Electronic copy available at: http://ssrn.

com/abstract=1746318
Exchange Rates and Oil Prices: A Multivariate Stochastic
Volatility Analysis
Minh Vo

Metropolitan State University

1501 Hennepin Ave. M2220
Minneapolis, MN 55403, U.S.A.
(minh.vo@metrostate.edu)
Liang Ding
Macalester College
1600 Grand Avenue
St. Paul, MN 55105, U.S.A.
(ding@macalester.edu)
Abstract
In contrast with the large literature that studies the interactions between the oil and the foreign
exchange (FX) markets at the return level, this paper examines their relationship at the risk level. We
employ the multivariate stochastic volatility (MSV) and the multivariate conditional correlation GARCH
(CC-MGARCH) framework to investigate the volatility interactions between the two in an attempt to
extract information intertwined in both markets for risk prediction. We oer three major ndings. First,
the volatility in each market is very persistent. It varies over time in a predictable manner, conditioned
on the past information. Second, the volatility in the oil market Granger-causes the volatility in the
FX markets but not the other way around. Thus, including the oil market into the information set will
improve the FX volatility forecast. Finally, the MSV models outperform the CC-MGARCH counterparts
in forecasting FX volatility.
Keywords: Oil price risk; Exchange rate risk; Multivariate stochastic volatility; Multivariate GARCH;
Volatility forecast

Corresponding author, phone: 612-659-7305

1
Electronic copy available at: http://ssrn.com/abstract=1746318
1 Introduction
It is well-known that oil prices and the U.S. dollar exchange rates are highly correlated. Given the fact that
oil is quoted in U.S. dollars, it is natural to hypothesize that exchange rates drive oil prices. More specically,
other things equal, when the U.S. dollar depreciates, oil-exporting countries would raise oil prices in order
to stabilize the purchasing power of their (U.S. dollar) export revenues in terms of their (predominately)
euro-denominated imports. This is equivalent to a reduction in supply or a leftward shift in the supply curve.
On the demand side, the U.S. dollar depreciation makes oil less expensive for consumers in other countries
(in local currency), thereby increasing their crude oil demand. Both eects, the reduction in supply and
the increase in demand, cause an increase in oil prices denominated in U.S. dollars. The exchange-rate-to-
oil-price causality relationship is supported by the empirical evidence found in Zhang, Fan, Tsai, and Wei
(2008), Krichene (2005) and Youse and Wirjanto (2004).
From the other perspective, exchange rates are believed to be determined by expected future fundamental
conditions, among which oil is surely an important factor. Increasing oil prices lead to stronger economies
for oil-exporters and higher production costs for oil-importers, hence it would cause the appreciation of oil-
exporter currencies relative to those of oil-importers. So, it is likely that the causality runs from oil prices to
the exchange rate. Benassy-Querea, Mignonb, and Penot (2007), Coudert, Mignon, and Penot (2007), Chen
and Chen (2007), Ayadi (2005), Chaudhuri and Daniel (1998) and Krugman (1984) all provide evidence
supporting this view.
These studies, despite their mixed implications, tend to suggest that oil prices and exchange rates prob-
ably both contain information that can aect each other. Accordingly, Chen, Rogo, and Rossi (2008) and
Groen and Pesenti (2010) use exchange rates to obtain a better forecast of oil (and other commodities) prices,
while Amano and Norden (1998) improve the exchange rate forecast by including oil price in the model.
The current literature that examines the relationship between oil price and exchange rate, as cited above,
mainly focuses on their returns. As noted by Clark (1973), Tauchen and Pitts (1983), and Ross (1989), the
volatility of an asset is also related to the rate of information ow across interacted markets. So the link
between the oil and the FX markets should appear not only in return but also in volatility. Examining the
volatility interaction between exchange rates and oil prices can shed light on the direction of the causality
relationship from a new perspective. Furthermore, if a signicant connection does exist, extracting and using
the information intertwined in both markets would improve prediction of exchange rate and oil risks, which
are critical in many areas of modern nance.
Instead of focusing on the relationship between exchange rate and oil returns, which many papers cited
above have investigated, this paper rst examines how the oil and the FX markets interact on the risk level.
Second, the paper attempts to extract information intertwined in the two markets, if detected, for a better
forecast of the exchange rate and oil volatility.
Based on the preliminary analysis of the data, we posit a bivariate model of vector of autoregression
VAR(5) with stochastic volatility for the joint processes governing the returns of various exchange rates and
2
Electronic copy available at: http://ssrn.com/abstract=1746318
the oil prices. We model the stochastic volatility using both the multivariate stochastic volatility (MSV)
and the conditional correlation multivariate GARCH (CC-MGARCH) models. We t two variants of the
models: the constant conditional correlation and the dynamic conditional correlation to identify the better
one for the data. For comprehensive discussion of the MSV and CC-MGARCH, the reader is referred to
Asai, McAleer, and Yu (2006) and Bauwens, Laurent, and Rombouts (2006). To estimate the MSV models,
we use the Bayesian Markov chain Monte Carlo (MCMC) method by Jacquier, Polson, and Rossi (1994) and
Kim, Shephard, and Chib (1998). For the CC-MGARCH models, we use the maximum likelihood methods
proposed by Bollerslev (1990) and Engle (2002).
On the return level, our empirical results show that the increase in oil prices Granger-causes the oil
exporting (importing) currencies to appreciate (depreciate), which is an intuitive result. On the contrary, we
did not nd consistently robust and signicant evidence showing that exchange rates Granger-cause oil prices
to change. With regard to volatility, our study oers three major ndings. First, the daily volatility in each
market is very persistent. It varies over time in a predictable manner, conditioned on the past information.
Second, the volatility in the oil market Granger-causes the volatility in FX markets but not the other way
around. In other words, innovations that hit the oil market also have some impact on the volatility of FX
market. Thus, including the oil market into the information set will improve the FX volatility forecast.
Third, the MSV models outperform the MGARCH counterparts in tting the data and forecasting exchange
rate volatility.
Overall, these results suggest that the causality relationship is more likely to run from oil markets to
FX markets. The possible mechanisms that exchange rates aect oil prices, such as the purchasing power
channel and the local price channel noted above, are based on supply and demand in the oil market, while the
mechanism of the opposite direction, such as the currency market expectation channel, is through market
trading behavior based on expectations of market participants. Oil prices and exchange rates, although
considered by many to be fundamental macro variables, are both asset prices that are determined mainly
by speculation-initiated transactions in the nancial markets. So the mechanisms based on the nancial
markets information expectation channel rather than the goods markets demand and supply framework is
more likely to dominate.
The remainder of the paper is organized as follows. Section 2 describes the data set and conducts a
preliminary analysis of the data. Section 3 discusses the models. Section 4 explains estimation methodology.
Section 5 presents estimation results. Section 6 checks the models in terms of goodness-of-t and evaluates
their forecast power, and Section 7 concludes the paper.
2 Data and Preliminary Analysis
The data set used in this study consists of a daily oil spot price time series and ve daily time series of
exchange rates against the U.S. dollar including the Canadian dollar (CAD), the Norwegian krone (NOK),
3
the Euro (EUR), the Indian rupee (INR), and the U.S. dollar trade weighted index. The Canadian dollar
and Norwegian krone are chosen to represent the exchange rates of oil exporting countries. The Euro and
the Indian rupee are to represent developed and developing oil consumers. The dollar index provides more
general sense to the results. Other major oil exporters (such as the Middle East countries) and importers
(such as China) are not included because their exchange rates are either pegged to the U.S. dollar or strictly
managed. All series span from July 28, 2005 to July 28, 2009 for a total of 1, 000 observations. The daily
oil spot price time series is based on the West Texas Intermediate (WTI) crude oil. It is obtained from the
historical database of the U.S. Department of Energy. The daily exchange rate data are extracted from the
Federal Reserve Bank of St. Louis website. Despite their dierent quoting habits, all the exchange rates are
converted to foreign currency per dollar. The daily return r
t
, in percentage, for each series j
t
is approximated
by r
t
= 100 [ln(j
t
) ln(j
t1
)].
Table 1 presents a wide range of descriptive statistics for all time series. All series have small mean and
standard deviations that are much greater than the mean in absolute value, indicating that the mean is
not signicantly dierent from zero. This is consistent with common knowledge that nancial time series
at this frequency usually follow a random walk. Except USD/INR, other exchange rate series have negative
skewness. The excess kurtosis for each is signicantly positive, indicating that they have heavy tails relative to
the normal distribution, which is also typical in these nancial data. Both Kolmogorov-Smirnov and Jarque-
Bera tests reject the null hypothesis that the return distributions are normal at 5% level of signicance.
Ljung-Box portmanteau tests on return and squared return series up to 6 and 12 lags indicate a high serial
correlation in the rst and second moments. The high autocorrelation in the second moment may be due
to changing conditional volatility over time. The 1CH tests at 6 and 12 lags reject the null hypothesis of
homoscedasticity in the data at 5% signicance level. Thus, the GARCH and SV framework is appropriate
to model the volatility of these time series.
[Insert Table 1 here]
Figures 1 and 2 graph the return time series and their distributions. We observe that for all series, the
return displays volatility clustering, another typical feature for high frequency nancial data. That is to say,
large changes tend to follow large changes, and small changes tend to follow small changes. However, the sign
of the change from one period to the next is unpredictable. The QQ-plots against the normal distribution
show that the distribution has heavier tail than normal.
[Insert Figure 1 here]
[Insert Figure 2 here]
In short, the preliminary analysis shows that the data exhibit heavy tails, autocorrelation and het-
eroscedasticity. This suggests the importance of using the stochastic volatility framework to model condi-
tional volatility of return in all series.
4
3 The Model
Based on the preliminary analysis in the previous section and using the optimal lag-length algorithm with
the Akaike information criterion (AIC), we posit the bivariate \ 1(5) with stochastic volatility models
for the joint processes governing the exchange rates and the oil returns. For comparison, we use both the
bivariate o\ and the bivariate G1CH (1, 1) to model the joint volatility process. Two variants of the
multivariate volatility models, the constant conditional correlation and the dynamic conditional correlation,
are used to t the data. This section discusses the model and the estimation methodology.
Let r
t
=
_
r
e
t
r
o
t
_
0
be the vector of returns of exchange rate and oil at time t (rate of change in
log-prices between time t 1 and time t). r
t
is modeled as a vector process consisting of a deterministic part
m
t
and a stochastic part e
t
:
r
t
= m
t
+e
t
. (1)
The deterministic part m
t
is specied as a vector autoregressive process of order ve \ 1(5):
m
t
= C+
5

k=1
B
k
r
tk
,
where C =
_
c
e
c
o
_
0
and B
k
is a (2 2) matrix with B
k

ij
= /
ij
k
for i, , = c, o. The subscripts c and
o stand for exchange rate and oil, respectively. The stochastic part e
t
=
_
c
e
t
c
o
t
_
0
involves the bivariate
stochastic volatility element of the model and depends on the model being used.
3.1 The Multivariate Stochastic Volatility Model (MSV)
The MSV model can be specied as follows:
e
t
=
t

t
, (2)
where,
t
, given
t
, is bivariate and normally distributed with mean 1 (z
t
[
t
) = 0
21
and variance
ar (
t
[
t
) = Z
t
=
_
_
1 j
t
j
t
1
_
_
.
The correlation coecient between c
e
t
and c
o
t
, j
t
, can be either constant or time-varying, depending on the
model. When j
t
is a constant, we have the constant conditional correlation MSV (CCC-MSV) model. When
it is time-varying, we have the dynamic conditional correlation (DCC-MSV) model.
t
is a (2 2) diagonal
matrix of standard deviation with the following form:

t
=
_
_
exp
_
h
e
t
2
_
0
0 exp
_
h
o
t
2
_
_
_
,
where /
e
t
, /
o
t
are conditional log-variances at time t of r
e
t
, r
o
t
, respectively.
5
3.1.1 The Constant Conditional Correlation MSV Model
The \ 1(5) model with constant conditional correlation multivariate stochastic volatility (CCC-MSV) can
be specied as follows:
r
t
= C+
5

k=1
B
k
r
tk
+
t

t
, (3)

t
[
t
~
_
_
0
21
,
_
_
1 j
j 1
_
_
_
_
(4)
h
t+1
= +(h
t
) +
t
, (5)
where h
t
=
_
/
e
t
/
o
t
_
0
is the vector of log-variances of r
e
t
and r
o
t
, =
_
j
e
j
o
_
0
is the unconditional
mean vector of h
t
. =
_
c
ij
_
, for i, , = c, o, represents the persistence and interaction between markets.
The error term
t
=
_
j
e
t
j
o
t
_
0
in the volatility equation is a bivariate normal random variable,
t
~

