -0.41 0.59 -0.64 1.20 -0.68 1.58 -0.76 1.91 -0.86 3.49
999 -0.32 0.46 -0.49 0.89 -0.52 1.39 -0.58 1.90 -0.65 2.75 999 -0.34 0.50 -0.53 0.96 -0.56 1.18 -0.62 1.71 -0.70 2.42
1000 -0.35 0.49 -0.54 0.94 -0.57 1.33 -0.63 1.75 -0.72 2.89 1000 -0.38 0.53 -0.59 0.99 -0.62 1.38 -0.69 1.90 -0.78 2.87
Tab. 11: Descriptive statistics of quantile estimates: MC simulation with M = 10000, N = 100, T = 1000.GHICAstatMCq.xpl
28
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-44
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGRACH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (hyp-a =0.0500)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-45
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGRACH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (hyp-a =0.0100)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-46
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGRACH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (hyp-a =0.0050)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-47
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGRACH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (hyp-a =0.0025)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-48
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGRACH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (hyp-a =0.0010)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-49
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGARCH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (nig-a =0.0500)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-50
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGARCH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (nig-a =0.0100)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-51
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGARCH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (nig-a =0.0050)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-52
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGARCH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (nig-a =0.0025)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-53
VaR time plots of the German stock portfolio with the equal weights. The dots are
the real portfolio returns, the RiskMetrics (red) and T(19)-deGARCH (green) results
are plotted as a dotted line and the GHICA as a straight line. The exceedances w.r.t.
the risk management models are displayed as cross on the bottom.
GHICASFMVaRplot.xplVaR plot (nig-a =0.0010)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-54
Backtesting
LR1 is the likelihood ratio w.r.t. the risk level test and LR2 w.r.t. the independence
test. * indicates the rejection at 99% level.
a = 5% a = 1% a = 0.5% a = 0.25% a = 0.1%n/T% LR1 LR2 n/T% LR1 LR2 n/T% LR1 LR2 n/T% LR1 LR2 n/T% LR1 LR2
HYP 4.9 0.02 *16.76 1.7 4.09 0.13 1.2 *7.06 0.04 0.8 *7.64 0.03 0.7 *15.27 0.00NIG 3.8 3.29 *7.14 1.4 1.43 0.05 0.8 1.52 0.03 0.7 5.43 0.00 0.4 5.09 0.00RiskMetr. 10.4 *47.46 *18.29 5.4 *96.11 *7.28 3.5 *77.12 3.92 2.8 *84.94 0.42 2.0 *82.19 0.15T4-deGARCG 12.7 *89.18 *12.28 3.2 *30.93 4.27 1.7 *17.75 0.13 1.2 *18.73 0.04 0.2 0.77 0.00T5-deGARCH 12.2 *79.24 *14.25 3.3 *33.33 4.15 2.0 *25.67 0.15 1.3 *21.97 0.05 0.6 *11.52 0.00T6-deGARCH 11.7 *69.77 *16.40 3.5 *38.33 3.92 2.1 *28.53 0.16 1.3 *21.97 0.05 0.9 *23.61 0.00T7-deGARCH 11.4 *64.32 *13.82 3.7 *43.56 6.50 2.2 *31.48 0.17 1.4 *25.37 0.10 1.1 *32.85 0.04T8-deGARCH 11.2 *60.78 *15.24 4.1 *54.68 5.88 2.4 *37.65 0.18 1.6 *32.58 0.12 1.2 *37.75 0.04T9-deGARCH 11.0 *57.33 *15.71 4.2 *57.59 5.74 2.5 *40.87 0.19 1.7 *36.38 0.13 1.3 *42.83 0.05T14-deGARCH 10.6 *50.66 *16.70 4.5 *66.61 5.34 2.9 *54.53 0.44 2.0 *48.48 0.15 1.3 *42.83 0.05T19-deGARCH 10.6 *50.66 *16.70 4.9 *79.30 *8.04 2.9 *54.53 0.44 2.1 *52.73 0.16 1.5 *53.43 0.11T24-deGARCH 10.6 *50.66 *16.70 4.9 *79.30 *8.04 3.2 *65.54 4.27 2.2 *57.07 0.17 1.6 *58.94 0.12
Tab. 13: Backtesting of the VaR forecasts of the German stock portfolios: MC simulation with M = 10000, N = 100, T = 1000. *indicates the rejection at 99% level.
