Introduction

ARCH Model
GARCH Models
Alternative Specifications

Modeling the Variance: ARCH - GARCH Models

Macro-prudential Surveillance Division

Financial Stability Department

State Bank of Pakistan - Karachi

October 2021

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 1/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Presentation Outline

1 Introduction

2 The ARCH Model

3 Generalized Autoregressive Conditional Hetero. Models

4 GARCH Zoo

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 2/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Introduction

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 3/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Introduction

Measurement of uncertainty or risk is an important concern of

investors
ARCH models are capable of analyzing uncertainty
Conventional econometrics assumes constant variance of error
term (assumption of homoscedasticity)
However, financial time series exhibit ‘wild’ and ‘calm’ periods
negating the assumption of homoscedasticity
The idea of volatility clustering means that large changes in
stock returns lead to further large changes

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 4/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Volatility clustering in daily stock return data

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 5/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Introduction

To model volatility clustering, Robert F. Engle (1982)

developed Autoregressive Conditional Heteroscedasticity
(ARCH) model
ARCH model relaxes the assumption of homscedasticity
Emphasizes the importance of conditional rather than overall
(unconditional) variance of errors
Proposes joint modelling of mean and variance equation
ARCH model assumes variance of error term depends upon
square of laggedresiduals

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 6/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

ARCH Model

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 7/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Autoregressive Conditional Heteroscedasticity (ARCH)

Consider the regression model

Yt = a + β ′ Xt + ut

where β is vector of k × 1 coefficients and Xt is a vector of

k × 1 explanatory variables.
Traditionally, we assume that
ut ∼ N(0, σ 2 )
However, ARCH(1) model assumes that

σt2 = γ0 + γ1 ut−1
2

Hence rather than constant variance, ARCH(1) model

conditions variance upon lagged residuals
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 8/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Autoregressive Conditional Heteroscedasticity (ARCH)

Formally, the ARCH(1) model may be represented as

Yt = a + β ′ Xt + ut
ut |Ωt ∼ N(0, ht2 )
ht2 = γ0 + γ1 ut−1
2

where Ωt represents information set as discussed in lecture 1

on heteroscedasticity.
First and second equations are called mean and variance
equations
Variance equation dictates that large shock leads to rise in
variance
To ensure positivity of variance, estimated γ coefficients must
be positive . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 9/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

ARCH(q) Model
Consider variance equations for ARCH(2), and (q) models
ht2 = 2
γ0 + γ1 ut−1 2
+ γ2 ut−2
ht2 = 2
γ0 + γ1 ut−1 2
+ γ2 ut−2 2
+ γ3 ut−3
...
ht2 = 2
γ0 + γ1 ut−1 2
+ γ2 ut−2 2
+ γ3 ut−3 2
+ ... + γq ut−q
∑q
2
= γ0 + γj ut−j
j=1

Accordingly, general form of ARCH(q) model is given as

Yt = a + β ′ Xt + ut
ut |Ωt ∼ N(0, ht2 )
∑ q
ht2 = γ0 + 2
γj ut−j
j=1

. . . . . . . . . . . . . . . . . . . .
The estimated γ coefficients have to be positive to ensure positive variance
. . . . . . . . . . . . . . . . . . . . 10/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Detecting ARCH Effects: ARCH LM Test

ARCH LM test may be used to test for presence of ARCH

effects through following steps:
Estimate the mean equation Yt = a + β ′ Xt + ut
Compute residuals ût and squared residuals ût2 series
Estimate the auxiliary regression

ût2 = γ0 + γ1 ût−1
2 2
+ γ2 ût−2 2
+ ... + γq ût−q

Compute the test statistic T × R 2 that follows χ2 distribution

with q degrees of freedom
Null hypothesis: H0 : γ0 = γ1 = γ2 = ... = γq = 0
Rejection of null indicates presence of ARCH effects
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 11/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation of ARCH model in EViews

