ARCH and MGARCH models
Christopher F Baum
EC 823: Applied Econometrics
Boston College, Spring 2013
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
1 / 38
ARCH models
Single-equation models
ARCH models
Heteroskedasticity can occur in time series models, just as it may in a
cross-sectional context. It has the same consequences: the OLS point
estimates are unbiased and consistent, but their standard errors will be
inconsistent, as will hypothesis test statistics and confidence intervals.
We may prevent that loss of consistency by using
heteroskedasticity-robust standard errors. The NeweyWest or HAC
standard errors available from newey in the OLS context or ivreg2 in
the instrumental variables context will be robust to arbitrary
heteroskedasticity in the error process as well as serial correlation.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
2 / 38
ARCH models
Single-equation models
The most common model of heteroskedasticity employed in the time
series context is that of autoregressive conditional heteroskedasticity,
or ARCH. As proposed by Nobel laureate Robert Engle in 1982, an
ARCH model starts from the premise that we have a static regression
model
yt = 0 + 1 zt + ut
and all of the GaussMarkov assumptions hold, so that the OLS
estimators are BLUE. This implies that Var (ut |Z ) is constant. But even
when this unconditional variance of ut is constant, we may have time
variation in the condiitional variance of ut :
2
E(ut2 |ut1 , ut2 , . . . ) = E(ut2 |ut1 ) = 0 + 1 ut1
so that the conditional variance of ut is a linear function of the squared
value of its predecessor.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
3 / 38
ARCH models
Single-equation models
If the original ut process is serially uncorrelated, the variance
conditioned on a single lag is identical to that conditioned on the entire
history of the series. We can rewrite this as
2
ht = 0 + 1 ut1
where ut = ht vt , vt (0, 1). This formulation represents the
ARCH(1) model, in which a single lagged u 2 enters the ARCH
equation. A higher-order ARCH equation would include additional lags
of u 2 . To ensure a positive variance, 0 > 0 and 1 > 0. When 1 > 0,
the squared errors are positively serially correlated even though the ut
themselves are not.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
4 / 38
ARCH models
Single-equation models
Since we could estimate this equation and derive OLS b which are
BLUE, why should we be concerned about ARCH? First, we could
derive consistent estimates of b which are asymptotically more
efficient than the OLS estimates, since the ARCH structure is no
longer a linear model.
Second, the dynamics of the conditional variance are important in
many contexts: particularly financial models, in which movements in
volatility are themselves important. Many researchers have found
ARCH effects" in higher-frequency financial data, and to the extent to
which they are present, we may want to take advantage of them. We
may test for the existence of ARCH effects in the residuals of a time
series regression by using the command estat archlm. The null
hypothesis is that of no ARCH effects; a rejection of the null implies the
existence of significant ARCH effects, or persistence in the squared
errors.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
5 / 38
ARCH models
Single-equation models
The ARCH model is inherently nonlinear. If we assume that the ut are
distributed Normally, we may use a maximum likelihood procedure
such as that implemented in Statas arch command to jointly estimate
its mean and conditional variance equation.
The ARCH model has been extended to a generalized form which has
proven to be much more appropriate in many contexts. In the simplest
example, we may write
2
ht = 0 + 1 ut1
+ 1 ht1
which is known as the GARCH(1,1) model since it involves a single lag
of both the ARCH term and the conditional variance term. We must
impose the additional constraint that 1 > 0 to ensure a positive
variance.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
6 / 38
ARCH models
Single-equation models
We may also have a so-called ARCH-in-mean model, in which the ht
term itself enters the regression equation. This sort of model would be
relevant if we had a theory that suggests that the level of a variable
might depend on its variance, which may be very plausible in financial
markets contexts or in terms of, say, inflation, where we often presume
that the level of inflation may be linked to inflation volatility. In such
instances we may want to specify a ARCH- or GARCH-in-mean model
and consider interactions of this sort in the conditional mean (level)
equation.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
7 / 38
ARCH models
Alternative GARCH specifications
Alternative GARCH specifications
A huge literature on alternative GARCH specifications exists; many of
these models are preprogrammed in Statas arch command, and
references for their analytical derivation are given in the Stata manual.