_
0
21
, diaq
_
o
2
e
, o
2
o
__
.
According to (3)(5) , conditioned on the conditional mean, C+
5

k=1
B
k
r
tk
, the return series of exchange
rate and oil are jointly normally distributed with variance-covariance matrix
t
Z
t

0
t
. The correlation between
the two series are assumed to be a constant j. Equation (5) models the time-varying cross-dependent
volatilities of the two series. In this equation, the diagonal parameters of the matrix (c
ee
and c
oo
) capture
the persistence of the volatility in each market while the o-diagonal parameters, c
eo
and c
oe
, measure the
dependence of the conditional volatility of the FX market on the oil market and vice versa. If c
eo
(c
oe
) is
dierent from zero, oil (FX) volatility may Granger-cause FX (oil) volatility.
3.1.2 The Dynamic Conditional Correlation MSV
When the correlation between the two returns is allowed to vary over time, we get the dynamic conditional
correlation MSV model (DCC-MSV). The DCC-MSV model with \ 1(5) in the mean has the following
form:
r
t
= C+
5

k=1
B
k
r
tk
+
t

t
, (6)
h
t+1
= +(h
t
) +
t
,

t
[
t
~
_
_
0
21
,
_
_
1 j
t
j
t
1
_
_
_
_
j
t
=
exp(
t
) 1
exp(
t
) + 1

t
= . + , (
t1
.) + o
q
.
t
, .
t
iid
~ (0, 1) .
6
This model captures the dynamic correlation between the two markets. To constrain j
t
to the interval
[1, 1], we use the Fisher transformation as suggested by Christodoulakis and Satchell (2002), in which j
t
is a function of
t
which follows an 1(1) stochastic process.
3.2 The Multivariate GARCH Model with Conditional Correlations
In addition to MSV, we also use the MGARCH model to t our data. The bivariate G1CH (1, 1) model
of exchange rate and oil returns has the following form:
r
t
= m
t
+e
t
,
e
t
= H
1=2
t

t
,
where H
1=2
t
is a 2 2 positive denite matrix and
t
is a 2 1 random vector with the following rst and
second moments:
1 (
t
) = 0
21
,
\ ar (
t
) = 1
2
,
where 1
2
is the identity matrix of order 2.
Thus, given the information at time t 1, the conditional variance-covariance matrix of r
t
is:
\ ar
t1
(r
t
) = \ ar (e
t
) = H
1=2
t
\ ar
t1
(
t
)
_
H
1=2
t
_
0
= H
t
.
The specications of H
t
depend on the specic MGARCH-type model. We examine the conditional
correlation multivariate GARCH models in this paper.
3.2.1 The Constant Conditional Correlation Multivariate GARCH (CCC-MGARCH)
The constant conditional correlation multivariate G1CH (1, 1) model of Bollerslev (1990) and Jeantheau
(1998) has the following form:
H
t
= 1
t
11
t
, (7)
1
t
= diaq
_
/
1=2
e;t
, /
1=2
o;t
_
,
where /
e;t
and /
o;t
can be considered as any univariate G1CH model of the two variables. 1 is their
constant correlation matrix which is a symmetric positive denite matrix with 1s on the diagonal.
7
3.2.2 The Dynamic Conditional Correlation Multivariate GARCH (DCC-MGARCH)
The DCC-MGARCH model of Engle (2002) is similar to the CCC-MGARCH model but it allows the cor-
relation matrix 1 in (7) to vary over time. Specically, 1
t
can be specied as follows for the bivariate
case:
1
t
= (Q
t
1
2
)
1=2
Q
t
(Q
t
1
2
)
1=2
where is the Hadamard product of two identically sized matrices, computed by element-by-element mul-
tiplication, 1
2
is the identity matrix of order 2, Q
t
is a 2 2 symmetric positive denite matrix, given
by
Q
t
= (1 c ,)

Q + ce
t1
e
0
t1
+ ,Q
t1
,
where

Q is the unconditional variance-covariance matrix of e
t
, and c and , are non-negative scalar parameters
satisfying the stationary condition: c + , < 1.
4 Estimation Methodology
4.1 MSV Models
4.1.1 The Algorithm
Estimating SV-type models is a challenge since they do not have closed-form likelihood functions due to
the latent structure of the variance. Several estimation methods have been proposed in literature, such
as, the quasi maximum likelihood (QML) of Harvey, Ruiz, and Shephard (1994), the generalized method
of moments (GMM) by Andersen and Sorensen (1996), Melino and Turnbull (1990), Sorensen (2000), the
ecient method of moments by Gallant, Hsieh, and Tauchen (1997), the simulated maximum likelihood by
Danielsson (1994), Durbin and Koopman (1997), and Sandmann and Koopman (1998), and the Bayesian
MCMC methods by Jacquier, Polson, and Rossi (1994), and Kim, Shephard, and Chib (1998). However,
Jacquier, Polson, and Rossi (1994) nd that the MCMC method is superior to both the QML and the GMM.
Furthermore, while most classical methods rely on asymptotic arguments to make inference, MCMC uses
exact posterior distributions of the parameters. This will yield more accurate results. In this paper, we use
the Bayesian MCMC for model estimation. For a comprehensive discussion of Bayesian inference and the
MCMC method, the reader is referred to Geweke (2005), Gammerman (1997), and Johannes and Polson
(2006).
The basic idea of Bayesian estimation lies on the Bayes theorem, which states that the posterior joint
distribution function of the parameters is proportional to the product of their prior distribution function and
the likelihood function of the data. In particular, let be the vector of parameters to be estimated, H
t
be
the vector of latent log volatility, and r
t
be the vector of observed data. Bayesian inference is then based on
8
the joint posterior distribution of unobservables
_
H
t
_
given the data r
t
. Let ) (.) be the probability
density function. Using the Bayes theorem, we get:
) (, H
t
[r
t
) _ ) (r
t
[, H
t
) ) (H
t
[) ) () .
In a sense, the posterior distribution function is a balance between the prior belief, ) (H
t
[) ) (), and the
likelihood of the data, 1(, H
t
[r
t
) = ) (r
t
[, H
t
). Inference is then conducted by evaluating its moments.
For simple and low-dimensional problems, the posterior distribution function may have tractable forms and
so evaluating its moments is a simple task. However, in complex and higher-dimensional problems, such
as the MSV model in this paper, the posterior distribution functions often do not have familiar functional
forms and we have to rely on simulation to make inferences.
The Markov chain Monte Carlo (MCMC) is a computational method to generate random samples from
a given distribution. It was rst introduced by Metropolis et al. (1953) and subsequently generalized by
Hastings (1970). It is based on the construction of a Markov chain in the parameter space. Under some mild
regularity conditions (see Tierney 1994) the chain asymptotically converges to the joint posterior distribution.
Thus, the realized value of the chain can be used to make inferences about the joint posterior distribution.
An MCMC algorithm called Gibbs sampler, introduced by Geman and Geman (1984), is based on the
Cliord-Hammersley theorem which states that a joint distribution can be characterized by its complete
conditional distributions. It is the natural choice in MCMC sampling when the conditionals from which
samples are drawn can be specied. This paper uses Gibbs sampler to generate Markov chains.
4.1.2 Estimation
The CCC-MSV model in (3) (5) can be rewritten elementwise as follows:
r
e
t
= c
e
+
5

i=1
/
ee
i
r
e
ti
+
5

i=1
/
eo
i
r
o
t1
+ c
e
t
, (8)
r
o
t
= c
o
+
5

i=1
/
oe
i
r
e
ti
+
5

i=1
/
oo
i
r
o
ti
+ c
o
t
, (9)
c
e
t
= exp
_
/
e
t
2
_
c
e
t
, (10)
c
o
t
= exp
_
/
o
t
2
_
c
o
t
, (11)
j = co (c
e
t
, c
o
t
) , (12)
/
e
t+1
= j
e
+ c
ee
(/
e
t
j
e
) + c
eo
(/
o
t
j
o
) + j
e
t
, j
e
t
iid
~
_
0, o
2
e
_
, (13)
/
o
t+1
= j
o
+ c
oe
(/
e
t
j
e
) + c
oo
(/
o
t
j
o
) + j
o
t
, j
o
t
iid
~
_
0, o
2
o
_
, (14)
To estimate this system, we rst use the maximum likelihood method to estimate the parameters C and
B in the mean equations (8)(9), then extract the residuals e
t
. In the second step, we employ the Bayesian
9
MCMC method with Gibbs sampling algorithm to estimate parameters in (10) (14) . To ensure that h
t
is
stationary, we constrain the persistent coecients c
ee
and c
oo
to the interval (1, 1) by setting c
ii
= 2c

ii
1
for i = c, o, where c

ii
has beta prior.
In the same manner, the DCC-MSV model (equation 6) can be written elementwise as follows:
r
e
t
= c
e
+
5

i=1
/
ee
i
r
e
ti
+
5

i=1
/
eo
i
r
o
t1
+ c
e
t
, (15)
r
o
t
= c
o
+
5

i=1
/
oe
i
r
e
ti
+
5

i=1
/
oo
i
r
o
ti
+ c
o
t
,
c
e
t
= exp
_
/
e
t
2
_
c
e
t
c
o
t
= exp
_
/
o
t
2
_
c
o
t
j
t
= co (c
e
t
, c
o
t
) =
exp(
t
) 1
exp(
t
) + 1
,

t
= . + , (
t1
.) + o
q
.
t
, .
t
iid
~ (0, 1) ,
/
e
t+1
= j
e
+ c
ee
(/
e
t
j
e
) + c
eo
(/
o
t
j
o
) + j
e
t
, j
e
t
iid
~
_
0, o
2
e
_
,
/
o
t+1
= j
o
+ c
oe
(/
e
t
j
s
) + c
oo
(/
o
t
j
o
) + j
o
t
, j
o
t
iid
~
_
0, o
2
o
_
.
The estimation strategy used in the constant correlation MSV model is also applied to this model.
4.2 CC-MGARCH Models
We use maximum likelihood to estimate the CC-MGARCH models. Like the MSV models, we rst estimate
\ 1(5) models for the mean and extract the residuals e
t
for volatility estimation. Let T
t1
be the o eld
generated by all the available information up to time t 1, then:
e
t
[T
t1
~ (0, H
t
) .
4.2.1 CCC-MGARCH Model
Given the CCC-MGARCH model:
e
t
[T
t1
~ (0, H
t
)
H
t
= 1
t
11
t
,
10
let 0 be the vector of all parameters in the model, the log likelihood function can be written as
1(0) =
1
2
T

t=1
_
ln(2) + ln[H
t
[ +e
0
t
H
1
t
e
t
_
=
1
2
T

t=1
_
ln(2) + ln[1
t
11
t
[ +e
0
t
(1
t
11
t
)
1
e
t
_
=
1
2
T

t=1
_
ln(2) + ln[1[ + 2 ln[1
t
[ + .
0
t
1
1
.
t
_
where is the number of time series, T is the sample size and .
t
= 1
1
t
c
t
is the standardized residual.
4.2.2 DCC-MGARCH Model
Given the DCC-MGARCH model:
e
t
[z
t1
~ (0, H
t
) ,
H
t
= 1
t
11
t
.
Let and be the vectors of parameters in 1
t
and 1
t
, respectively. The log likelihood for this estimator
can be written as:
1(, ) =
1
2
T

t=1
_
ln(2) + ln[H
t
[ +e
0
t
H
1
t
e
t
_
(16)
=
1
2
T

t=1
_
ln(2) + ln[1
t
1
t
1
t
[ +e
0
t
1
1
t
1
1
t
1
1
t
e
t
_
=
1
2
T

t=1
_
ln(2) + 2 ln[1
t
[ + ln[1
t
[ + .
0
t
1
1
t
.
t
_
=
1
2
T

t=1
_
ln(2) + 2 ln[1
t
[ + c
0
t
1
1
t
1
1
t
c
t
.
0
t
.
t
+ ln[1
t
[ + .
0
t
1
1
t
.
t
_
,
where is the number of time series, T is the sample size and .
t
= 1
1
t
c
t
is the standardized residual.
1(, ) can be rewritten as a sum of a volatility component 1
v
() and a correlation component 1
c
(, ):
1(, ) = 1
v
() + 1
c
(, )
where
1
v
() =
1
2
T

t=1
_
ln(2) + 2 ln[1
t
[ + c
0
t
1
2
t
c
t
_
is the sum of individual GARCH likelihoods, and
1
c
(, ) =
1
2
T

