APPLIED

ECONOMETRIC TIME
SERIES 3RD ED.

Chapter 3: Modeling Volatility

Chapter 3: Modeling Volatility

WALTER ENDERS, UNIVERSITY OF ALABAMA

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 1

ECONOMIC TIME SERIES:

THE STYLIZED FACTS

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 15000

12500
Trillions of 2005 dollars

10000

7500

5000

2500

0
1950 1960 1970 1980 1990 2000 2010

GDP Potential Consumption Investment

Figure 3.1 Real GDP, Consumption and Investment

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 20

15

10
Percent per year

-5

-10

-15
1950 1960 1970 1980 1990 2000 2010
Figure 3.2 Annualized Growth Rate of Real GDP

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 12.5

10.0

7.5

5.0
percentage change

2.5

0.0

-2.5

-5.0

-7.5

-10.0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Figure 3.3: Percentage Change in the NYSE US 100: (Jan 4, 2000 - July 16, 2012)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 12.5

10.0

7.5

5.0
percentage change

2.5

0.0

-2.5

-5.0

-7.5

-10.0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Figure 3.3: Percentage Change in the NYSE US 100: (Jan 4, 2000 - July 16, 2012)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 16

14

12
percent per year

10

0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

T -bill 5-year

Figure 3.4 Short- and Long-Term Interest Rates

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 2.25

2.00

1.75 Pound
currency per dollar

1.50

1.25

Euro
1.00

0.75
Sw. Franc

0.50
2000 2002 2004 2006 2008 2010 2012
Figure 3.5: Daily Exchange Rates (Jan 3, 2000 - April 4, 2013)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 150

125

100
dollars per barrel

75

50

25

0
2000 2002 2004 2006 2008 2010 2012
Figure 3.6: Weekly Values of the Spot Price of Oil: (May 15, 1987 - Nov 1, 2013)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
ARCH Processes
The GARCH Model

2. ARCH AND GARCH

PROCESSES

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Other Methods
Let t = demeanded daily return. One method is to use 30-
day moving average /30

Implicit volatility

Logs can stabilize volatility

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
One simple strategy is to model the conditional variance as
an AR(q) process using squares of the estimated residuals

In contrast to the moving average, here the weights need

not equal1/30 (or 1/N).

The forecasts are:

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Properties of the Simple ARCH Model
t vt 0 1 t21
Since vt and t-1 are independent:

Et = E[ vt(0 + 1t-12 )]1/2 ] = 0

Et-1t = Et-1vtEt-1 [0 + 1t-12 ]1/2 ] = 0

Et t-i = 0 ( i 0)

Et 2= E[ vt 2(0 + 1t-12 )] = 0 + 1E(t-1) 2

= 0/( 1 - 1 )

Et-1t 2= Et-1 Copyright

[ vt 2(2015 + (
0 John, Wiley
1 &t-1 ) 2
)] = 0 +
Sons, Inc. All rights reserved.
(
1 t-1 ) 2
ARCH Interactions with the Mean
Consider: yt =a0 + a1yt1 + t

Var(yt yt1, yt2, ) = Et1(yt a0 a1yt1)2

= Et1(t)2 = 0 + 1(t1)2

Unconditional Variance:

a0
Since: t 1 a 1 t i
var(yt ) = a12i var( t i )
i
y a
1 i 0 i=0

0 1
=
1 1 1 a
1
2

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Figure 3.7: Simulated ARCH Processes
White Noise Process vt t vt 1 0.8 t21
8
8

0
0

8
0 50 100 8
0 50 100
20
(a) 20
(b)

0
0

20
0 20 40 60 80 100 20
0 20 40 60 80 100
(c)
yt = 0.2yt-1 + t (d)
yt = 0.9yt-1 + t
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Other Processes
q
ARCH(q) t = v t 0 + i t -i
2

i=1

GARCH(p, q) t = vt ht
q p
ht 0 2
i t i i ht i
i 1 i 1

The benefits of the GARCH model should be clear; a high-order

ARCH model may have a more parsimonious GARCH
representation that is much easier to identify and estimate. This is
particularly true since all coefficients must be positive.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Testing For ARCH
Step 1: Estimate the {yt} sequence using the "best fitting" ARMA
model (or regression model) and obtain the squares of the fitted errors .
Consider the regression equation:

t2 a0 a1 t21 a2 t2 2 ...
If there are no ARCH effects a1 = a2 = = 0
All the coefficients should be statistically significant
No simple way to distinguish between various ARCH and GARCH
models

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Testing for ARCH II
Examine the ACF of the squared residuals:
Calculate and plot the sample autocorrelations of the
squared residuals

LjungBox Q-statistics can be used to test for groups of

significant coefficients.

n
Q = T (T + 2)i2 /(T i )
i=1

Q has an asymptotic 2 distribution with n degrees of freedom

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Engle's Model of U.K. Inflation
Let = inflation and r = real wage

t = 0.0257 + 0.334t1 + 0.408t4 0.404t5 + 0.0559rt1 + t

the variance of t is ht = 8.9 x 10-5

t = 0.0328 + 0.162t1 + 0.264t4 0.325t5 + 0.0707rt1 + t

ht = 1.4 x 105 + 0.955(0.4 + 0.3L + 0.2L2 + 0.1L3 )t-1 )2

(8.5 x 10-6) (0.298)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
A GARCH Model of Oil Prices
Volatility Moderation
A GARCH Model of the Spread

4. THREE EXAMPLES OF
GARCH MODELS

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
A GARCH Model of Oil Prices
Use OIL.XLS to create
pt = 100.0*[log(spott) log(spott1)].