Estimate the mean equation

r_ftse c r_ftse(-1)
Go to View/Residual Diagnostics/Heterosceasticity
Tests/ARCH
To test for ARCH(1) effects, insert 1 at lags box and press OK
Results show that LM stat is significant showing presence of
ARCH(1) effects

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 12/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Result of ARCH LM Test

Heteroskedasticity Test: ARCH

F-statistic 46.84671 Prob. F(1,2607) 0.0000

Obs*R-squared 46.05506 Prob. Chi-Square(1) 0.0000

Test Equation:
Dependent Variable: RESID^2

Variable Coefficient Std. Error t-Statistic Prob.

C 7.62E − 05 3.76E − 06 20.27023 0.0000

RESID^2(-1) 0.132858 0.019411 6.844466 0.0000
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 13/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation of ARCH Model through ML Approach

OLS assumes constant variance of error term and does not

recognizes ARCH effects
Explicit recognition of heteroscedasticity requires joint
estimation of mean and variance equations
This estimation is performed through Maximum Likelihood
estimation procedure
We assume true model and error distribution are known
Initial values of estimate parameters are assumed and
likelihood of data is calculated
Using computer algorithms, different values of parameters are
tried to maximize the likelihood
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 14/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation of ARCH Model in EViews

In EViews workfile, go to Quick/Estimate Equation and

change estimation Method from "LS Least Squares..." to
"ARCH Autoregressive Conditional..."
Enter mean equation that may include AR and MA terms
r_ftse c r_ftse(-1)
To estimate ARCH(1), insert in ARCH box and insert 0 in
GARCH box
Leave other options unchanged and press OK
For higher order ARCH processes, simply enter desired number
of lags in variance equation specification
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 15/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Results of ARCH(1) Estimation

Dependent Variable: R_FTSE

Method: ML ARCH - Normal distribution (BFGS / Marquardt steps)

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000401 0.000178 2.257706 0.0240

R_FTSE(-1) 0.075197 0.019210 3.914550 0.0001

Variance Equation

C 7.39E − 05 2.11E − 06 35.07442 0.0000

RESID(-1)^2 0.161295 0.020232 7.972125 0.0000
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 16/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Results of ARCH(6) Estimation

Dependent Variable: R_FTSE

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000399 0.000162 2.456115 0.0140

R_FTSE(-1) 0.069678 0.019753 3.527472 0.0004

Variance Equation

C 3.52E − 05 2.58E − 06 13.65522 0.0000

RESID(-1)^2 0.080459 0.014865 5.412805 0.0000
RESID(-2)^2 0.131237 0.024881 5.274594 0.0000
RESID(-3)^2 0.107556 0.022741 4.729591 0.0000
RESID(-4)^2 0.081070 0.022648 3.579532 0.0003
RESID(-5)^2 0.089834 0.022985 3.908369 0.0001
RESID(-6)^2 0.123533 0.023890 5.170870 0.0000
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 17/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Retrieving series of conditional variance/SD

To generate plot of conditional variance, go to View/GARCH

Graphs/Conditional Variance Graph in estimation window
To generate series of ARCH variance, go to Proc/Make
GARCH variance series and give name of your choice
Series of ARCH SD can be calculated using the series of
ARCH variance as:
genr sd_arch1=arch1ˆ(1/2)

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 18/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Slightly more mathematical treatment of ARCH model

Consider the mean equation of a stationary process

Yt = a + β ′ Xt + ut

where classical OLS assumes that Var (ut ) = σ 2 is constant.