t=1
_
ln[1
t
[ + .
0
t
1
1
t
.
t
.
0
t
.
t
_
is the likelihood for correlations.
11
We use two-step estimation methodology proposed by Engle (2002). In the rst step, we estimate
^
= arg max [1
v
()]
and then we take
^
as given in the second step to estimate :
^
= arg max
_
1
c
_
^
,
__
5 Empirical Results
5.1 Mean Estimates
Tables 2 and 3 report the results of estimating the \ 1 for the mean. We use the optimal lag length of
5 as determined by the AIC. For the case of the USD/CAD (Table 2) the coecients of lags 1 and 3 of oil
returns are small but negative at 5% level of signicance in the FX equation, indicating that oil prices have
some positive but small impact on the CAD.
1
This could be due to the fact that Canada is one of the major
oil exporting countries. Other things equal, an increase in oil prices can be considered as a positive shock
to the Canadian economic fundamentals, which would therefore cause the Canadian dollar to appreciate.
This is also the case for Norway, another major oil exporting country, albeit at dierent lags. In a similar
argument, the increase in oil prices would raise production costs for oil importers. This is certainly a negative
fundamental shock for the biggest oil consumers in the world the U.S., which causes the USD to depreciate.
This is reected in the case of the USD index in the rst and second lags of oil returns and the rst lag for
the USD/EUR.
[Insert Table 2 here]
[Insert Table 3 here]
The impact, however, is insignicant for the USD/INR. Although the Indian rupee is market determined,
the Reserve Bank of India (the central bank of India) trades actively in the USD/INR currency market to
impact the eective exchange rates. Such managed oat rate might reect the policy makers goals more
than the markets interpretation of the fundamental news such as oil price shocks.
In the other direction, we see that the coecient on the third lag of USD/CAD in the oil equation is
signicant, as is the case for the USD index and USD/INR. The coecients for the USD/NOK and the
USD/EUR however are insignicant. Overall, the causality relationship from exchange rate to oil price is
not as signicant and robust as the relationship in the opposite direction.
The ARCH tests of Engle (1982) on residuals at 1, 6, and 12 lags reject the null hypothesis of homoscedas-
ticity for all cases. Thus, the MSV and MGARCH are appropriate for modeling volatility.
1
A decrease in the USD/CAD exchange rate implies the CAD appreciates relative to the USD.
12
5.2 Volatility Estimates
5.2.1 CCC-MSV Model
Table 4 reports the estimates of the following CCC-MSV model:
c
e
t
= exp
_
/
e
t
2
_
c
e
t
,
c
o
t
= exp
_
/
o
t
2
_
c
o
t
,
j = co (c
e
t
, c
o
t
)
/
e
t+1
= j
e
+ c
ee
(/
e
t
j
e
) + c
eo
(/
o
t
j
o
) + j
e
t
, j
e
t
iid
~
_
0, o
2
e
_
,
/
o
t+1
= j
o
+ c
oe
(/
e
t
j
s
) + c
oo
(/
o
t
j
o
) + j
o
t
, j
o
t
iid
~
_
0, o
2
o
_
.
[Insert Table 4 here]
We observe that both exchange rate and oil returns are very persistent. The persistent coecient for
exchange rates c
ee
ranges from .8439 for USD/INR to .9728 for USD/CAD. For oil, c
oo
varies from .9005
to .9855. These results are consistent with the fact that exchange rate and oil price volatilities are often
clustered.
A majority of the transmission coecients (c
eo
and c
oe
) are signicant but small (the 95% condence
interval does not contain 0.) The volatility transmission coecient from the oil markets to the FX markets
c
eo
is positive and signicant in all cases, suggesting that shocks to volatility of oil markets cause the volatility
in the FX market to increase at a later date. c
eo
is the largest for the USD/EUR and the USD/INR. Other
things equal, a shock that increases oil price volatility by 1% will increase the volatilities of USD/EUR
and USD/INR by approximately .1% and .14%, respectively. Intuitively, oil shocks increase uncertainty of
economic fundamentals which causes exchange rates to be more volatile.
For the transmission coecient from FX markets to oil markets, the result is mixed. c
oe
is not signicantly
dierent from zero for the USD index, the USD/INR and the USD/EUR at 5% signicance level. Thus, it
is unclear whether shocks to volatility of FX market cause volatility in oil markets to rise.
Except for USD/INR, the contemporaneous correlation j between oil and exchange rates is negative and
quite large in absolute value. This result is consistent with empirical ndings in other studies that dollar
value and oil price are overall inversely correlated. Figures 3 and 4 show the posterior densities of parameters
estimated in the model for the cases of the USD/CAD and the USD index.
[Insert Figure 3 here]
[Insert Figure 4 here]
Figures 5 and 6 show the volatility estimates of exchange rates and oil. To the naked eye it seems that
the estimates reect the variations of the residuals.
13
[Insert Figure 5 here]
[Insert Figure 6 here]
5.2.2 DCC-MSV Model
Table 5 reports the estimates of the DCC-MSV model:
c
e
t
= exp
_
/
e
t
2
_
c
e
t
,
c
o
t
= exp
_
/
o
t
2
_
c
o
t
,
j
t
= co (c
e
t
, c
o
t
) =
exp(
t
) 1
exp(
t
) + 1
,

t
= . + , (
t1
.) + o
q
.
t
, .
t
iid
~ (0, 1) ,
/
e
t+1
= j
e
+ c
ee
(/
e
t
j
e
) + c
eo
(/
o
t
j
o
) + j
e
t
, j
e
t
iid
~
_
0, o
2
e
_
,
/
o
t+1
= j
o
+ c
oe
(/
e
t
j
e
) + c
oo
(/
o
t
j
o
) + j
o
t
, j
o
t
iid
~
_
0, o
2
o
_
.
[Insert Table 5 here]
Like the estimates of the CCC-MSV model, the persistent coecients (c
ee
and c
oo
) are quite close to
one, except the case of USD/NOK where c
ee
drops from .9139 in the CCC-MSV to .669 in the DCC-MSV
model. The transmission coecient c
eo
are all found to be signicantly dierent from zero. Except for the
USD index, this coecient signicantly increases compared to CCC-MSV model. Like the CCC-MSV case,
the transmission coecient from FX markets to oil market, c
oe
, is signicant in most cases except for the
USD/INR and the USD/CAD, suggesting exchange rate volatility also contains information that can aect
oil market volatility in later dates.
We also observe that in all cases, the DIC (deviance information criterion) is smaller in the DCC-MSV
than the CCC-MSV, suggesting that the former ts the data better. Considering that regime switch and
structural breaks are often identied in time series of exchange rate and oil price, their relationship is very
likely time-varying so that DCC model can capture such a feature better. Nevertheless, which model is more
appropriate depends on its forecasting power. We will assess the forecasting performance of each model
later in the paper. Figures 7 and 8 show the posterior densities of parameters estimated for the cases of
USD/CAD and the USD index.
[Insert Figure 7 here]
[Insert Figure 8 here]
Figures 9 and 10 graph the estimated volatilities of all exchange rates and oil and their correlation
coecients for DCC-MSV model. As in the CCC-MSV model, we see that the estimated volatilities reect
14
quite well the variation of the residuals. Except for USD/INR, the correlation coecients are negative and
quite volatile, varying from around .1 to around .5. For USD/INR, it ranges from .2 to .4.
[Insert Figure 9 here]
[Insert Figure 10 here]
5.2.3 CCC-MGARCH Model
Table 6 reports the parameter estimates for the following CCC-MGARCH(1,1) model, written in elementwise
form:
c
e;t
=
_
/
e;t
c
e;t
,
c
o;t
=
_
/
o;t
c
o;t
,
j = co (c
e;t
, c
o;t
)
_
_
/
e;t
/
o;t
_
_
=
_
_
c
1
c
2
_
_
+
_
_
a
11
a
12
a
21
a
22
_
_
_
_
c
2
e;t1
c
2
o;t1
_
_
+
_
_
/
11
/
12
/
21
/
22
_
_
_
_
/
e;t1
/
o;t1
_
_
.
[Insert Table 6 here]
Parameters a
ii
, /
ii
, i = 1, 2 represent the persistence of the volatility, while a
ij
, /
ij
, i, , = 1, 2 and i ,= ,,
are transmission coecients, representing the impact of volatility in one market on the other. We observe
that as in the MSV models, volatilities are very persistent (the sum a
ii
+ /
ii
is close to unity), once again
conrming the clustered volatilities in these nancial time series.
Unlike the MSV models, the transmission coecients a
12
and /
12
, from the oil market to the FX market,
are not signicantly dierent from zero for all cases, suggesting that volatility in the oil market does not have
signicant impacts on the FX market. However, the coecients a
21
and /
21
, from the FX to the oil market
are signicant in most cases. a
21
is signicant for CAN/USD, USD/EUR and USD/INR; /
21
is signicant
for all but USD/INR. It seems that the MGARCH and MSV are quite inconsistent in terms of the direction
of the volatility causality. A further evaluation of the model selection later would help us to judge which
one is more reliable.
The estimated correlations between exchange rates and oil prices are comparable those in the CCC-MSV.
The pattern that rising oil price is usually associated with depreciation of the dollar seems to be robust to
both MSV and MGARCH frameworks.
5.2.4 DCC-MGARCH Model
Table 7 reports the estimates of the DCC-MGARCH(1,1) model. This model is similar to the CCC-
MGARCH(1,1), except that the coecient j between c
e
t
and c
o
t
is not a constant but time-varying. The
result is similar to that in the CCC-MGARCH model.
15
[Insert Table 7 here]
Figures 11 and 12 graph the dynamic correlations.
[Insert Figure 11 here]
[Insert Figure 12 here]
6 Comparing Models
6.1 Model Checking
In this section, we check the goodness-of-t of each model and compare them. To check the goodness-of-t,
we test whether the volatility in each market indicated by each model is sucient to explain the stylized facts,
such as heavy tail, volatility clustering, etc. In particular, we test if the residuals in the mean equations,
having been standardized by the corresponding volatility in the variance equations, are standard normal.
Tables 8 and 9 present the test results.
Table 8 reports various statistics of the standardized residual for CCC-MSV and DCC-MSV models.
We observe that for all cases, the mean in each market is close to zero while the standard deviation is
close to one. Furthermore, the kurtosis for both markets in both models are close to 3, indicating that the
thick tail has been removed. We use both the Jarque-Bera and the Kolmogorov-Smirnov methods to test
for normality. In all cases, the Kolmogorov-Smirnov test cannot reject the null hypothesis of normality at
5% level of signicance. The results for the Jarque-Bera tests are mixed, however. For CCC-MSV, the test
rejects the null at 5% level of signicance for exchange rate in the cases of the USD index and the USD/EUR.
For DCC-MSV, it rejects the null for exchange rate in the USD index case. Despite that, in a majority of
cases, it does not reject the null hypothesis of normality. Thus, both CCC-MSV and DCC-MSV models do
quite well in modelling heteroscedasticity in the data.
[Insert Table 8 here]
To compare CCC-MSV and DCC-MSV, in terms of goodness-of-t, we use the deviance information
criterion (DIC) proposed by Spiegelhalter, Best, Carlin, and der Linde (2002) which is a Bayesian version
of the Akaike Information Criterion (AIC) (Akaike 1973) and closely related to the Bayesian (or Schwarz)
Information Criterion (BIC) (Schwarz 1978). Like the AIC and the BIC, it trades o the model adequacy,
measured by the log-likelihood, against the model complexity, measured by the number of free parameters.
The smaller the DIC, the better the model. For a discussion of the use of the DIC to compare SV models,
the reader is referred to Berg, Myer, and Yu (2004) . We see that in all cases, the two models have similar
DICs, although the DCC-MSV model has slightly lower DIC than the CCC-MSV model; thus, it is hard
to tell which model is better for the data. Later in the paper, we will revisit the issue by comparing their
forecasting performance.
16
Table 9 reports similar statistics for CCC-MGARCH and DCC-MGARCH models. We observe that the
tail thickness has been reduced (kurtosis is close to 3); however, it is still thicker than in the cases of MSV
models. Jarque-Bera tests reject the null hypothesis of normality at 5% level of signicance in most cases.
Thus, it seems that MSV models outperform MGARCH counterparts in terms of goodness-of-t. This in
turn implies that the volatility causality detected by MSV should be more reliable than MGARCH.
[Insert Table 9 here]
Comparing DCC-MGARCH with CCC-MGARCH, in all cases, the value of log likelihood is much higher
in DCC-MGARCH than in CCC-MGARCH, suggesting that the DCC-MGARCH model better t the data
than the CCC-MGARCH.
6.2 Forecast Performance
To evaluate the forecasting power of a volatility model, we rst need realized volatility. There are a number
of ways in literature to obtain realized volatility (see, for example Merton 1980; Perry 1982; Akgiray 1989;
Ding, Granger, and Engle 1993.) In this paper, we use the approach of Merton (1980) and Perry (1982) by
simply using the square of daily return as a proxy of realized daily return.
To evaluate the forecast performance, we use two popular metrics: the root mean square error (RMSE)
and the mean absolute error (MAE). They are given by:
1'o1 =