One of particular interest is Nelsons (1991) exponential GARCH, or
EGARCH. He proposed:
log ht = +
j tj E tj + tj
j=1
which is then parameterized as a rational lag of two finiteorder
polynomials, just as in Bollerslevs GARCH.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
8 / 38
ARCH models
Alternative GARCH specifications
Advantages of the EGARCH specification include the positive nature of
ht irregardless of the estimated parameters, and the asymmetric
nature of the impact of innovations: with 6= 0, a positive shock will
have a different effect on volatility than will a negative shock, mirroring
findings in equity market research about the impact of bad news and
good news on market volatility. For instance, a simple EGARCH(1,1)
model will provide a variance equation such as
p
log ht = 0 + 1 zt1 + 2 zt1 2/ + 3 log ht1
where zt = t /t , which is distributed as N(0, 1).
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
9 / 38
ARCH models
Alternative GARCH specifications
Nelsons model is only one of several extensions of GARCH that allow
for asymmetry, or consider nonlinearities in the process generating the
conditional variance: for instance, the threshold ARCH model of
Zakoian (1990) and the Glosten et al. model (1993).
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
10 / 38
ARCH models
Implementation
Stata 12 provides a suite of commands to estimate time series models
in the ARCH (Autoregressive Conditional Heteroskedasticity) family.
The command arch is used to estimate single-equation models. Its
options allow the specification of over a dozen models from the
literature, including ARCH, GARCH, ARCH-in-mean, GARCH with
ARMA errors, EGARCH (exponential GARCH), TARCH (threshold
ARCH), GJR (Glosten et al., 1993), SAARCH (simple asymmetric
ARCH), PARCH (power ARCH), NARCH (nonlinear ARCH), APARCH
(asymmetric power ARCH) and NPARCH (nonlinear power ARCH).
Errors may be specified as Gaussian, t, or GED (generalized error
distribution).
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
11 / 38
ARCH models
Implementation
To estimate an ARCH model, you give the arch varname command,
followed by (optionally) the independent variables in the mean
equation and the options indicating the type of model. For instance, to
fit a GARCH(1,1) to the mean regression of cpi on wage,
arch cpi wage, arch(1) garch(1)
It is important to note that a GARCH(2,1) model would be specified
with the option arch(1/2). If the option was given as arch(2), only
the second-order term would be included in the conditional variance
equation.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
12 / 38
ARCH models
Implementation
A test for ARCH effects in a linear regression can be conducted with
the estat archlm command. Using Statas urate dataset of
monthly unemployment rates for several US states:
. webuse urates, clear
. qui reg D.tenn LD.tenn
. estat archlm, lags(3)
LM test for autoregressive conditional heteroskedasticity (ARCH)
lags(p)
chi2
11.195
H0: no ARCH effects
df
Prob > chi2
3
vs.
0.0107
H1: ARCH(p) disturbance
The LM test indicates the presence of significant ARCH effects.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
13 / 38
ARCH models
Implementation
We estimate a GARCH(1,1) model:
. arch D.tenn LD.tenn, arch(1) garch(1) nolog vsquish
ARCH family regression
Sample: 1978m3 - 2003m12
Number of obs
Distribution: Gaussian
Wald chi2(1)
Log likelihood = 127.4172
Prob > chi2
OPG
Std. Err.
P>|z|
=
=
=
310
9.39
0.0022
D.tenn
Coef.
[95% Conf. Interval]
tenn
LD.
_cons
.2129528
-.0155809
.0694996
.0085746
3.06
-1.82
0.002
0.069
.076736
-.0323868
.3491695
.0012251
.1929262
.0675544
2.86
0.004
.0605219
.3253305
.7138542
.0028566
.0923551
.0016481
7.73
1.73
0.000
0.083
.5328415
-.0003736
.894867
.0060868
tenn
ARCH
arch
L1.
garch
L1.