However, ARCH model recognizes the volatility clustering
phenomena and models it through conditional variance.
In ARCH model, ut is modeled as a combination of a
systematic component and a random component
√
ut = zt ht

where zt ∼ N(0, 1) is a standard normally distributed random

variable and ht is a scaling factor that propagates the impact
of shock. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 19/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Slightly more mathematical treatment of ARCH model

Shock propagation depends upon past value of shocks

2
ht = γ0 + γ1 ut−1

Substituting the value of value of ut in mean equation

√
Yt = a + β ′ Xt + zt ht
√
= a + β ′ Xt + zt γ0 + γ1 ut−1
2

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 20/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Unconditional mean of error term E (ut ) = 0 as E (zt ) = 0

Unconditional variance of error term
γ0
Var (ut ) = E (zt2 )(ht ) = 1−γ 1
To ensure stationarity, we must have γ0 > 0 and 0 < γ1 < 1
If impact of shock is transmitted in a delayed manner, the a
higher order ARCH specification will be preferable
However, estimation of higher order ARCH models could be
technically challenging
GARCH models provide a straightforward solution in such
conditions

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 21/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

GARCH Models

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 22/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Generalized Autoregressive Conditional Hetero. Models

Tim Bollerslev (1986) generalized the variance equation of
ARCH model by inclusion of AR terms
This simply shows that scaling variable of variance can be
influenced by itself apart from being affected by lagged
residuals
General form of GARCH(p,q) model is given as
Yt = a + β ′ Xt + ut
ut |Ωt ∼ N(0, ht )
∑p
∑
q
2
ht = γ0 + δi ht−i + γj ut−j
i=1 j=1

Variance equation of GARCH(1,1) model

2
ht = γ0 + δ1 ht−1 + γ1 ut−1
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 23/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Conversion of GARCH(1,1) into an infinite ARCH process

GARCH(1,1) is a parsimonious alternative to ARCH model with
large number of lags. To show this, consider GARCH(1,1) variance
equation
2
ht = γ0 + δ1 ht−1 + γ1 ut−1
2 2
= γ0 + δ1 (γ0 + δ1 ht−2 + γ1 ut−2 ) + γ1 ut−1
= γ0 + δ1 γ0 + δ12 ht−2 + δ1 γ1 ut−2
2 2
+ γ1 ut−1
= γ0 + δ1 γ0 + δ12 (γ0 + δ1 ht−3 + γ1 ut−3
2 2
) + δ1 γ1 ut−2 2
+ γ1 ut−1
= γ0 + δ1 γ0 + δ12 γ0 + δ13 ht−3 + δ12 γ1 ut−3
2 2
+ δ1 γ1 ut−2 2
+ γ1 ut−1
= γ0 (1 + δ1 + δ12 ) + δ13 ht−3 + γ1 (δ12 ut−3
2 2
+ δ1 ut−2 2
+ ut−1 )
...
∑ j−1 ∞ ∑ j−1 ∞
γ0 γ0
= + δ1∞ ht−T + γ1 2
δ1 ut−j = + γ1 2
δ1 ut−j
1 − δ1 1 − δ1
j=1 . . . . . . . . . . . . . . . j=1.
. . . .
. . . . . . . . . . . . . . . . . . . . 24/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Hence ARCH(∞) can be represented as GARCH(1,1) where

coefficients of MA terms decline geometrically.

Estimating GARCH models is technically easier and

statistically more efficient when compared against higher order
ARCH models

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 25/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation of GARCH models in EViews

In EViews workfile, go to Quick/Estimate Equation and

change estimation Method from "LS Least Squares..." to
"ARCH Autoregressive Conditional..."
Enter mean equation that may include AR and MA terms
r_ftse c r_ftse(-1)

Insert desired number of ARCH and GARCH terms in relevant

boxes
Leave other options unchanged and press OK

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 26/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

GARCH Estimation Results

Dependent Variable: R_FTSE
GARCH = C(3) + C(4)*RESID(-1)^2 + C(5)*GARCH(-1)

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000433 0.000158 2.731784 0.0063

R_FTSE(-1) 0.062542 0.020696 3.021886 0.0025

Variance Equation

C 8.22E − 07 2.42E − 07 3.392755 0.0007

RESID(-1)^2 0.050862 0.006657 7.640484 0.0000
GARCH(-1) 0.940265 0.007970 117.9695 0.0000
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 27/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Alternative Specifications