_
1

i=1
_
^ o
2
i
o
2
i
_
2
,
'1 =
1

i=1

^ o
2
i
o
2
i

,
where ^ o
2
i
is the forecast volatility of date i indicated by the model, o
2
i
is the realized volatility at date i.
is the number of observations in the sample. The lower the metrics, the better is the model in terms of
forecast performance.
To conduct the forecast experiment, we use the estimates for each model in the previous section to
forecast volatilities for the next two weeks and calculate the two metrics. To see how multivariate models
perform compared to the univariate counterparts, we also include GARCH(1,1) and the univariate stochastic
volatility model in the experiment. Table 10 reports the results.
[Insert Table 10 here]
For exchange rates, we observe that both univariate and multivariate SV models signicantly outperform
the GARCH counterparts in all cases. Among them, the CCC-MSV is ranked number one in three cases,
USD/CAD, USD/NOK and USD index. DCC-MSV is ranked number one in the other two cases. However,
17
both RMSE and MAE metrics indicate that there is no signicant dierence between the two models. The
univariate SV is ranked number 3 in all cases. We also see that in most cases, multivariate GARCH models
outperform the univariate GARCH in forecasting FX volatility. For oil, the result is completely dierent,
and the univariate GARCH outperforms all others. Thus, oil volatility helps predict exchange rate volatility
but not the other way around. This result strongly suggests that their causality relationship runs from oil
market to the FX market, not the opposite. This is consistent with the empirical result of the MSV models.
Recall potential mechanisms connecting exchange rate and oil price mentioned in the literature: From
exchange rate to oil price, purchasing power channel argues that the depreciation of the USD reduces
purchasing power of oil exporters so that they want to raise price for compensation. The local price channel
suggests that the U.S. dollar depreciation makes oil less expensive for consumers in non-U.S. dollar regions
(in local currency), thereby increasing their crude oil demand, which eventually causes adjustments in the
oil price denominated in U.S. dollars. From oil price to exchange rate, a possible mechanism is through the
currency market channel. The exchange rate is determined by expected future fundamental conditions of two
countries. Increasing oil prices lead to stronger economies of oil exporters, hence causing the appreciation
of their currencies. On the other hand, increasing oil price implies higher production costs for oil importers
and hence should cause the depreciation of their currencies.
Apparently, the mechanisms from exchange rates to oil prices mentioned above are based on supply
and demand in the oil market, while the mechanism of the opposite direction is through market trading
behavior based on expectations of market participants. Oil prices and exchange rates, though considered
by many to be fundamental macroeconomic variables, are both asset prices that are determined mainly by
speculation-initiated transactions in nancial markets. So the mechanisms based on supply and demand
in goods markets are probably not as strong as the mechanisms based on nancial markets information
expectation channel. The empirical results found in this paper tends to imply this intuition.
7 Conclusion
This paper studies the possible connection between the oil market and the FX market. Instead of analyzing
the relationship of their returns which has been done extensively in the literature, we focus on the volatility
association of both markets in an attempt to extract information intertwined in the two for better risk
predictions. We model the volatility interactions with the multivariate stochastic volatility as well as the
multivariate GARCH framework, allowing the volatility in each market to Granger-cause that in the other.
We oer three major ndings. First, conditioned on the past information, the daily volatility in each market
is very persistent. Second, the volatility in the oil market Granger-causes the volatility in the FX market
but not the other way around. In other words, innovations that hit the oil market also have some impact on
the volatility of FX market and thus using such a dependence signicantly enhances the forecasting power
of the models. Third, the MSV models do a better job than the MGARCH counterparts in tting the data
18
and forecasting exchange rate volatility.
References
Akaike, H. (1973). Information theory and an extension of the maximum likelihood principle. In B. N.
Petrov and F. Csaki (Eds.), Proceedings of the Second International Symposium on Information Theory,
Budapest, pp. 267281. Akademiai Kiado.
Akgiray, V. (1989). Conditional heteroscedasticity in time series of stock returns: Evidence and forecast.
Journal of Business 62, 5580.
Amano, R. A. and S. Norden (1998). Oil prices and the rise and fall of the US real exchange rate. Journal
of International Money and Finance 17, 299316.
Andersen, T. and B. Sorensen (1996). GMM estimation of a stochastic volatility model: A Monte Carlo
study. Journal of Business and Economic Statistics 14, 329352.
Asai, M., M. McAleer, and J. Yu (2006). Multivariate stochastic volatility: A review. Econometric Re-
view 25, 145175.
Ayadi, F. O. (2005). Oil price uctuations and the Nigerian economy. OPEC Review 29, 199217.
Bauwens, L., S. Laurent, and J. V. K. Rombouts (2006). Multivariate GARCH models: A survey. Journal
of Applied Econometrics 21, 79109.
Benassy-Querea, A., V. Mignonb, and A. Penot (2007). China and the relationship between the oil price
and the dollar. Energy Policy 35, 57955805.
Berg, A., R. Myer, and J. Yu (2004). Deviance information criterion for comparing stochastic volatility
models. Journal of Business and Economic Statistics 22, 107120.
Bollerslev, T. (1990). Modeling the coherence in short-run nominal exchange rates: A multivariate gener-
alized ARCH model. Review of Economics and Statistics 72, 498505.
Chaudhuri, K. and B. C. Daniel (1998). Long-run equilibrium real exchange rates and oil prices. Economics
Letters 58, 231238.
Chen, S. S. and H. C. Chen (2007). Oil prices and real exchange rates. Energy Economics 29, 390404.
Chen, Y. C., K. Rogo, and B. Rossi (2008). Can exchange rates forecast commodity prices? NBER
Working Paper No. 13901.
Christodoulakis, G. A. and S. E. Satchell (2002). Correlated ARCH (CorrARCH): Modelling the time-
varying conditional correlation between nancial asset returns. European Journal of Operational Re-
search 139, 351370.
19
Clark, P. (1973). A subordinate stochastic process model with nite variance for speculative prices. Econo-
metrica 41, 135155.
Coudert, V., V. Mignon, and A. Penot (2007). Oil price and the dollar. Energy Studies Review 15, 4865.
Danielsson, J. (1994). Stochastic volatility in asset prices: Estimation with simulated maximum likelihood.
Journal of Econometrics 64, 375400.
Ding, Z., C. W. J. Granger, and R. F. Engle (1993). A long memory of stock returns and a new model.
Journal of Empirical Finance 1, 83106.
Durbin, J. and S. Koopman (1997). Monte Carlo maximum likelihood estimation for non-Gaussian state
space models. Biometrika 84, 669684.
Engle, R. (2002). Dynamic conditional correlation: A simple class of multivariate generalized autore-
gressive conditional heteroskedasticity models. Journal of Business and Economic Statistics 20(3),
339350.
Engle, R. F. (1982). Autoregressive conditional heteroskedasticity with estimates of the variance of U.K.
ination. Econometrica 50, 9871007.
Gallant, A. R., D. Hsieh, and G. Tauchen (1997). Estimation of stochastic volatility models with diagnos-
tics. Journal of Econometrics 81, 159192.
Gammerman, D. (1997). Markov Chain Monte Carlo: Stochastic Simulation for Bayesian Inference. Chap-
man & Hall.
Geman, S. and D. Geman (1984). Stochastic relaxation, Gibbs distributions, and the Bayesian restoration
images. IEEE Transactions on Pattern Analysis and Machine Intelligence 6, 721741.
Geweke, J. (2005). Contemporary Bayesian Econometrics and Statistics. Wiley & Sons.
Groen, J. J. J. and P. A. Pesenti (2010). Commodity prices, commodity currencies and global economic
developments. NBER Working Paper 15743.
Harvey, A. C., E. Ruiz, and N. Shephard (1994). Multivariate stochastic variance models. Review of
Economic Studies 61, 247264.
Hastings, W. K. (1970). Monte Carlo sampling methods using Markov chains and their applications.
Biometrika 57(1), 97109.
Jacquier, E., N. G. Polson, and P. E. Rossi (1994). Bayesian analysis of stochastic volatility models.
Journal of Business and Economic Statistics 12, 371389.
Jeantheau, T. (1998). Strong consistency of estimators for multivariate ARCH models. Econometric The-
ory 14, 7086.
Johannes, M. and N. Polson (2006). MCMC methods for continuous-time nancial econometrics. In Y. At-
Sahalia and L. Hansen (Eds.), Handbook of Financial Econometrics. Elsevier.
20
Kim, S., N. Shephard, and S. Chib (1998). Stochastic volatility: Likelihood inference and comparison with
ARCH models. Review of Economic Studies 65, 363393.
Krichene, N. (2005). A simultaneous equations model for world crude oil and natural gas markets. IMF
Working Paper.
Krugman, P. R. (1984). Oil shocks and exchange rate dynamics. Frankel, J. A. (ed.). Exchange Rates and
International Macroeconomics. University of Chicago Press.
Melino, A. and S. M. Turnbull (1990). Pricing foreign currency options with with stochastic volatility.
Journal of Econometrics 45, 239265.
Merton, R. (1980). On estimating the expected return on the market: An explanatory investigation.
Journal of Financial Economics 8, 323361.
Metropolis, N., A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller (1953). Equations of
state calculations by fast computing machines. Journal of Chemical Physics 21(6), 10871092.
Perry, P. (1982). The time-variance relationship of security returns: Implications for the return-generating
stochastic process. Journal of Finance 37, 857870.
Ross, S. (1989). Information and volatility: The no-arbitrage martingale approach to timing and resolution
irrelevancy. Journal of Finance 44, 117.
Sandmann, G. and S. J. Koopman (1998). Estimation of stochastic volatility models via Monte Carlo
maximum likelihood. Journal of Econometrics 87, 271301.
Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics 6, 461464.
Sorensen, M. (2000). Prediction-based estimating functions. Econometrics Journal 3, 123147.
Spiegelhalter, D. J., N. G. Best, B. P. Carlin, and A. V. der Linde (2002). Bayesian measures of model
complexity and t. Journal of the Royal Statistical Society 64, 583639.
Tauchen, G. and M. Pitts (1983). The price variability-volume relationship on speculative markets. Econo-
metrica 51, 485505.
Tierney, L. (1994). Markov chains for exploring posterior distributions. Annals of Statistics 22(4), 1701
1762.
Youse, A. and T. S. Wirjanto (2004). The empirical role of the exchange rate on the crude-oil price
formation. Energy Economics 26, 783799.
Zhang, Y. J., Y. Fan, H. T. Tsai, and Y. M. Wei (2008). Spillover eect of US dollar exchange rate on oil
prices. Journal of Policy Modeling 30, 97391.
21
-
4
-
2
0
2
4
Time
C
A
D
/ U
S
D
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
4
-
2
0
2
4
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-4 -2 0 2 4
0
. 0
0
. 2
0
. 4
0
. 6
0
. 8
D
e
n
s
i t y
CAD/USD
Nor mal
-
6
-
2
0
2
4
Time
N
O
K
/ U
S
D
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
6
-
2
0
2
4
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-6 -4 -2 0 2 4
0
. 0
0
. 2
0
. 4
0
. 6
D
e
n
s
i t y
NOK/USD
Nor mal
-
4
-
2
0
2
Time
E
U
R
/ U
S
D
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
4
-
2
0
2
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-4 -2 0 2
0
. 0
0
. 4
0
. 8
D
e
n
s
i t y
EUR/USD
Nor mal
Figure 1: Exchange rate returns and their distributions for CAD, NOK and EUR
-
4
-
2
0
2
4
Time
I N
R
/ U
S
D
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
4
-
2
0
2
4
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-4 -2 0 2 4
0
. 0
0
. 4
0
. 8
1
. 2
D
e
n
s
i t y
INR/USD
Nor mal
-
4
-
2
0
1
2
Time
U
S
D
I n
d
e
x
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
4
-
2
0
1
2
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-4 -3 -2 -1 0 1 2
0
. 0
0
. 4
0
. 8
D
e
n
s
i t y
USD inde x
Nor mal
-
1
0
0
5
1
5
Time
O
i l
2006 2007 2008 2009
-3 -2 -1 0 1 2 3
-
1
0
0
5
1
5
Theoretical Quantiles
S
a
m
p
l e
Q
u
a
n
t i l e
s
-15 -5 0 5 10 15
0
. 0
0
0
. 0
5
0
. 1
0
0
. 1
5
D
e
n
s
i t y Oil
Nor mal
Figure 2: Exchange rate returns and their distributions for INR, USD index and oil
22
-2.5 -2.0 -1.5 -1.0 -0.5
0
. 0
0
. 5
1
. 0
1
. 5
Density of e
e
D
e
n
s
i t y
0.0 0.5 1.0 1.5 2.0 2.5
0
. 0
0
. 5
1
. 0
1
. 5
Density of o
o
D
e
n
s
i t y
0.90 0.92 0.94 0.96 0.98 1.00
0
5
1
0
1
5
2
0
2
5
Density of ee
ee
D
e
n
s
i t y
0.85 0.90 0.95 1.00
0
5
1
0
1
5
2
0
Density of oo
oo
D
e
n
s
i t y
-0.02 0.00 0.02 0.04 0.06 0.08
0
5
1
0
1
5
2
0
2
5
3
0
3
5
Density of eo
eo
D
e
n
s
i t y
0.00 0.05 0.10 0.15
0
5
1
0
1
5
2
0
Density of oe
oe
D
e
n
s
i t y
-0.40 -0.30
0
2
4
6
8
1
0
1
2
1
4
Density of

