ARCH and GARCH Estimation

Dr. Chen, Jo-Hui

 The types of serial correlation discussed so far refer only to the
error term ut. In AR(p) we postulated that ut depends linearly on
the p past errors ut-1, ut-2, ……,ut-p. Another type of serial
correlation is often encountered in time series data, especially
when forecasts are generated. Some forecasters have observed
that the variance of prediction errors is not a constant but differs
from period to period. For instance, when the Federal Reserve
Board switched to controlling money growth rather than the
interest rate, as was done before, interest rates became quite
volatile (that is, they began to vary a great deal around the
mean). Forecast errors associated with interest rate predictions
were thus heteroscedastic.
 A similar heteroscedasticity was observed when exchange r
ate policy switched from fixed exchange rates to flexible ex
change rates. In the latter case, exchange rates fluctuated a
great deal, making their forecast variances larger. In monet
ary theory and the theory of finance, financial asset portfoli
os are functions of the expected means and variances of the
rates of returns. Increased volatility of security prices or rat
es of return are often indicators that the variances are not co
nstant over time. Engle (1982) introduced a new approach
to modeling heteroscedasticity in a time series context.
The structure of this topic

A. The ARCH Specification

B. Estimating ARCH models in EViews
C. Asymmetric ARCH Models
D. EViews Example
A. The ARCH Specification

 Autoregressive Conditional Heteroskedasticity (ARCH) models are

specifically designed to model and forecast conditional variances.
The variance of the dependent variable is modeled as a function of
past values of the dependent variable and independent, or
exogenous variables. In developing an ARCH model, you will have
to provide two distinct specifications—one for the conditional mean
and one for the conditional variance.
A. The ARCH Specification

1. The GARCH(1,1) Model

yt  xt' r   t (1)...... Mean equation

 t2       t21     t21 (2).......The conditioal variance

where

 t2 = the conditional variance equation is a function of three terms

w = the mean.
= News about volatility from the previous period, measured a
st the
2
1
lag of the squared residual from the mean equation. (ARCH t
erm)
2
t 1 = last period’s forecast variance. (GARCH term)
A. The ARCH Specification

1. The GARCH(1,1) Model

 The (1,1) in GARCH(1,1) refers to the presence of a first-order

GARCH term and a first-order ARCH term. An ordinary ARCH
model is a special case of a GARCH specification in which there are
no lagged forecast variances in the conditional variance equation.

 For example, if the asset return was unexpectedly large in either the
upward or the downward direction, then the trader will increase the
estimate of the variance for the next period. This model is consistent
with the volatility clustering often seen in financial returns data, where
large changes in returns are likely to be followed by further large
changes.
A. The ARCH Specification

2. The ARCH-M model

 The x’s in equation (1) represent exogenous or predetermined variables

that are included in the mean equation. If we introduce the conditional
variance into the mean equation, we get the ARCH-in-Mean (ARCH-
M) model:

y t  xt'  r   t2  ~
r  t

 The ARCH-M model is often used in financial applications where the

expected return on an asset is related to the expected asset risk. The
estimated coefficient on the expected risk is a measure of the risk-
return tradeoff.
A. The ARCH Specification

3. The GARCH(p,q) model

 Higher order GARCH models, denoted GARCH(p,q), can be estimated

by choosing either p or q greater than 1. The representation of the
GARCH(p,q) variance is
q p
 t2     i   t2i    j   t2 j
i 1 j 1

where p is the order of the GARCH terms and q is the order of the
ARCH term.

 The error variance at time t is assumed to depend on previous squared

error terms. Also, the variance at time t is conditional on those in
previous periods and hence the term conditional heteroscedasticity.
The ARCH test is one the null hypothesis H0:α1=α2=…..= αq=0.
B. Estimating ARCH models in EViews

a) Quick/Estimate Equation/ARCH
b) Option:
Heteroskedasticity Consistent Covariances: You should use this option if you s
uspect that the residuals are not conditionally normally distributed.
c) The Mean Equation:
You can enter the specification in list form by listing the dependent variable fo
llowed by the regressors. You should add the C to your specification if you wi
sh to include a constant. If your specification includes an ARCH-M term, you
should click on the appropriate radio button in the upper right-hand side of the
dialog.
d) The variance Equation
①Under the ARCH specification label, you should choose the number of AR
CH and GARCH terms.
②In the edit box labeled Variance Regressors, you may optionally list variabl
es you wish to include in the variance specification. Note that EViews will a
lways include a constant as a variance regressor so that you do not need to a
dd C to the list.
B. Estimating ARCH models in EViews

Examples 1

 GARCH(1,1)
To estimate a standard GARCH(1,1) model with no regressors in the mean and
variance equations.
Rt  c   t
 t2       t21     t21