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 28/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The GARCH in Mean or GARCH-M model

GARCH-M models allow the conditional mean to depend on

its own conditional variance
GARCH-M may facilitate estimation required rate of return
(risk premium) based on conditional heteroscedasticity
GARCH-M models can be linked with asset–pricing models like
the capital asset–pricing models (CAPM) for many financial
applications
General form of GARCH-M(p,q) is given by
Yt = a + β ′ Xt + θht + ut
ut |Ωt ∼ N(0, ht )
∑p
∑
q
2
ht = γ0 + δi ht−i + γj ut−j
i=1 j=1
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 29/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

GARCH-M

Another variant of the GARCH-M model captures risk through

standard deviation rather than variance

√
Yt = a + β ′ Xt + θ ht + ut
ut |Ωt ∼ N(0, ht )
∑p ∑
q
2
ht = γ0 + δi ht−i + γj ut−j
i=1 j=1

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 30/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimating GARCH-M models in EViews

Open GARCH estimation window in EViews and enter mean

equation

Choose appropriate option from the GARCH-M menu

(variance, SD or log(var)) and press OK

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 31/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

GARCH-M Estimation Results

Dependent Variable: R_FTSE

GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*GARCH(-1)

Variable Coefficient Std. Error z-Statistic Prob.

GARCH 5.192169 4.115375 1.261652 0.2071

C 8.81E − 05 0.000316 0.278608 0.7805
R_FTSE(-1) 0.061704 0.020696 2.981425 0.0029

Variance Equation

C 8.59E − 07 2.53E − 07 3.396541 0.0007

RESID(-1)^2 0.052104 0.006802 7.659535 0.0000
GARCH(-1) 0.938595 0.008206 114.3776 0.0000

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 32/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The Threshold GARCH (TGARCH) model

Squaring the residuals makes positive and negative shocks

indistinguishable

Assuming symmetry is a major limitation of ARCH and

GARCH models

Negative shocks (or ‘bad news’) have a larger impact on

volatility than do positive shocks (or ‘good news’)

Threshold GARCH model was introduced by the works of

Zakoian (1990) and Glosten et al. (1993)

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 33/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The Threshold GARCH (TGARCH) model

Main contribution is to correctly estimate the impact of bad

news on volatility

TGARCH adds a multiplicative dummy variable into the

variance equation

Consider the variance equation of TGARCH(1,1)

2 2
ht = γ0 + δht−1 + γut−1 + θut−1 dt−1

where the negative shock dummy dt = 1 if ut < 0 and dt = 0

otherwise.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 34/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The Threshold GARCH (TGARCH) model

Good news has impact γ while bad news has impact γ + θ

If θ is significant and greater than 0, it implies that bad

news/negative shock has more impact on volatility

General form of TGARCH is give as

∑
p ∑
q
2
ht = γ0 + δi ht−i + (γj + θj )ut−j
i=1 j=1

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 35/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimating TGARCH models in EViews

Open GARCH estimation window in EViews and enter mean

equation

Choose appropriate lags for ARCH and GARCH terms

Change Threshold order from 0 to 1 and press OK

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 36/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Results of TGARCH Estimation

Dependent Variable: R_FTSE

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000318 0.000159 1.997651 0.0458

R_FTSE(-1) 0.058004 0.020591 2.816961 0.0048

Variance Equation

C 6.44E − 07 1.75E − 07 3.690089 0.0002

RESID(-1)^2 0.014062 0.006510 2.160235 0.0308
RESID(-1)^2*(RESID(-1)<0) 0.050685 0.009166 5.529872 0.0000
GARCH(-1) 0.953861 0.006301 151.3722 0.0000

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 37/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The exponential GARCH (EGARCH) model