D
e
n
s
i t y
0.05 0.10 0.15 0.20
0
5
1
0
1
5
Density of e
e
D
e
n
s
i t y
0.10 0.15 0.20 0.25
0
2
4
6
8
1
0
1
2
1
4
Density of o
o
D
e
n
s
i t y
Figure 3: Posterior densities of parameters estimated for CCC-MSV model for USD/CAD
-2.5 -2.0 -1.5 -1.0
0
. 0
0
. 5
1
. 0
1
. 5
Density of e
e
D
e
n
s
i t y
1.0 1.5 2.0
0
. 0
0
. 5
1
. 0
1
. 5
2
. 0
Density of o
o
D
e
n
s
i t y
0.6 0.7 0.8 0.9 1.0
0
5
1
0
1
5
Density of ee
ee
D
e
n
s
i t y
0.90 0.94 0.98
0
5
1
0
1
5
2
0
2
5
3
0
Density of oo
oo
D
e
n
s
i t y
0.0 0.1 0.2 0.3
0
5
1
0
1
5
2
0
Density of eo
eo
D
e
n
s
i t y
-0.02 0.02 0.06 0.10
0
5
1
0
1
5
2
0
2
5
Density of oe
oe
D
e
n
s
i t y
-0.45 -0.35 -0.25
0
2
4
6
8
1
0
Density of

D
e
n
s
i t y
0.0 0.1 0.2 0.3 0.4 0.5
0
2
4
6
8
Density of e
e
D
e
n
s
i t y
0.08 0.10 0.12 0.14 0.16
0
5
1
0
1
5
2
0
2
5
3
0
Density of o
o
D
e
n
s
i t y
Figure 4: Posterior densities of parameters estimated for CCC-MSV for USD index
23
0
1
2
3
4
5
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
CAD/USD
0
. 4
0
. 6
0
. 8
1
. 0
1
. 2
1
. 4
1
. 6
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
0
1
2
3
4
5
6
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
NOK/USD
0
. 5
1
. 0
1
. 5
2
. 0
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
0
2
4
6
8
1
0
1
2
1
4
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
USD/EUR
2
3
4
5
6
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
Figure 5: Estimated volatilities of USD/CAD, USD/NOK and USD/EUR for CCC-MSV model
0
1
2
3
4
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
USD Inde x
0
. 4
0
. 6
0
. 8
1
. 0
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
0
1
2
3
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
INR/USD
0
. 5
1
. 0
1
. 5
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
0
5
1
0
1
5
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
Oil
2
3
4
5
6
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
Figure 6: Estimated volatilities of the USD index, USD/INR and oil for CCC-MSV model
24
-2.0 -1.5 -1.0 -0.5
0
. 0
0
. 5
1
. 0
1
. 5
Density of e
e
D
e
n
s
i t y
0.5 1.0 1.5 2.0
0
. 0
0
. 5
1
. 0
1
. 5
Density of o
o
D
e
n
s
i t y
0.90 0.95 1.00
0
5
1
0
2
0
Density of ee
ee
D
e
n
s
i t y
0.85 0.90 0.95 1.00
0
5
1
0
2
0
Density of oo
oo
D
e
n
s
i t y
0.00 0.04 0.08 0.12
0
5
1
0
2
0
Density of eo
eo
D
e
n
s
i t y
0.00 0.05 0.10 0.15
0
5
1
0
2
0
Density of oe
oe
D
e
n
s
i t y
0.10 0.15 0.20 0.25 0.30
0
5
1
0
1
5
2
0
2
5
Density of e
e
D
e
n
s
i t y
0.08 0.12 0.16 0.20
0
5
1
0
1
5
Density of o
o
D
e
n
s
i t y
-0.8 -0.6 -0.4 -0.2
0
1
2
3
4
Density of

D
e
n
s
i t y
0.08 0.12 0.16 0.20
0
5
1
0
1
5
Density of

D
e
n
s
i t y
0.1 0.2 0.3 0.4
0
2
4
6
8
1
0
Density of q
q
D
e
n
s
i t y
Figure 7: Posterior densities of parameters estimated for DCC-MSV for USD/CAD
-2.5 -2.0 -1.5 -1.0
0
. 0
0
. 5
1
. 0
1
. 5
Density of e
e
D
e
n
s
i t y
1.0 1.5 2.0
0
. 0
0
. 5
1
. 0
1
. 5
Density of o
o
D
e
n
s
i t y
0.70 0.80 0.90 1.00
0
5
1
0
1
5
Density of ee
ee
D
e
n
s
i t y
0.90 0.95 1.00
0
5
1
0
2
0
Density of oo
oo
D
e
n
s
i t y
0.00 0.10 0.20
0
5
1
0
1
5
Density of eo
eo
D
e
n
s
i t y
0.00 0.05 0.10
0
5
1
0
2
0
Density of oe
oe
D
e
n
s
i t y
0.05 0.15 0.25
0
2
4
6
8
1
2
Density of e
e
D
e
n
s
i t y
0.10 0.15 0.20
0
5
1
0
2
0
Density of o
o
D
e
n
s
i t y
-0.8 -0.6 -0.4 -0.2
0
1
2
3
4
Density of

D
e
n
s
i t y
0.10 0.15 0.20
0
5
1
0
2
0
Density of

D
e
n
s
i t y
0.0 0.1 0.2 0.3 0.4 0.5 0.6
0
1
2
3
4
5
Density of q
q
D
e
n
s
i t y
Figure 8: Posterior densities of parameters estimated for DCC-MSV for USD index
25
0
1
2
3
4
5
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
CAD/USD
0
. 4
0
. 6
0
. 8
1
. 0
1
. 2
1
. 4
1
. 6
1
. 8
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
0
1
2
3
4
5
6
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
NOK/USD
0
. 5
1
. 0
1
. 5
2
. 0
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
- 0
. 5
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
0
1
2
3
4
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
USD/EUR
0
. 4
0
. 6
0
. 8
1
. 0
1
. 2
1
. 4
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
- 0
. 4
0
- 0
. 3
0
- 0
. 2
0
- 0
. 1
0
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
Figure 9: Estimated volatilities of CAD/USD, NOK/USD and USD/EUR for DCC-MSV model
0
1
2
3
4
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
USD Inde x
0
. 4
0
. 6
0
. 8
1
. 0
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
0
1
2
3
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
INR/USD
0
. 5
1
. 0
1
. 5
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
- 0
. 3
- 0
. 2
- 0
. 1
0
. 0
0
. 1
0
. 2
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
0
5
1
0
1
5
Time
A
b
s
. R
e
s
.
2006 2007 2008 2009
Oil
2
3
4
5
6
Time
V
o
l a
t i l i t y
2006 2007 2008 2009
Figure 10: Estimated volatilities of the USD index, INR/USD and oil for DCC-MSV model
26
- 4
- 2
0
2
4
Time
C
A
D
/ U
S
D
2006 2007 2008 2009
CAD/USD
- 1
0
- 5
0
5
1
0
1
5
Time
O
i l
2006 2007 2008 2009
- 0
. 4
5
- 0
. 3
5
- 0
. 2
5
- 0
. 1
5
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
- 6
- 4
- 2
0
2
4
Time
N
O
K
/ U
S
D
2006 2007 2008 2009
NOK/USD
- 1
0
- 5
0
5
1
0
1
5
Time
O
i l
2006 2007 2008 2009
- 0
. 5
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
- 4
- 2
0
2
Time
U
S
D
/ E
U
R
2006 2007 2008 2009
USD/EUR
- 1
0
- 5
0
5
1
0
1
5
Time
O
i l
2006 2007 2008 2009
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
0
. 0
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
Figure 11: Estimated time-varying correlations between oil and CAD/USD, NOK/USD, USD/EUR in the
DCC-MGARCH(1,1) model
- 4
- 3
- 2
- 1
0
1
Time
I n
d
e
x
2006 2007 2008 2009
USD Inde x
- 1
0
- 5
0
5
1
0
1
5
Time
O
i l
2006 2007 2008 2009
- 0
. 4
- 0
. 3
- 0
. 2
- 0
. 1
0
. 0
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
- 4
- 2
0
2
4
Time
I N
R
/ U
S
D
2006 2007 2008 2009
INR/USD
- 1
0
- 5
0
5
1
0
1
5
Time
O
i l
2006 2007 2008 2009
- 0
. 2
5
- 0
. 1
5
- 0
. 0
5
Time
C
o
r r e
l a
t i o
n
2006 2007 2008 2009
Figure 12: Estimated time-varying correlations between oil and USD index, INR/USD in the DCC-
MGARCH(1,1) model
27
T
a
b
l
e
1
:
D
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
e
x
c
h
a
n
g
e
r
a
t
e
r
e
t
u
r
n
t
i
m
e
s
e
r
i
e
s
S
t
a
t
i
s
t
i
c
s
O
i
l
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
/
E
U
R
U
S
D
/
I
N
R
U
S
D
I
n
d
e
x
M
e
a
n
(
%
)
.
0
1
2
8

.
0
1
3
3

.
0
0
6
5

.
0
1
6
6
.
0
1
0
6

.
0
1
3
4
S
t
d
.
D
e
v
.
(
%
)
2
.
9
0
6
4
.
7
5
4
7
.
9
5
6
9
.
6
7
8
3
.
5
3
8
5
.
5
3
2
4
S
k
e
w
n
e
s
s
.
0
7
4
8

.
2
7
5
8

.
1
6
8
1

.
4
1
1
9
.
1
6
9
8

.
7
4
1
9
E
x
c
e
s
s
K
u
r
t
o
s
i
s
4
.
3
5
2
2
6
.
3
5
1
1
4
.
9
5
9
6
4
.
9
7
8
0
9
.
4
4
2
8
6
.
1
5
3
5
K
o
l
m
o
g
o
r
o
v
-
S
m
i
r
n
o
v
.
1
9
4
6

.
1
3
9
3

.
0
8
7
6

.
1
4
9
9
.
2
0
9
7

.
1
9
3
3

J
a
r
q
u
e
-
B
e
r
a
7
9
4
.
7
2

1
7
0
1
.
6
6

1
0
3
5
.
1
8
1
0
6
6
.
4
6

3
7
3
6
.
0
1

1
6
7
7
.
5
0

1
C
H
(
6
)
4
4
3
.
4
8

1
9
2
.
5
0

1
0
5
.
8
6

1
4
8
.
3
6

6
9
.
7
8

7
7
.
5
5

1
C
H
(
1
2
)
7
0
5
.
7
6

4
5
3
.
3
2

2
8
0
.
5
0

2
8
9
.
9
6

8
3
.
6
7

1
6
5
.
6
4

L
j
u
n
g
-
B
o
x
(
6
)
7
5
.
2
9

2
1
.
7
3

1
9
.
3
0

2
2
.
4
2

1
4
.
8
3

1
8
.
5
8

L
j
u
n
g
-
B
o
x
(
1
2
)
1
0
3
.
2
7

7
1
.
0
7

7
5
.
9
1

5
3
.
9
3

3
2
.
7
6

5
3
.
0
9

L
j
u
n
g
-
B
o
x
(
6
)
o
n
s
q
u
a
r
e
d
r
e
t
u
r
n
s
3
6
0
.
5
8

2
3
2
.
0
8

1
3
5
.
5
7

2
0
7
.
9
6

4
4
.
8
6

8
3
.
9
1

L
j
u
n
g
-
B
o
x
(
1
2
)
o
n
s
q
u
a
r
e
d
r
e
t
u
r
n
s
7
5
8
.
7
5

4
4
1
.
8
6

3
1
0
.
5
5

3
4
4
.
8
1

1
3
6
.
2
8

1
9
1
.
6
6

N
o
t
e
:
*
i
n
d
i
c
a
t
e
s
s
i
g
n
i

c
a
n
c
e
a
t
t
h
e
5
%
l
e
v
e
l
.
28
T
a
b
l
e
2
:
E
s
t
i
m
a
t
e
s
f
r
o
m
V
A
R
m
o
d
e
l
s
f
o
r
e
x
c
h
a
n
g
e
r
a
t
e
a
n
d
o
i
l
r
e
t
u
r
n
s
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
I
n
d
e
x
P
a
r
a
m
e
t
e
r
s
L
a
g
F
X
O
i
l
F
X
O
i
l
F
X
O
i
l
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
E
x
c
h
a
n
g
e
r
a
t
e