GHICASFMVaRplot.xpl
31
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Empirical Study 4-55
Backtesting
LR1 is the likelihood ratio w.r.t. the risk level test and LR2 w.r.t. the independence
test. * indicates the rejection at 99% level.
T(4)-deGARCH (green).
VaR plot (nig-a =0.0010)
5 10
X*E2
-1.5
-1-0
.50
0.5
Y
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
Conclusion 5-1
Conclusion and Outlook
� GHICA X
� Advanced ICA 1:Gaussian ICs (∈ IRG ) + non-Gaussian ICs (∈ IRNG ) withG >> NG
� Advanced ICA 2:Localization of ICA: yt = Wtxt
� Further GHICA application: Portfolio including bond andderivatives
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-1
Reference
Back, A. and Weigend, A.A first application of independent component analysis toextracting structure from stock returns.International Journal of Neural Systems, 8: 473-484, 1998
Barndorff-Nielsen, O.Exponentially decreasing distributions for the logarithm ofparticle size.Proceedings of the Royal Society of London, A 353: 401-419,1977.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-2
Reference
Barndorff-Nielsen, O. and Blæsild, P.Hyperbolic distribution and ramifications: Contributions totheory and applications.Statistical Distributions in Scientific Work, C. Taillie, P. Patiland A. Baldessari (eds).Vol. 4, D. Reidel, pp. 19-44, 1981.
Barndorff-Nielsen, O. and and Shephard, N.Modelling by levy processes for financial econometrics.Levy Processes : Theory and Applications, Barndorff-Nielsen,T. Mikosch and S. Resnik (eds).Birkhauser Boston, 2001.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-3
Reference
Chen, Y., Hardle, W. and Jeong, S.Nonparametric risk management with generalized hyperbolicdistributions.SFB 649, discussion paper, 2005
Christoffersen, P. F.Evaluating interval forecast.International Economic Review, 39: 841-862, 1998.
Cover, T. and Thomas, J.Elements of Information TheoryJohn Wiley & Sons, 1991.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-4
Reference
Duann, J., Jung, T., Kuo, W., Yeh, T., Makeig, S., Hsieh, J.and Sejnowski, T.Single-trial variability in event-related bold signals.NeuroImage, 15: 823-835, 2002.
Eberlein, E. and Keller, U.Hyperbolic distributions in finance.Bernoulli, 1: 281-299, 1995.
Franke, J., Hardle, W. and Hafner, C.Statistics of Financial MarketsSpringer- Verlag Berlin Heidelberg New York, 2004.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-5
Reference
Hardle, W., Kleinow, T. and Stahl, G.Applied Quantitative FinanceSpringer- Verlag Berlin Heidelberg New York, 2002.
Hyvarinen, A.New Approximations of Differential Entropy for IndependentComponent Analysis and Projection PursuitMIT Press, pp. 273-279, 1998.
Hyvarinen, A. and Oja, E.Independent component analysis: Algorithms and applicationsNeural Networks, 13: 411-430, 1999.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-6
Reference
Hyvarinen, A., Karhunen, J. and Oja, E.Independent Component AnalysisJohn Wiley & Sons, Inc. 2001.
Jaschke, S. and Jiang, Y.Approximating value at risk in conditional gaussian modelsApplied Quantitative Finance, W. Hardle, T. Kleinow and G.Stahl (eds).Springer Verlag, 2002.
Jorion, P.Value at RiskMcGraw-Hill, 2001.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy
References 6-7
Reference
Mercurio, D. and Spokoiny, V.Statistical inference for time inhomogeneous volatility models.The Annals of Statistics, 32: 577-602, 2004.
GHICA
Distribution comparison
-5 0 5
X
00.
10.
20.
30.
40.
5
Y
NIGLaplaceNormalt(5)Cauchy