The following MA model works well:

pt = 0.127 + t + 0.177t1 + 0.095t3

The McLeodLi (1983) test for ARCH errors using four lags:

t2 a0 a1 t21 a2 t2 2 ...
The F-statistic for the null hypothesis that the coefficients 1 through 4 all
equal zero is 26.42. With 4 numerator and 1372 denominator degrees of
freedom, we reject the null hypothesis of no ARCH errors at any
conventional significance level.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The GARCH(1,1) Model

pt = 0.130 + t + 0.225t1

ht = 0.402 + 0.097 (t1)2+ 0.881ht1

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Volatility Moderation
Use the file RGDP.XLS to construct the growth rate of real U.S. GDP

yt = log(RGDPt/RGDPt1).

A reasonable model is:

yt = 0.005 + 0.371yt1 + t
(6.80) (6.44)

After checking for GARCH errors

yt = 0.004 + 0.398yt1 + t
(7.50) (6.76)
ht = 1.10x104 + 0.182 8.76x105Dt
(7.87) (2.89) (6.14)

The intercept of the variance equation was 1.10 104 prior to 1984Q1 and
experienced a significant decline to 2.22 105 (= 1.10 104 8.76 105)
beginning in 1984Q1.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Figure 3.8: Forecasts of the Spread
6

-2

-4

-6
1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
_____ 1-Step Ahead Forecast - - - - - 2 Conditional Standard Deviations

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 5

A GARCH MODEL OF RISK

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Holt and Aradhyula (1990)
The study examines the extent to which producers in the U.S. broiler (i.e.,
chicken) industry exhibit risk averse behavior.
The supply function for the U.S. broiler industry takes the form:

qt = a0 + a1pet - a2ht - a3pfeedt-1 + a4hatcht-1 + a5qt-4 + 1t

qt = quantity of broiler production (in millions of pounds) in t;

pet = Et-1pt = expected real price of broilers at t
ht = expected variance of the price of broilers in t
pfeedt-1 = real price of broiler feed (in cents per pound) at t-1;
hatcht-1 = hatch of broiler-type chicks in t-1;
1t = supply shock in t;

and the length of the time period is one quarter.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Note the negative effect of the conditional variance of price
on broiler supply.
The timing of the production process is such that feed and
other production costs must be incurred before output is sold
in the market.
Producers must forecast the price that will prevail two months
hence.
The greater pte, the greater the number of chicks that will be
fed and brought to market.
If price variability is very low, these forecasts can be held
with confidence. Increased price variability decreases the
accuracy of the forecasts and decreases broiler supply.
Risk-averse producers will opt to raise and market fewer
broilers when the conditional volatility of price is high.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The Price equation
(1 - 0.511L - 0.129L2 - 0.130L3 - 0.138L4)pt = 1.632 + 2t

ht = 1.353 + 0.1622t-1 + 0.591ht-1

The paper assumes producers use these equations to form

their price expectations. The supply equation:

qt = 2.767pet - 0.521Et-1ht - 4.325pfeedt-1

+ 1.887hatcht-1 + 0.603pt-4 + 1t

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 6: THE ARCH-M MODEL

Engle, Lilien, and Robins let

yt = t + t
where yt = excess return from holding a long-term asset
relative to a one-period treasury bill and t is a time-
varying risk premium: Et1yt = t
The risk premium is:
t = + ht , > 0
q

ht 0 i t i
2

i 1

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Figure 3.9: Simulated ARCH-M Processes

White noise process ht = 0 + 1t-1)2

2 3

0 2

2 1
0 20 40 60 0 20 40 60

(a) (b)

yt = -4 + 4ht + t yt = -1 + ht + t
10 4

5 2

0 0

5 2
0 20 40 60 0 20 40 60

(c) (d)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Diagnostic Checks for Model Adequacy
Forecasting the Conditional Variance

7. ADDITIONAL
PROPERTIES OF GARCH
PROCESS
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Forecasting with the GARCH(1, 1)
1. ET T2 1 hT 1 0 1 T2 1hT
The 1-step ahead forecast can be calculated directly.
2. Variance: t2 vt2 ( 0 1 t21 1ht 1 )
E t2 Evt2 ( 0 1 E t21 1 Eht 1 )

Note : E t21 E ( Et 2 t21 ) Eht 1

E t2 0 (1 1 ) E t21 0 /(1 1 1 )

3. Mult-Step Forecasts: Note that Et t2 j Et ht j

Et ht j 0 1 Et t2 j 1 1 Et ht j 1
Etht+j = 0 + (1 + 1)Etht+j1

Et ht j 0 [1 (1 1 ) (1 1 ) 2 ... (1 1 ) j 1 ] (1 1 ) j ht

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Volatility Persistence
Large values of both 1 and 1 act to increase the conditional
volatility but they do so in different ways.