_cons
Following estimation, we may use predict with the variance option
to produce the conditional variance series.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
14 / 38
ARCH models
Implementation
Conditional variance from GARCH(1,1)
Conditional variance, one-step
.1
.08
.06
.04
.02
1980m1
1985m1
Christopher F Baum (BC / DIW)
1990m1
Month
1995m1
ARCH and MGARCH models
2000m1
2005m1
Boston College, Spring 2013
15 / 38
ARCH models
Implementation
We may also fit a model with additional variables in the mean equation:
. arch D.tenn LD.tenn LD.indiana LD.arkansas, arch(1) garch(1) nolog vsquish
ARCH family regression
Sample: 1978m3 - 2003m12
Number of obs
=
310
Distribution: Gaussian
Wald chi2(3)
=
41.31
Log likelihood = 135.1611
Prob > chi2
=
0.0000
D.tenn
Coef.
OPG
Std. Err.
P>|z|
[95% Conf. Interval]
tenn
tenn
LD.
indiana
LD.
arkansas
LD.
_cons
.1459972
.0723994
2.02
0.044
.004097
.2878974
.1751591
.047494
3.69
0.000
.0820727
.2682455
.1170958
-.0078106
.0757688
.0087075
1.55
-0.90
0.122
0.370
-.0314083
-.0248769
.2655999
.0092558
.1627143
.0712808
2.28
0.022
.0230064
.3024221
.6793291
.0042064
.1388493
.0026923
4.89
1.56
0.000
0.118
.4071896
-.0010704
.9514687
.0094832
ARCH
arch
L1.
garch
L1.
_cons
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
16 / 38
ARCH models
Implementation
Following estimation, we may test hypotheses on the coefficients of the
conditional variance equation: for instance, that they sum to unity,
indicating integrated GARCH:
. test [ARCH]L.arch + [ARCH]L.garch == 1
( 1) [ARCH]L.arch + [ARCH]L.garch = 1
chi2( 1) =
2.30
Prob > chi2 =
0.1297
In this case, that hypothesis cannot be rejected at 90%.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
17 / 38
ARCH models
Multiple-equation models
Multiple-equation GARCH models
Multivariate GARCH models allow the conditional covariance matrix of
the dependent variables to follow a flexible dynamic structure and allow
the conditional mean to follow a vector autoregressive (VAR) structure.
The general MGARCH model can be written as
yt
t
= Cxt + t
=
1/2
Ht t
where yt is a m-vector of dependent variables, C is a m k parameter
matrix, xt is a k-vector of explanatory variables, possibly including lags
1/2
of yt , Ht is the Cholesky factor of the time-varying conditional
covariance matrix Ht , and t is a m-vector of zero-mean, unit-variance
i.i.d. innovations.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
18 / 38
ARCH models
Multiple-equation models
In this general framework, Ht is a matrix generalization of univariate
GARCH models. For example, a general MGARCH(1,1)) model may
be written as:
vech(Ht ) = s + A vech(t1 0t1 ) + B vech(Ht1 )
where the vech() function returns a vector containing the unique
elements of its matrix argument. The various parameterizations of
MGARCH provide alternative restrictions on H, the conditional
covariance matrix, which must be positive definite for all t.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
19 / 38
ARCH models
Implementation
Implementation
Statas mgarch command estimates multivariate GARCH models,
allowing both the conditional mean and conditional covariance matrix
to be dynamic. Four commonly used parameterizations are supported:
the diagonal vech (DVECH) model
the constant conditional correlation (CCC) model
the dynamic conditional correlation (DCC) model
the varying conditional correlation (VCC) model
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
20 / 38
ARCH models
Parameterizations
Alternative parameterizations differ in terms of flexibility, allowing for
more complex H processes, and parsimony, allowing the model to be
specified with fewer parameters.