1
.
0
0
3
1
(
.
9
2
8
7
)
.
0
6
3
8
(
.
6
2
6
4
)

.
1
3
2
5
(
.
0
0
0
1
)
.
0
8
7
9
(
.
4
0
5
5
)

.
0
4
2
8
(
.
2
0
1
4
)
.
3
3
4
0
(
.
0
6
6
2
)

.
0
0
4
7
(
.
8
9
1
9
)

.
0
5
3
3
(
.
6
8
4
1
)

.
0
5
6
6
(
.
1
0
3
8
)

.
1
1
5
1
(
.
2
7
8
6
)

.
0
4
8
8
(
.
1
4
5
0
)

.
0
5
1
8
(
.
7
7
5
2
)

3
.
0
0
7
2
(
.
8
3
3
0
)
.
2
9
4
2
(
.
0
2
4
4
)

.
0
1
9
8
(
.
5
7
0
6
)
.
1
6
5
0
(
.
1
2
1
1
)
.
0
1
2
4
(
.
7
1
1
7
)
.
3
6
1
3
(
.
0
4
6
7
)

4
.
0
1
3
6
(
.
6
9
1
0
)

.
0
3
9
6
(
.
7
6
2
3
)

.
0
3
0
7
(
.
3
7
7
3
)
.
0
9
3
7
(
.
3
7
8
2
)
.
0
1
8
9
(
.
5
7
2
4
)
.
0
7
5
8
(
.
6
7
5
8
)

.
0
3
4
9
(
.
3
0
7
6
)
.
1
4
8
3
(
.
2
5
5
9
)

.
0
6
4
3
(
.
0
6
0
2
)

.
0
1
4
8
(
.
8
8
7
3
)

.
0
0
5
5
(
.
8
6
8
0
)
.
0
4
7
8
(
.
7
9
1
0
)
O
i
l

.
0
1
9
8
(
.
0
2
7
2
)

.
0
1
0
8
(
.
7
5
0
9
)

.
0
5
9
5
(
.
0
0
0
0
)

.
0
0
4
8
(
.
8
8
9
6
)

.
0
2
0
3
(
.
0
0
1
0
)
.
0
0
0
2
(
.
9
9
4
6
)

.
0
0
9
5
(
.
2
8
8
3
)

.
0
1
2
1
(
.
7
2
4
7
)

.
0
1
3
8
(
.
2
2
9
0
)

.
0
1
4
8
(
.
6
7
3
2
)

.
0
1
4
5
(
.
0
1
8
9
)

.
0
0
4
9
(
.
8
8
3
9
)

.
0
2
1
7
(
.
0
1
4
8
)
.
1
5
7
0
(
.
0
0
0
0
)

.
0
2
7
0
(
.
0
1
8
3
)
.
1
4
2
8
(
.
0
0
0
0
)

.
0
1
1
6
(
.
0
5
8
3
)
.
1
5
0
7
(
.
0
0
0
0
)

.
0
1
1
6
(
.
1
9
9
2
)

.
0
2
2
9
(
.
5
0
5
1
)

.
0
2
5
1
(
.
0
2
9
2
)

.
0
0
5
0
(
.
8
8
6
3
)

.
0
1
0
8
(
.
0
8
0
3
)

.
0
1
3
3
(
.
6
9
2
4
)

.
0
0
4
4
(
.
6
2
8
3
)

.
0
9
8
7
(
.
0
0
4
1
)

.
0
1
1
4
(
.
3
2
1
7
)

.
1
1
1
6
(
.
0
0
1
5
)

.
0
0
2
4
(
.
6
9
8
9
)

.
1
0
3
3
(
.
0
0
2
1
)
C
o
n
s
t
a
n
t

.
0
1
1
2
(
.
6
3
8
9
)
.
0
1
5
0
(
.
8
6
9
5
)

.
0
0
3
1
(
.
9
1
8
2
)
.
0
1
1
1
(
.
9
0
3
2
)

.
0
1
1
7
(
.
4
8
4
5
)
.
0
1
9
5
(
.
8
3
0
6
)
F
t
e
s
t
s
o
f
l
a
g
s
1
.
8
7
6
(
.
0
4
4
8
)
3
.
9
3
(
.
0
0
0
0
)
4
.
8
2
9
(
.
0
0
0
0
)
3
.
8
1
(
.
0
0
0
0
)
2
.
6
1
4
(
.
0
0
3
9
)
3
.
9
9
9
(
.
0
0
0
0
)
A
R
C
H
t
e
s
t
s

1
C
H
(
1
)
1
8
.
5
4
9
8
(
.
0
0
0
0
)
5
8
.
1
9
8
5
(
.
0
0
0
0
)
5
.
9
2
5
4
(
.
0
1
4
9
)
5
6
.
1
6
1
4
(
.
0
0
0
0
)
3
.
6
3
0
9
(
.
0
5
6
7
)
5
3
.
6
6
1
(
.
0
0
0
0
)

1
C
H
(
6
)
8
7
.
2
0
3
1
(
.
0
0
0
0
)
1
6
4
.
3
8
2
6
(
.
0
0
0
0
)
5
4
.
8
0
8
8
(
.
0
0
0
0
)
1
6
4
.
6
6
1
9
(
.
0
0
0
0
)
4
4
.
1
7
2
(
.
0
0
0
0
)
1
5
4
.
9
8
9
8
(
.
0
0
0
0
)

1
C
H
(
1
2
)
1
6
7
.
4
0
6
6
(
.
0
0
0
0
)
1
8
1
.
5
9
2
1
(
.
0
0
0
0
)
1
0
6
.
9
2
7
6
(
.
0
0
0
0
)
1
8
3
.
7
7
0
1
(
.
0
0
0
0
)
7
9
.
5
2
0
4
(
.
0
0
0
0
)
1
7
4
.
9
2
9
5
(
.
0
0
0
0
)
29
T
a
b
l
e
3
:
E
s
t
i
m
a
t
e
s
f
r
o
m
V
A
R
m
o
d
e
l
s
f
o
r
e
x
c
h
a
n
g
e
r
a
t
e
s
a
n
d
o
i
l
r
e
t
u
r
n
s
(
c
o
n
t
.
)
U
S
D
/
E
U
R
U
S
D
/
I
N
R
P
a
r
a
m
e
t
e
r
s
L
a
g
F
X
O
i
l
F
X
O
i
l
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
C
o
e
f
.
p
-
v
a
l
u
e
E
x
c
h
a
n
g
e
r
a
t
e

.
0
0
9
5
(
.
7
7
4
9
)
.
1
8
3
8
(
.
1
9
2
2
)

.
1
0
4
8
(
.
0
0
1
3
)

.
1
2
3
0
(
.
4
7
7
0
)

.
0
5
8
3
(
.
0
7
8
7
)

.
0
6
0
0
(
.
6
7
0
1
)

.
0
0
0
5
(
.
9
8
9
0
)
.
3
0
6
5
(
.
0
7
8
1
)

3
.
0
2
8
7
(
.
3
8
6
2
)
.
2
1
5
3
(
.
1
2
6
6
)
.
0
2
8
9
(
.
3
7
6
1
)
.
3
5
7
6
(
.
0
4
0
0
)

4
.
0
2
2
7
(
.
4
9
2
8
)

.
0
1
8
6
(
.
8
9
4
9
)
.
0
2
0
8
(
.
5
2
5
6
)
.
0
8
5
7
(
.
6
2
3
4
)

.
0
1
5
0
(
.
6
4
9
4
)

.
0
4
6
9
(
.
7
3
8
3
)
.
0
1
8
7
(
.
5
6
5
4
)
.
0
4
8
0
(
.
7
8
2
2
)
O
i
l

.
0
1
5
4
(
.
0
4
7
9
)

.
0
0
4
9
(
.
8
8
2
4
)
.
0
0
1
5
(
.
8
0
5
8
)

.
0
2
5
6
(
.
4
2
7
0
)

.
0
1
3
6
(
.
0
7
9
9
)

.
0
0
6
3
(
.
8
4
9
8
)

.
0
0
3
6
(
.
5
5
5
8
)
.
0
0
7
0
(
.
8
2
6
9
)

.
0
0
9
4
(
.
2
2
5
5
)
.
1
4
3
0
(
.
0
0
0
0
)

.
0
0
8
2
(
.
1
7
2
5
)
.
1
3
8
2
(
.
0
0
0
0
)

.
0
1
0
5
(
.
1
7
8
5
)

.
0
2
2
4
(
.
4
9
9
9
)

.
0
0
4
3
(
.
4
7
7
0
)

.
0
2
2
1
(
.
4
9
1
8
)

5
.
0
0
0
8
(
.
9
1
6
6
)

.
1
1
4
9
(
.
0
0
0
5
)

.
0
0
1
4
(
.
8
1
7
5
)

.
1
1
3
2
(
.
0
0
0
4
)
C
o
n
s
t
a
n
t

.
0
1
4
3
(
.
5
0
6
4
)
.
0
1
1
1
(
.
5
1
5
4
)
.
0
0
3
2
(
.
9
7
2
2
)
F
t
e
s
t
s
o
f
l
a
g
s
1
.
5
7
(
.
1
1
0
3
)
3
.
6
6
4
(
.
0
0
0
0
)
1
.
5
5
3
(
.
1
1
5
7
)
4
.
0
1
9
(
.
0
0
0
0
)
A
R
C
H
t
e
s
t
s

1
C
H
(
1
)
2
.
8
6
4
7
(
.
0
9
0
5
)
5
6
.
9
8
5
(
.
0
0
0
0
)
1
8
.
4
2
1
1
(
.
0
0
0
0
)
5
8
.
4
7
0
9
(
.
0
0
0
0
)

1
C
H
(
6
)
8
3
.
8
6
3
9
(
.
0
0
0
0
)
1
6
2
.
7
1
3
3
(
.
0
0
0
0
)
6
1
.
1
2
4
6
(
.
0
0
0
0
)
1
6
2
.
9
0
3
5
(
.
0
0
0
0
)