(a) File/Open/workfile/c:\program file\EView4\example file\data\gerus

(b) Quick/Estimate Equation/ARCH
(c) R c
(d) Enter 1 for the number of ARCH terms, and 1 for the number of GARCH
terms, and select GARCH (symmetric) Select None for the ARCH-M
(e) Leave blank the Variance Regressors edit box.
Mean Eq.
B. Estimating ARCH models in EViews

Examples 2

 To estimate the ARCH(4)-M model:

NYS t  r0  r1 DUM t  r2   t   t
 t2    1   t21   2   t2 2   3   t23   4   t24  r3 DUM t

a) Quick/Estimate Equation/ARCH
b) NYS c DUM
c) Enter 4 for the ARCH term and 0 for the GARCH term, and select
GARCH (symmetric).
d) Select std. Dev. for the ARCH-M term
e) enter DUM in the Variance Regressors edit box.
Click
B. Estimating ARCH models in EViews

Examples 3

 To estimate ARCH(3) model:

 Supposed that annual data for the U.S. for 1960-1995 on the Federal Reserve
discount rate (in percent), money supply (M2 in billions of dollars), and Federal
deficit (D) in billions of current dollars. The model relating discount rates (r) to
the money supply (M) and government budget deficits (D) lagged twice is as
follows:
rt  1   2 M t 1   3 M t 2   4 Dt 1   5 Dt 2  ut
 t2   0  1ut21   2 ut22   3ut23
B. Estimating ARCH models in EViews

Examples 3

 Steps:
 The interest equation was first estimated, its estimated residuals squared, and th
e auxiliary regression for the variance estimated. R2 for this regression was 0.12
6, but nR2=3.91 has p-value 0.27, which in not significant for x2 test. Thus, AR
CH(3) is not supported. However, the ARCH(1) term was significant at the lev
el 0.09, so an ARCH(1) specification was tested next. The p-value for this test
was 0.048, which indicates significant at the 5 percent level. Because the defici
t term D2 was insignificant, it was excluded from the model in order to improve
the precision of the other variables and to reduce any multicollinearity that may
be present. The final model is given below with p-values in parentheses:

rˆt  3.38  0.0283M t 1  0.0262 M t 2  0.0347 Dt 1

( 0.01) ( 0.01) ( 0.01) ( 0.01)
R 2  .464
B. Estimating ARCH models in EViews

1. ARCH Estimation Output

 The output from ARCH estimation is divided into two sections:

1) The upper part provides the standard output for the mean equation.
2) The lower part, labeled “Variance Equation” contains the coefficients, standard
errors, z-statistics and p-values for the coefficients of the variance equation. Th
e ARCH parameters correspond toαand the GARCH parameters toβ.
3) Note that measures such as R2 may not be meaningful if there are no regressors
in the mean equation. Here, for example, the R2 is negative.
4) The sum of the ARCH and GARCH coefficient (α+β) is very close to one, indi
cating that volatility shocks are quite persistent.
B. Estimating ARCH models in EViews

2. Working with ARCH Model

 Once your model has been estimated, EViews provides a variety of views and p
rocedures for inference and diagnostic checking.

Views of ARCH Models

Views/actual, fitted, residual
/Conditional SD Graph
/Covariance Matrix
/Coefficient Tests
/Residual Tests/Correlogram-Q-statistics
/Correlogram Squared Residuals
/Histogram-Normality Test
/ARCH LM Test
B. Estimating ARCH models in EViews

2. Working with ARCH Model

 The ARCH LM test statistic is computed from an auxiliary test regressio
n. To test the null hypothesis that there is no ARCH up to order q in the
residuals, we run the regression
 q 
et2   0     s et2 s   vt ,
 s 1 
 where e is the residual. This is a regression of the squared residuals on
a constant and lagged squared residuals up to order q. EViews reports t
wo test statistics for this test regression. The F-statistic is an omitted va
riable test for the joint significance of all lagged squared residuals. The
Obs*R-squared statistic is Engle’s LM test statistic, computed as the nu
mber of observations times the R2 from the test regression. The exact fi
nite sample distribution of the F-statistic under H0 is not known but the
LM test statistic is asymptotically distributed x2(q) under quite general c
onditions.
 3. Working with GARCH Model
(1)Mean equation:
By using ACF, PACF, and Q test, we can evaluate the form of mean equation
via ARMA procedure.
View/Residual Tests/Correlogram Q-statistics

This view can be used to test for remaining serial correlation in the mean equa
tion and to check the specification of the mean equation. If the mean equation
is correctly specified, all Q-statistics should not be significant.
q
(2) Variance equation:  (i )
We can also apply Ljung-Box Q2 test:
Q 2
( q )  T (T  2 ) T i
i 1
and use ACF and PACF to decide the number of lag for GARCH (p, q ).