The oldest and simplest parameterization is the diagonal vech
(DVECH) of Bollerslev, Engle, Wooldridge (JPE, 1988), which restricts
the A and B matrices to be diagonal. The number of parameters grows
rapidly with the size of the model. For instance, there are 3m(m + 1)/2
parameters in a DVECH(1, 1) with m series.
Despite the large number of parameters, the diagonal structure implies
that each conditional variance and covariance depends only on its own
past, and not on past values of other elements. For a DVECH(1,1),
hij,t = sij + aij i,t1 j,t1 + bij hij,t1
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
21 / 38
ARCH models
Parameterizations
Conditional correlation models
Conditional correlation (CC) models use nonlinear combinations of
univariate GARCH models to represent the conditional covariances in
H. They often have less difficulty with satisfying the restrictions on the
estimated H, and their number of parameters grows more slowly than
in the DVECH specification.
In CC models, Ht is decomposed into a matrix of conditional
correlations Rt and a diagonal matrix of conditional variances, Dt :
1/2
Ht = Dt
1/2
Rt Dt
implying that hij,t = ij,t i,t j,t , where i,t is modeled as a univariate
GARCH process. The CC models differ in how they parameterize Rt .
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
22 / 38
ARCH models
Parameterizations
The constant CC model of Bollerslev (REStat, 1990) specifies the
correlation matrix as time invariant:
q
hij,t = ij hii,t hjj,t
where the diagonal elements follow univariate GARCH processes, and
ij is a time-invariant weight.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
23 / 38
ARCH models
Parameterizations
Engles (JBES, 2002) extension, the dynamic CC model, allows the
conditional correlations (technically, quasicorrelations) to follow a
GARCH(1,1)-like process:
q
hij,t = ij,t hii,t hjj,t
where now the parameters follow a dynamic process.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
24 / 38
ARCH models
Parameterizations
Tse and Tsuis (JBES, 2002) variant, the varying CC model, expresses
the conditional correlations using a time-invariant component, a
measure of recent correlations among the residuals, and last periods
values. It differs from the DCC model in terms of the dynamic process
followed by the parameters.
In Stata, the four MGARCH specifications are invoked with the mgarch
command, with a first argument being the model specification: dvech,
ccc, dcc or vcc.
To illustrate, we use Statas stocks dataset, and model daily Toyota
and Honda equity returns as AR(1) processes with the ccc and dcc
specifications.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
25 / 38
ARCH models
Parameterizations
The estimated Toyota mean and conditional variance equations:
. webuse stocks, clear
(Data from Yahoo! Finance)
. mgarch ccc (toyota honda = L.toyota L.honda), arch(1) garch(1) nolog vsquish
Constant conditional correlation MGARCH model
Sample: 1 - 2015
Number of obs
=
2014
Distribution: Gaussian
Wald chi2(4)
=
4.34
Log likelihood = 11602.61
Prob > chi2
=
0.3620
Coef.
Std. Err.
P>|z|
[95% Conf. Interval]
toyota
toyota
L1.
honda
L1.
_cons
ARCH_toyota
arch
L1.
garch
L1.
_cons
-.03374
.032697
-1.03
0.302
-.097825
.030345
-.005188
.0004523
.0288975
.0003094
-0.18
1.46
0.858
0.144
-.0618261
-.0001542
.0514502
.0010587
.0661046
.0095018
6.96
0.000
.0474814
.0847279
.916793
4.50e-06
.0117942
1.19e-06
77.73
3.78
0.000
0.000
.8936769
2.17e-06
.9399092
6.83e-06
...
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
26 / 38
ARCH models
Parameterizations
The estimated Honda mean and conditional variance equations, and
correlation estimate:
honda
toyota
L1.
honda
L1.
_cons
ARCH_honda
arch
L1.
garch
L1.