1
C
H
(
1
2
)
1
3
2
.
9
4
1
3
(
.
0
0
0
0
)
1
8
0
.
2
0
1
1
(
.
0
0
0
0
)
6
4
.
4
4
1
8
(
.
0
0
0
0
)
1
8
1
.
7
4
6
3
(
.
0
0
0
0
)
30
Table 4: Posterior estimates of CCC-MSV models for exchange rates and oil returns
Para. Stat. USD/CAD USD/NOK USD Index USD/EUR USD/INR
j
e
'ca: 1.2697 .5860 1.7533 1.3465 2.537
95%C1 (1.83, .09) (1.07, .21) (2.22, 1.23) (1.86, .85) (3.24, 1.91)
c
ee
'ca: .9728 .9139 .9211 .8853 .8439
95%C1 (.93, .99) (.80, .98) (.72, .98) (.52, .98) (.75, .91)
c
eo
'ca: .0219 .0757 .0681 .1026 .1393
95%C1 (.00, .05) (.02, .18) (.05, .22) (.01, .45) (.07, .22)
o
e
'ca: .1107 .1369 .1264 .1277 .6255
95%C1 (.03, .17) (.10, .20) (.06, .33) (.08, .24) (.07, .77)
j 'ca: .3355 .3470 .2955 .2680 .0938
95%C1 (.39, .28) (.41, .29) (.38, .22) (.32, .21) (.16, .02)
j
o
'ca: 1.3987 1.6157 1.4548 1.4556 1.2702
95%C1 (.26, 1.84) (1.10, 2.00) (1.05, 1.89) (.94, 1.32) (.69, 1.91)
c
oo
'ca: .9524 .9005 .9713 .9272 .9855
95%C1 (.02, .98) (.78, .97) (.94, .99) (.81, .99) (.97, .99)
c
oe
'ca: .0392 .0953 .0240 .0698 .0071
95%C1 (.01, .1) (.02, .22) (.00, .07) (.00, .19) (.00, .02)
o
o
'ca: .1573 .1333 .1088 .1298 .1167
95%C1 (.03, .22) (.09, .20) (.02, .14) (.08, .20) (.08, .15)
11C 6192 6683 5622 6075 5228
31
Table 5: Posterior estimates of DCC-MSV model for exchange rates and oil returns
Para. Stat. USD/CAD USD/NOK USD Index USD/EUR USD/INR
j
e
'ca: 1.2314 .6848 1.8780 1.2757 2.3684
95%C1 (1.77, .78) (1.13, .2) (2.28, 1.41) (1.75, .74) (3.14, 1.77)
c
ee
'ca: .9658 .6690 .9386 .8733 .8434
95%C1 (.91, .99) (.32, .85) (.8, .98) (.62, .99) (.76, .91)
c
eo
'ca: .0283 .3222 .0545 .1138 .1413
95%C1 (.00, .07) (.14, .70) (.01, .17) (.01, .35) (.06, .24)
o
e
'ca: .1441 .2739 .1314 .1475 .6227
95%C1 (.1, .25) (.15, .43) (.08, .25) (.07, .26) (.49, .78)
j
o
'ca: 1.5033 1.5584 1.4955 1.6152 1.4524
95%C1 (1.01, 1.91) (1.13, 2.04) (1.11, 1.94) (.23, 2.05) (.81, 1.99)
c
oo
'ca: .9641 .9205 .9576 .909 .9887
95%C1 (.89, .99) (.8, .98) (.91, .98) (.67, .98) (.97, .99)
c
oe
'ca: .0283 .0735 .0349 .0912 .0042
95%C1 (.00, .11) (.01, .20) (.01, .08) (.01, .37) (.00, .02)
o
o
'ca: .1289 .0911 .1209 .1085 .1022
95%C1 (.09, .18) (.04, .13) (.08, .18) (.07, .16) (.07, .14)
. 'ca: .5857 .5521 .4801 .5454 .1243
95%C1 (.77, .32) (.89, .20) (.67, .27) (.69, .3) (.36, .18)
, 'ca: .8672 .9065 .8412 .8855 .9246
95%C1 (.67, .97) (.78, .99) (.53, .96) (.66, .97) (.76, .98)
o
q
'ca: .1219 .1710 .1834 .12 .1607
95%C1 (.06, .26) (.07, .32) (.06, .43) (.03, .21) (.09, .28)
11C 6173 6636 5567 6057 5191
32
T
a
b
l
e
6
:
E
s
t
i
m
a
t
e
s
o
f
C
C
C
-
M
G
A
R
C
H
(
1
,
1
)
m
o
d
e
l
P
a
r
a
.
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
I
n
d
e
x
U
S
D
/
E
U
R
U
S
D
/
I
N
R
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
a
1
0
.
0
0
0
5
(
.
4
2
9
7
)
.
0
2
0
6
(
.
3
8
9
3
)
.
0
0
0
0
(
.
4
9
9
8
)
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
0
9
(
.
3
0
7
3
)
a
2
0
.
1
5
6
2
(
.
0
0
0
0
)
.
0
6
1
0
(
.
0
7
9
8
)
.
0
4
9
1
(
.
0
2
4
6
)
.
1
1
9
7
(
.
0
0
0
0
)
.
0
6
7
1
(
.
0
0
0
4
)
a
1
1
.
0
6
0
9
(
.
0
0
0
0
)
.
0
0
0
0
(
.
0
1
3
9
)
.
0
0
0
0
(
.
4
9
4
9
)
.
0
3
1
9
(
.
0
0
0
0
)
.
1
1
7
4
(
.
0
0
0
0
)
a
1
2
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
2
4
(
.
4
9
6
2
)
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
0
0
(
.
5
0
0
0
)
a
2
1
.
1
7
8
1
(
.
0
0
0
0
)
.
2
1
2
7
(
.
1
6
5
3
)
.
6
2
7
6
(
.
2
7
0
0
)
.
1
5
1
1
(
.
0
0
0
0
)
.
1
2
3
7
(
.
0
0
0
0
)
a
2
2
.
0
9
4
4
(
.
0
8
7
8
)
.
0
4
0
0
(
.
4
2
0
8
)
.
0
3
5
9
(
.
2
5
0
8
)
.
0
9
1
9
(
.
1
0
7
1
)
.
0
6
1
7
(
.
0
4
7
8
)
/
1
1
.
9
2
3
0
(
.
0
0
0
0
)
.
0
0
1
6
(
.
0
1
3
7
)
.
0
1
0
7
(
.
4
9
1
1
)
.
9
4
1
0
(
.
0
0
0
5
)
.
8
6
4
5
(
.
0
0
0
0
)
/
1
2
.
0
0
1
2
(
.
4
9
8
6
)
.
1
0
4
1
(
.
4
9
4
4
)
.
0
3
4
5
(
.
4
9
9
3
)
.
0
0
1
5
(
.
4
9
9
1
)
.
0
0
1
1
(
.
4
9
8
5
)
/
2
1
.
0
9
7
4
(
.
0
0
0
0
)
.
0
7
3
9
(
.
0
0
0
0
)
.
7
4
5
8
(
.
0
0
0
0
)
.
6
7
9
5
(
.
0
0
0
0
)
.
0
0
0
0
(
.
4
9
9
9
)
/
2
2
.
8
6
1
9
(
.
0
0
0
0
)
.
9
1
8
2
(
.
2
4
3
7
)
.
9
0
8
4
(
.
1
1
1
2
)
.
8
4
2
0
(
.
0
0
0
0
)
.
9
2
3
2
(
.
0
0
0
0
)
j

.
3
2
1
8
(
.
0
0
0
0
)

.
3
4
3
3
(
.
0
0
0
0
)

.
2
8
5
8
(
.
0
0
0
0
)

.
2
7
0
6
(
.
0
0
0
0
)

.
1
0
0
2
(
.
0
0
0
0
)
|
o
q
|
i
/

3
0
8
7
0

1
1
2
1
6
8

2
0
9
6
8
1

7
4
4
6
2

2
8
6
9
3
33
T
a
b
l
e
7
:
E
s
t
i
m
a
t
e
s
o
f
D
C
C
-
M
G
A
R
C
H
(
1
,
1
)
m
o
d
e
l
P
a
r
a
.
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
I
n
d
e
x
U
S
D
/
E
U
R
U
S
D
/
I
N
R
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
M
e
a
n
p
-
v
a
l
u
e
a
1
0
.
0
0
0
0
(
.
4
9
9
9
)
.
0
2
8
7
(
.
3
1
6
8
)
.
0
0
0
4
(
.
4
7
9
5
)
.
0
0
3
1
(
.
4
5
3
4
)
.
0
0
0
4
(
.
4
4
0
9
)
a
2
0
.
0
8
6
2
(
.
0
0
0
0
)
.
0
0
4
1
(
.
4
3
2
8
)
.
0
4
1
7
(
.
0
2
1
1
)
.
0
2
2
2
(
.
1
4
4
5
)
.
0
6
0
4
(
.
0
2
8
2
)
a
1
1
.
0
6
2
2
(
.
0
0
0
0
)
.
0
0
1
3
(
.
3
7
9
4
)
.
0
0
0
5
(
.
2
8
7
0
)
.
0
0
0
0
(
.
4
9
1
8
)
.
1
1
7
4
(
.
0
0
0
0
)
a
1
2
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
2
9
(
.
4
8
8
2
)
.
0
0
0
0
(
.
5
0
0
0
)
.
0
0
0
0
(
.
4
9
9
8
)
.
0
0
0
0
(
.
5
0
0
0
)
a
2
1
.
1
4
8
7
(
.
0
0
0
0
)
.
2
6
7
5
(
.
3
8
8
2
)
.
6
8
8
4
(
.
0
3
2
5
)
.
4
0
1
2
(
.
3
8
7
8
)
.
0
1
2
9
(
.
3
8
0
8
)
a
2
2
.
0
7
7
3
(
.
0
7
4
6
)
.
0
4
7
2
(
.
4
3
5
1
)
.
0
4
9
0
(
.
2
1
6
4
)
.
0
3
8
3
(
.
2
6
4
0
)
.
0
6
2
6
(
.
0
5
4
4
)
/
1
1
.
9
2
2
5
(
.
0
0
0
0
)
.
4
3
4
9
(
.
0
7
2
7
)
.
6
5
4
9
(
.
1
3
6
7
)
.
1
5
3
1
(
.
3
8
7
5
)
.
8
6
5
7
(
.
0
0
0
0
)
/
1
2
.
0
0
1
2
(
.
4
9
8
4
)
.
0
5
4
3
(
.
4
9
6
5
)
.
0
1
2
0
(
.
4
9
9
4
)
.
0
4
6
9
(
.
4
9
9
0
)
.
0
0
1
1
(
.
4
9
8
6
)
/
2
1
.
0
3
4
5
(
.
1
0
7
0
)
.
5
9
5
5
(
.
0
0
0
0
)
1
.
4
0
0
1
(
.
0
0
0
0
)
.
8
9
1
9
(
.
0
0
0
0
)
.
1
3
3
5
(
.
0
0
0
0
)
/
2
2
.
8
9
7
4
(
.
0
0
0
0
)
.
8
5
9
0
(
.
1
0
5
6
)
.
8
7
1
9
(
.
0
0
6
5
)
.
8
8
6
3
(
.
2
0
4
0
)
.
9
2
2
3
(
.
0
0
0
0
)
c
.
0
0
7
8
(
.
0
5
9
1
)
.
0
2
3
9
(
.
0
4
1
1
)
.
0
3
1
7
(
.
0
6
6
8
)
.
0
2
3
9
(
.
0
6
7
0
)
.
0
0
6
5
(
.
0
9
8
5
)
,
.
9
8
8
6
(
.
0
0
0
0
)
.
9
3
4
6
(
.
0
0
0
0
)
.
8
6
7
2
(
.
0
0
0
0
)
.
9
2
8
7
(
.
0
0
0
0
)
.
9
9
0
7
(
.
0
0
0
0
)
|
o
q
|
i
/

3
5
5
8
.
8
6
3

4
0
5
9
.
2
7
2

2
9
4
6
.
3
8
6

3
4
1
5
.
7
3
9

2
9
6
9
.
2
2
2
34
T
a
b
l
e
8
:
M
S
V
m
o
d
e
l
c
h
e
c
k
i
n
g
P
a
r
a
.
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
I
n
d
e
x
U
S
D
/
E
U
R
U
S
D
/
I
N
R
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
C
C
C
-
M
S
V
M
e
a
n

.
0
1
1
1
.
0
2
3
8

.
0
1
3
7
.
0
2
3
2

.
0
0
7
5
.
0
2
5
1

.
0
1
1
7
.
0
2
7
1

.
0
1
0
4
.
0
2
7
5
S
t
d
.
9
5
7
6
.
9
7
3
7
.
9
8
7
1
.
9
8
2
7
.
9
3
3
8
.
9
8
9
7
.
9
7
7
8
.
9
9
9
4
.
9
9
8
8
.
9
7
8
9
S
k
e
w
n
e
s
s