View/Residual Tests/Correlogram Squared Residuals

This view can be used to test for remaining ARCH in the variance equation an
d to check the specification of the variance equation. If the variance equation
is correctly specified, all Q-statistics should not be significant.
Q-statistics Q 2 -statistics
C. Asymmetric ARCH Models

 For Equities, it is often observed that downward movements in the

market are followed by higher volatilities than upward movements of
the same magnitude. EViews estimates two models that allow for
asymmetric shocks to volatility: TARCH and EGARCH.

Volatility

News
C. Asymmetric ARCH Models

1. The TARCH Model: Threshold ARCH

 t2       t21  r   t21  d t 1     t21

where dt=1 if εt<0 , and 0 otherwise.

 In this model, good news (εt>0 ), and bad new (εt<0 ), have differential
effects on the conditional variance—good news has an impact of α, wh
ile bad news has an impact of α+ r. If r>0 we say that the leverage eff
ect exists. If r≠0, the news impact is asymmetric.
C. Asymmetric ARCH Models

1. The TARCH Model: Threshold ARCH

 For higher order specifications of the TARCH model, EViews

estimates
q p
 t2     i   t2i r   t21  d t 1    j   t2 j
i 1 j 1

1) Quick/Estimate Equation
2) ARCH specification.
3) Rc
4) TARCH

 If the leverage effect term (r), represented by (RESID<0)*ARCH(1) in

the output, is significantly positive but there appears to be no
asymmetric effect.
C. Asymmetric ARCH Models

2. The EGARCH Model

 The specification for the conditional variance is

   
log  t2      log  t21  
 t 1
 t 1

 r t 1
 t 1

 Note that the left-hand side is the log of the conditional variance. This
implies that the leverage effect is exponential, and that forecasts of the
conditional variance are guaranteed to be nonnegative. The presence
of leverage effects can be tested by the hypothesis that r ＞ 0. The im
pact is asymmetric if r≠ 0.
C. Asymmetric ARCH Models

2. The EGARCH Model

 The higher order specifications of EGARCH model:

  t i 
       
p q
log  t2 2
j log  t  j    i  ri t i 

j 1 i 1   t i  t i 

 To estimate an EGARCH model, simply select the EGARCH radio button

under the ARCH specification settings.
 The leverage effect term (r), denoted as RES/SQR[GARCH](1) in the
output, is negative and statistically different from zero, indicating the
existence of the leverage effect in future bonds returns R during the
sample period.
C. Asymmetric ARCH Models

3. Plotting the Estimated News Impact Curve

 Our goal is to plot the volatilityσ2, against z =ε/σthe impact, where

      log   z
   
log  t2  t21 t 1  r z t 1

a) Procs/Make GARCH Variance Series

D. EViews Example

 Modeling the Volatility in the US Dollar/Deutschmark exchange rates

from 1971 to 1993.
 st 
r  log  
1. The dependent variable is the daily nominal return, t  st 1 
where s is the spot rate. The return series clearly shows volatility
clustering, especially early in the sample.

2. GARCH(1,1)-MA(1) model
Rt   0   1   t2   t   t 1
 t2     t21   t21  r  MON t
 The mean equation includes the conditional variance and the errors
are MA(1). The MON series in the conditional variance equation is
a dummy variable for Monday, which is meant to capture the non-
trading period effect during the weekend.
D. EViews Example

3. Test
a) Open/workfile/Example/data/gerus
b) Quick/Estimate Equation/ARCH
c) ARCH specification: R c ma(1)
d) Variance Regressors: mon
e) 11
f) Select Variance for the ARCH-M
g) View/Residual Tests/ARCH LM Test/7
 The top part of the output from testing up to an ARCH(7) is given by
ARCH TEST:
F-statistic 0.1163 Probability 0.9970
Obs*R-squared 0.8836 Probability 0.9964

Indicating that there does not appear to be any ARCH up to order 7.

D. EViews Example

3. Test
h) View/Residual Tests/Historgram-Normality test
 The residuals are highly leptokurtic and the Jarque-Bera (JB) test decisi
vely rejects the normal distribution.
 JB test is a test statistic for testing whether the series is normally distrib
uted. The test statistic measures the difference of the skewness and kur
tosis of the series with those from the normal distribution.

N  k  2  K  3 
2

 Jarque-Bera= S  
6  4 


 Where S is the skewness, K is the kurtosis, and k represents the number

of estimated coefficients used to create the series.
 Under the null hypothesis of a normal distribution, the JB statistic is dis
tributed as x2 with 2 degrees of freedom.
h) Procs/Make Residual Series/Standardized/View/Distribution Graphs/
Quantile-Quantile/normal distribution
 If the residuals are normally distributed, the QQ-plots should lie
on a straight line. The plot shows that it is primarily a few large
outliers that are driving the departure from normality.