_cons
Correlation
toyota
honda
-.0066352
.0343028
-0.19
0.847
-.0738675
.0605971
-.0332976
.0006128
.0316213
.0003394
-1.05
1.81
0.292
0.071
-.0952743
-.0000524
.028679
.0012781
.0498417
.0080311
6.21
0.000
.0341009
.0655824
.9321435
5.26e-06
.0111601
1.41e-06
83.52
3.73
0.000
0.000
.9102701
2.50e-06
.9540168
8.02e-06
.7176095
.0108477
66.15
0.000
.6963483
.7388707
In this CCC specification, the sizable correlation indicates the
interaction between the two equations error processes.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
27 / 38
ARCH models
Parameterizations
In the DCC model, the diagonal elements of Ht are modeled as
univariate GARCH models. The off-diagonal elements are modeled as
nonlinear functions of the diagonal terms:
q
hij,t = ij,t hii,t hjj,t
where ij,t follows a dynamic process, rather than being constrained to
be constant as in the CCC specification.
Two additional parameters, 1 and 2 , are adjustment parameters that
govern the evolution of the conditional quasicorrelations. They must be
positive and sum to less than one. A test for the sum of these
parameters equalling zero tests the DCC model against the special
case of the CCC model.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
28 / 38
ARCH models
Parameterizations
The DCC model may be written as
yt
= Cxt + t
t
= Ht
1/2
Ht
Rt
Qt
1/2
1/2
Dt Rt Dt
diag(Qt )1/2 Qt diag(Qt )1/2
(1 1 2 )R + 1 t1 0t1 +
2 Qt1
where Dt is a diagonal matrix of conditional variances,
Rt is a matrix of conditional quasicorrelations,
1/2
and t is a vector of standardized residuals, Dt
t .
R is a weighted average of the unconditional VCE of the standardized
residuals and the unconditional mean of Qt .
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
29 / 38
ARCH models
Parameterizations
With the DCC specification:
. mgarch dcc (toyota honda = L.toyota L.honda), arch(1) garch(1) nolog vsquish
Dynamic conditional correlation MGARCH model
Sample: 1 - 2015
Number of obs
=
2014
Distribution: Gaussian
Wald chi2(4)
=
4.81
Log likelihood = 11624.54
Prob > chi2
=
0.3074
Coef.
Std. Err.
P>|z|
[95% Conf. Interval]
toyota
toyota
L1.
honda
L1.
_cons
ARCH_toyota
arch
L1.
garch
L1.
_cons
-.0346653
.0319267
-1.09
0.278
-.0972404
.0279098
-.0069742
.000373
.0284872
.0003108
-0.24
1.20
0.807
0.230
-.0628081
-.0002362
.0488597
.0009821
.0629146
.0093309
6.74
0.000
.0446263
.0812029
.9208039
4.32e-06
.0116908
1.16e-06
78.76
3.72
0.000
0.000
.8978904
2.04e-06
.9437175
6.60e-06
...
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
30 / 38
ARCH models
Parameterizations
honda
toyota
L1.
honda
L1.
_cons
.0030378
.0339118
0.09
0.929
-.0634281
.0695036
-.0367691
.0005624
.0316091
.000341
-1.16
1.65
0.245
0.099
-.0987219
-.0001059
.0251836
.0012307
.0536899
.008511
6.31
0.000
.0370087
.0703711
.928433
5.43e-06
.0115932
1.44e-06
80.08
3.77
0.000
0.000
.9057107
2.61e-06
.9511554
8.26e-06
Correlation
toyota
honda
.7264858
.0132659
54.76
0.000
.7004852
.7524864
Adjustment
lambda1
lambda2
.0528653
.746622
.014217
.0746374
3.72
10.00
0.000
0.000
.0250005
.6003354
.0807301
.8929085
ARCH_honda
arch
L1.
garch
L1.
_cons
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
31 / 38
ARCH models
Parameterizations
In both the CCC and DCC specifications, the mean equations indicate
that lagged daily returns of both stocks are not significant determinants
of current returns, as is implied by efficient markets theory.