.
0
8
1
6
.
0
0
3
4
.
0
4
5
0
.
0
0
8
9

.
1
6
6
3
.
0
3
5
1

.
1
0
3
7
.
0
2
8
6
.
1
0
5
4
.
0
0
7
1
K
u
r
t
o
s
i
s
3
.
0
8
7
2
.
6
5
3
9
3
.
2
3
2
1
2
.
7
7
6
6
3
.
4
4
7
6
2
.
7
8
5
0
3
.
3
4
8
5
2
.
7
5
7
7
3
.
2
3
9
3
2
.
7
7
7
9
K
o
l
m
o
g
o
r
o
v
-
S
m
i
r
n
o
v
.
0
2
8
6
.
0
1
9
7
.
0
2
3
3
.
0
2
1
4
.
0
3
8
3
.
0
2
2
7
.
0
3
0
1
.
0
2
1
4
.
0
2
5
4
.
0
1
9
9
(
j

a
|
n
c
)
(
.
3
8
9
1
)
(
.
8
3
4
8
)
(
.
6
5
6
1
)
(
.
7
5
1
5
)
(
.
1
0
9
2
)
(
.
6
8
6
)
(
.
3
3
)
(
.
7
5
2
)
(
.
5
4
5
7
)
(
.
8
2
7
3
)
J
a
r
q
u
e
-
B
e
r
a
1
.
4
7
0
2
4
.
8
1
3
0
2
.
6
9
1
8
1
.
9
7
6
1
1
3
.
1
4
2
0
2
.
0
1
8
7
7
.
0
0
4
2
.
4
5
4
1
4
.
3
4
4
7
1
.
9
4
7
1
(
j

a
|
n
c
)
(
.
4
7
9
4
)
(
.
0
9
0
1
)
(
.
2
6
0
3
)
(
.
3
7
2
3
)
(
.
0
0
1
4
)
(
.
3
6
4
5
)
(
.
0
3
0
1
)
(
.
2
9
3
2
)
(
.
1
1
3
9
)
(
.
3
7
7
7
)
D
I
C
6
2
7
5
6
7
5
4
5
6
9
6
6
1
4
6
5
4
3
9
D
C
C
-
M
S
V
M
e
a
n

.
0
1
2
6
.
0
2
3
8

.
0
1
6
5
.
0
2
3
6

.
0
0
9
9
.
0
2
4
9

.
0
1
3
0
.
0
2
5
8

.
0
1
0
4
.
0
2
6
4
S
t
d
.
9
7
8
6
.
9
8
3
3
1
.
0
0
1
4
.
9
9
8
1
1
.
0
0
0
4
.
9
9
0
0
.
9
8
1
7
.
9
7
9
1
.
9
9
8
2
.
9
7
6
1
S
k
e
w
n
e
s
s

.
1
0
1
3
.
0
1
6
5
.
0
2
6
5
.
0
3
1
1

.
1
4
9
9
.
0
4
0
6

.
0
8
7
2
.
0
3
4
6
.
1
2
0
9
.
0
1
8
6
K
u
r
t
o
s
i
s
2
.
9
1
9
2
2
.
7
3
3
1
3
.
2
4
8
0
2
.
8
9
1
5
3
.
2
5
6
6
2
.
7
9
3
8
3
.
2
3
0
0
2
.
8
1
4
5
3
.
2
7
5
1
2
.
8
0
0
2
K
o
l
m
o
g
o
r
o
v
-
S
m
i
r
n
o
v
.
0
2
2
4
.
0
2
1
7
.
0
1
9
5
.
0
1
9
6
.
0
2
2
4
.
0
2
4
.
0
3
0
5
.
0
1
8
6
.
0
2
7
.
0
1
7
4
(
j

a
|
n
c
)
(
.
6
9
9
8
)
(
.
7
3
6
4
)
(
.
8
4
5
8
)
(
.
8
3
8
5
)
(
.
7
0
3
6
)
(
.
6
1
4
5
)
(
.
3
1
5
3
)
(
.
8
8
0
8
)
(
.
4
6
3
9
)
(
.
9
2
4
5
)
J
a
r
q
u
e
-
B
e
r
a
1
.
9
3
7
2
2
.
8
7
2
2
2
.
7
9
8
4
.
5
9
6
6
6
.
5
9
3
5
1
.
9
3
8
9
3
.
5
7
7
8
1
.
5
3
6
6
5
.
7
0
8
8
1
.
6
1
7
1
(
j

a
|
n
c
)
(
.
3
7
9
6
)
(
.
2
3
7
9
)
(
.
2
4
6
8
)
(
.
7
4
2
1
)
(
.
0
3
7
0
)
(
.
3
7
9
3
)
(
.
1
6
7
1
)
(
.
4
6
3
8
)
(
.
0
5
7
6
)
(
.
4
4
5
5
)
D
I
C
6
2
7
5
6
7
4
9
5
6
6
4
6
1
4
2
5
4
1
7
35
T
a
b
l
e
9
:
M
G
A
R
C
H
m
o
d
e
l
c
h
e
c
k
i
n
g
P
a
r
a
.
U
S
D
/
C
A
D
U
S
D
/
N
O
K
U
S
D
I
n
d
e
x
U
S
D
/
E
U
R
U
S
D
/
I
N
R
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
E
x
.
r
a
t
e
O
i
l
C
C
C
-
M
G
A
R
C
H
M
e
a
n

.
0
1
1
0
.
0
1
9
3

.
0
1
0
1
.
0
1
9
3

.
0
1
2
9
.
0
1
9
7

.
0
1
2
5
.
0
1
9
9

.
0
1
0
5
.
0
1
7
5
S
t
d
.
9
9
6
6
.
9
9
8
2
.
9
9
8
9
.
9
9
8
9
.
9
9
9
9
.
9
9
6
2
.
9
9
8
1
.
9
9
8
9
.
9
9
5
9
.
9
9
6
4
S
k
e
w
n
e
s
s

.
1
4
0
8
.
0
4
7
8
.
0
3
2
9
.
0
6
6
2

.
2
5
0
6
.
1
0
2
3

.
1
4
7
5
.
0
6
2
4

.
2
5
5
4
.
0
4
7
9
K
u
r
t
o
s
i
s
3
.
6
3
8
6
3
.
1
8
8
1
3
.
7
0
9
9
3
.
2
0
1
5
4
.
0
6
3
1
3
.
4
5
8
9
3
.
8
3
2
3
.
2
0
0
7
8
.
0
5
6
2
3
.
3
5
2
7
K
o
l
m
o
g
o
r
o
v
-
S
m
i
r
n
o
v
.
0
3
0
2
.
0
2
0
2
.
0
2
7
1
.
0
1
9
3
.
0
3
7
1
.
0
2
1
4
.
0
3
4
1
.
0
1
9
3
.
0
7
6
8
.
0
2
4
6
(
j

a
|
n
c
)
(
.
3
2
6
1
)
(
.
8
1
2
3
)
(
.
4
5
7
8
)
(
.
8
5
3
6
)
(
.
1
2
9
5
)
(
.
7
5
3
6
)
(
.
1
9
8
9
)
(
.
8
5
4
8
)
(
.
0
0
0
0
)
(
.
5
8
2
5
)
J
a
r
q
u
e
-
B
e
r
a
2
0
.
5
7
1
9
1
.
9
4
6
6
2
1
.
4
9
5
1
2
.
5
1
8
5
5
7
.
9
7
3
2
1
0
.
7
2
5
9
3
2
.
8
3
5
3
2
.
4
2
3
2
1
0
7
.
6
4
7
2
5
.
7
3
4
1
(
j

a
|
n
c
)
(
.
0
0
0
0
)
(
.
3
7
7
8
)
(
.
0
0
0
0
)
(
.
2
8
3
9
)
(
.
0
0
0
0
)
(
.
0
0
4
7
)
(
.
0
0
0
0
)
(
.
2
9
7
7
)
(
.
0
0
0
0
)
(
.
0
5
6
9
)
L
o
g
l
i
k

3
0
,
8
7
0

7
0
4
,
4
0
2

2
0
9
,
6
8
1

7
4
,
4
6
2

2
8
,
6
9
3
D
C
C
-
M
G
A
R
C
H
M
e
a
n

.
0
1
1
1
.
0
1
9
5

.
0
1
0
9
.
0
1
9
7

.
0
1
3
4
.
0
2
0
7

.
0
1
3
2
.
0
2
0
5

.
0
1
0
7
.
0
1
8
3
S
t
d
.
9
9
5
4
.
9
9
7
2
.
9
9
5
4
.
9
9
8
1
.
9
9
9
8
.
9
9
8
4
.
9
9
8
2
.
9
9
9
5
.
9
9
5
6
.
9
9
6
2
S
k
e
w
n
e
s
s

.
1
3
8
7
.
0
5
8
1
.
0
3
2
8
.
0
7
4
3

.
2
2
5
8
.
0
9
4
2

.
1
6
3
1
.
1
0
4
5

.
2
4
9
9
.
0
5
5
2
K
u
r
t
o
s
i
s
3
.
6
2
1
2
3
.
2
2
2
6
3
.
6
8
7
7
3
.
2
9
2
4
3
.
9
8
7
1
3
.
3
6
5
2
3
.
8
1
1
6
3
.
4
3
3
1
8
.
0
2
3
0
3
.
3
6
3
3
K
o
l
m
o
g
o
r
o
v
-
S
m
i
r
n
o
v
.
0
3
1
1
.
0
2
0
2
.
0
2
7
7
.
0
1
9
9
.
0
3
9
6
.
0
2
0
8
.
0
3
4
6
.
0
2
0
0
.
0
7
5
8
.
0
2
5
1
(
j

a
|
n
c
)
(
.
2
9
0
2
)
(
.
8
1
2
2
)
(
.
4
3
0
3
)
(
.
8
2
8
1
)
(
.
0
8
8
9
)
(
.
7
8
0
8
)
(
.
1
8
3
7
)
(
.
8
2
1
4
)
(
.
0
0
0
0
)
(
.
5
5
9
1
)
J
a
r
q
u
e
-
B
e
r
a
1
9
.
5
5
7
9
2
.
7
3
4
3
2
0
.
1
9
2
5
4
.
6
2
0
9
4
9
.
4
8
3
1
7
.
2
0
4
6
3
2
.
2
2
5
1
9
.
8
3
3
6
1
0
6
2
.
0
6
1
6
.
1
7
8
7
(
j

a
|
n
c
)
(
.
0
0
0
0
)
(
.
2
5
4
8
)
(
.
0
0
0
0
)
(
.
0
9
9
2
)
(
.
0
0
0
0
)
(
.
0
2
7
3
)
(
.
0
0
0
0
)
(
.
0
0
7
3
)
(
.
0
0
0
0
)
(
.
0
4
5
5
)
L
o
g
l
i
k

3
,
5
5
9
-
4
,
0
5
9

2
,
9
4
6

3
,
4
1
6

2
,
9
6
9
36
Table 10: Model forecasting performance
Exchange Rate Oil
Model RMSE MAE RMSE MAE
Value Rank Value Rank Value Rank Value Rank
USD/CAD CCC-MSV .5425 1 .4334 1 10.6509 5 6.8949 2
DCC-MSV .5429 2 .4399 2 10.3945 4 7.2344 3
CCC-MGARCH .8284 4 .6000 4 8.4170 3 8.0047 5
DCC-MGARCH .8312 5 .6023 5 8.2613 2 7.8063 4
GARCH(1,1) .8838 6 .6567 6 5.1838 1 4.3367 1
SV .6104 3 .4979 3 12.5194 6 9.0289 6
USD/NOK CCC-MSV 1.0138 1 .7737 1 10.305 4 6.839 3
DCC-MSV 1.0297 2 .7762 2 10.3557 5 6.7891 2
CCC-MGARCH 1.3279 4 1.0288 4 7.9864 2 7.5625 4
DCC-MGARCH 1.3402 5 1.0373 5 8.3701 3 7.8320 5
GARCH(1,1) 1.5271 6 1.1424 6 5.1838 1 4.3367 1
SV 1.2205 3 1.0062 3 12.5194 6 9.0289 6
USD Index CCC-MSV .4387 1 .2939 1 10.2942 4 6.6903 3
DCC-MSV .4469 2 .2870 2 10.3090 5 6.6867 2
CCC-MGARCH .6778 5 .5185 5 8.3692 2 7.8413 4
DCC-MGARCH .666 4 .5076 4 8.3951 3 7.9133 5
GARCH(1,1) .7069 6 .5472 6 5.1838 1 4.3367 1
SV .4901 3 .4073 3 12.5194 6 9.0289 6
USD/EUR CCC-MSV .7197 2 .4842 2 10.7311 5 6.5977 2
DCC-MSV .7033 1 .5123 1 10.3638 4 6.9997 3
CCC-MGARCH .892 5 .7217 5 8.605 3 8.2355 5
DCC-MGARCH .8711 4 .6982 4 8.1633 2 7.6469 4
GARCH(1,1) .9151 6 .7424 6 5.1838 1 4.3367 1
SV .8189 3 .6783 3 12.5194 6 9.0289 6
USD/INR CCC-MSV .4041 2 .2066 2 10.7792 5 6.7188 2
DCC-MSV .4019 1 .2059 1 10.6686 4 6.8429 3
CCC-MGARCH .6541 5 .4558 4 7.4508 2 6.9134 4
DCC-MGARCH .6557 6 .4563 5 7.5285 3 7.0149 5
GARCH(1,1) .6530 4 .4598 6 5.1838 1 4.3367 1
SV .4468 3 .2671 3 12.5194 6 9.0289 6
37