Exponential GARCH uses log specification in variance equation

EGARCH makes the leverage effect exponential rather than

quadratic

This ensures that the estimates of the conditional variance are

non-negative

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 38/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

The exponential GARCH (EGARCH) model

General equation for EGARCH variance equation:

q 
 
∑ 
 t−j  ∑
p
u
log(ht ) = γ0 + ζi √ + δi log(ht−i )
 ht−j 
j=1 i=1

The EGARCH model can accommodate assymetric impact of

negative shocks
q 
 
∑  ∑q ∑
p
 ut−j  ut−j
log(ht ) = γ0 +ζi  √  +ξ √ + δi log(ht−i )
 ht−j  i ht−j i=1
j=1 j=1

If ξi < 0, then then positive shocks (good news) generate less

volatility than negative shocks (bad news)
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 39/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation of EGARCH in EViews

Open GARCH estimation window in EViews and enter mean

equation

Choose appropriate lags for ARCH and GARCH terms

In "Model" menue, change ARCH/GARCH to EGARCH

Change Threshold order from 0 to 1 if EGARCH with

asymmetric effects isrequired

Press OK

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 40/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Estimation results of EGARCH

Dependent Variable: R_FTSE

LOG(GARCH) = C(3) + C(4)*ABS(RESID(-1)/@SQRT(GARCH(-1))) + C(5)
*RESID(-1)/@SQRT(GARCH(-1)) + C(6)*LOG(GARCH(-1))

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000308 0.000156 1.971303 0.0487

R_FTSE(-1) 0.055499 0.020197 2.747828 0.0060

Variance Equation

C(3) −0.153863 0.028280 −5.440702 0.0000

C(4) 0.086418 0.012960 6.667851 0.0000
C(5) −0.044196 0.007397 −5.974427 0.0000
C(6) 0.990901 0.002380 416.2739 0.0000
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 41/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Adding explanatory variables in the mean and variance

equations

So far, we assumed that mean equation comprises only AR(1)

process

Richer specifications of mean equation higher order AR, MA

terms and exogenous exogenous variables are possible

Similarly, exogenous variables can be used to improve the fit of

variance equation

If we use exogenous variables in mean and/or variance

equation, we have to provide values of these exogenous
variables during forecasting exercise
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 42/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Adding explanatory variables in the mean and variance

equations

Open GARCH estimation window in EViews and enter mean

equation
We assume that Gulf war in 1994 affected mean and variance
of rturns
Construct a dummy vriable "gulf" such that gulf = 1 for 1994
and 0 otherwise
Choose appropriate lags for ARCH and GARCH terms
Insert the dummy variable in mean equation and "Variance
regressors" box
Press OK
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 43/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Results of exogenous regressors inclusion

Dependent Variable: R_FTSE

GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*GARCH(-1) + C(7)*GULF

Variable Coefficient Std. Error z-Statistic Prob.

C 0.000463 0.000167 2.769287 0.0056

R_FTSE(-1) 0.062656 0.020694 3.027718 0.0025
GULF −0.000243 0.000514 −0.471862 0.6370

Variance Equation

C 8.07E − 07 2.36E − 07 3.410444 0.0006

RESID(-1)^2 0.049484 0.006494 7.619819 0.0000
GARCH(-1) 0.942280 0.007694 122.4774 0.0000
GULF −3.52E − 07 3.37E − 07 −1.042517 0.2972

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 44/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

References

Asteriou, D., & Hall, S. G. (2011). Applied econometrics

second edition. Hampshire: Palgrave Macmillan.
Brooks, C. (2014). Introductory Econometrics for Finance,
Cambridge University Press, 3rd Edition
Enders, W (2015). Applied Econometric Time Series, 4th
Edition, John Wiley & Sons.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 45/46
FSD - SBP GARCH Models
 Introduction
ARCH Model
GARCH Models
Alternative Specifications

Thank you!

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 46/46
FSD - SBP GARCH Models