There are very significant GARCH effects in both specifications. A
sizable correlation parameter appears, as it did in the CCC
specification. The magnitudes of the lambda parameters indicate that
the evolution of the conditional covariances depends more on their
past values than on lagged residuals innovations.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
32 / 38
ARCH models
Parameterizations
The VCC model of Tse and Tsui can be written as
yt
= Cxt + t
t
= Ht
Ht
= Dt
Rt
= (1 1 2 )R + 1 t1 + 2 Rt1
1/2
1/2
Rt Dt
1/2
where Dt is a diagonal matrix of conditional variances,
Rt is a matrix of conditional correlations,
R is the matrix of means to which the dynamic process reverts, and
t is the rolling estimator of the covariance matrix of the standardized
residuals t .
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
33 / 38
ARCH models
Parameterizations
We illustrate the VCC model with two companies shares, assumed to
have no mean equation per previous findings, but with their ARCH and
GARCH parameters constrained to be equal.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
34 / 38
ARCH models
Parameterizations
. constraint 1 _b[ARCH_toyota:L.arch] = _b[ARCH_nissan:L.arch]
. constraint 2 _b[ARCH_toyota:L.garch] = _b[ARCH_nissan:L.garch]
. mgarch vcc (toyota nissan =, noconstant), arch(1) garch(1) constraints(1 2) n
> olog vsquish
Varying conditional correlation MGARCH model
Sample: 1 - 2015
Number of obs
=
2015
Distribution: Gaussian
Wald chi2(.)
=
.
Log likelihood = 11282.46
Prob > chi2
=
.
( 1) [ARCH_toyota]L.arch - [ARCH_nissan]L.arch = 0
( 2) [ARCH_toyota]L.garch - [ARCH_nissan]L.garch = 0
Coef.
ARCH_toyota
arch
L1.
garch
L1.
_cons
ARCH_nissan
arch
L1.
garch
L1.
_cons
Std. Err.
P>|z|
[95% Conf. Interval]
.0797459
.0101634
7.85
0.000
.059826
.0996659
.9063808
4.24e-06
.0118211
1.10e-06
76.67
3.85
0.000
0.000
.883212
2.08e-06
.9295497
6.40e-06
.0797459
.0101634
7.85
0.000
.059826
.0996659
.9063808
5.91e-06
.0118211
1.47e-06
76.67
4.03
0.000
0.000
.883212
3.03e-06
.9295497
8.79e-06
...
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
35 / 38
ARCH models
Parameterizations
...
Correlation
toyota
nissan
.6720056
.0162585
41.33
0.000
.6401394
.7038718
Adjustment
lambda1
lambda2
.0343012
.7945548
.0128097
.101067
2.68
7.86
0.007
0.000
.0091945
.596467
.0594078
.9926425
The validity of the constraints could be established with a likelihood
ratio test against the unconstrained model.
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
36 / 38
ARCH models
Parameterizations
We can produce predictions of the three series in the conditional VCE,
ex post and ex ante. Notice that the ex ante predictions (beyond the
sample period, ending in day 2015) quickly converge in the absence of
additional information, as these are dynamic forecasts.
. tsappend, add(50)
. predict H*, variance dynamic(2016)
. lab var H_toyota_toyota CV_Toy
. lab var H_nissan_nissan CV_Nis
. lab var H_nissan_toyota CCov_Toy_Nis
. lab var t "Trading Day"
. tsline H* in 1800/l, leg(rows(1)) xline(2015) ylab(,angle(0))
Christopher F Baum (BC / DIW)
ARCH and MGARCH models
Boston College, Spring 2013
37 / 38
ARCH models
Parameterizations
.001
.0008
.0006
.0004
.0002
0
1800
1850
CV_Toy
Christopher F Baum (BC / DIW)
1900
1950
Trading Day
CCov_Toy_Nis
ARCH and MGARCH models
2000
2050
CV_Nis
Boston College, Spring 2013
38 / 38