Figure 3.10: Persistence in the GARCH(1, 1) Model

70

60

50

40

30

20

10

0
25 50 75 100 125 150 175 200 225 250
Model 1 Model 2

ht 1 0.6 t21 0.2ht 1 ht 1 0.2 t21 0.6ht 1

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Assessing the Fit
Standardized residuals:
T
RSS' ( t2 / ht )
T
RSS' v 2
t
t 1 t 1

AIC' = 2 ln L + 2n
SBC' = 2ln L + n ln(T)

where L likelihood function and n is the

number of estimated parameters.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Diagnostic Checks for Model Adequacy
If there is any serial correlation in the standardized
residuals--the {st} sequence--the model of the mean is not
properly specified.
To test for remaining GARCH effects, form the LjungBox
Q-statistics of the squared standardized residuals.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 8

MAXIMUM LIKELIHOOD
ESTIMATION

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Maximum Likelihood and a Regression
Under the usual normality assumption, the log likelihood of
observation t is:

1
(1/ 2) ln(2 ) (1/ 2) ln 2 2 ( y x ) 2

2 2
t t

With T independent observations:

T T 1 T
log L ln( 2 ) ln
2 2
2
(
2 2 t 1 t
y xt )
2

We want to select and 2 so as to maximize L

2 t2 / T xt yt / xt2

Nonlinear Estimation 37
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 The Likelihood Function with ARCH errors
1 t2
For observation t Lt exp
2 ht 2 ht

T T
T
ln L = ln (2 ) 0.5 ln ht 0.5( t2 / ht )
2 t 1 t=1

Now substitute for t and ht

T 1 T
1 T
ln (2 ) 0.5 ln( 0 1 t 1 ) [ yt xt / ( 0 1 t21 )]
2
ln L = 2

2 t= 2 2 t=2

There are no analytic solutions to the first-order conditions for a

maximum.
Nonlinear Estimation 38
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 9

OTHER MODELS OF
CONDITIONAL VARIANCE

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
IGARCH
The IGARCH Model: Nelson (1990) argued that constraining 1 +
1 to equal unity can yield a very parsimonious representation of the
distribution of an assets return.

Etht+1 = 0 + ht
and
Etht+j = j0 + ht

ht 0 (1 1 ) t21 1Lht

ht 0 /(1 1 ) (1 1 ) 1i t21 i
i 0

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
RiskMetrics
RiskMetrics assumes that the continually compounded daily return of
a portfolio follows a conditional normal distribution.

The assumption is that: rt|It-1 ~ N(0, ht)

ht = (t-1)2 + ( 1 - )(ht-1) ; > 0.9

Note: (Sometimes rt-1 is used). This is an IGARCH without an

intercept.

Suppose that a loss occurs when the price falls. If the

probability is 5%, RiskMetrics uses 1.65ht+1 to measure the
risk of the portfolio. The Value at Risk (VaR) is:

VaR = Amount of Position x 1.65(ht+1)1/2 and for k days is

VaR(k) = Amount of Position x 1.65(k ht+1)1/2

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Why Do We Care About ARCH Effects?
1. We care about the higher moments of the distribution.

2. The estimates of the coefficients of the mean are not correctly

estimated if there are ARCH errors. Consider
T
T T 1
ln L = ln (2 ) - ln 2 -
2 2 2 2
(y
t=1
t xt ) 2

T T
T
ln L = ln (2 ) 0.5 ln ht 0.5 [( y xt ) 2 / ht ]
2 t 1 t=1

3. We want to place conditional confidence intervals around our

forecasts
(see next page)

Nonlinear Estimation 42
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Example from Tsay
Suppose rt = 0.00066 0.0247rt-2 + t
ht = 0.00000389 + 0.0799(t-1)2 + 0.9073(ht-1)

Given current and past returns, suppose:

ET(rT+1) = 0.00071
and
ET(hT+1) = 0.0003211

The 5% quantile is 0.00071 1.6449*(0.0003211)1/2 = -0.02877

The VaR for a portfolio size of $10,000,000 with probability 0.05

is

($10,000,000 )(-0.02877) = $287,700

i.e., with 95% chance, the potential loss of the portfolio is

$287,700 or less.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.

21t

Models with Explanatory Variables

To model the effects of 9/11 on stock returns,
create a dummy variable Dt equal to 0 before 9/11
and equal to 1 thereafter. Let

ht = 0 + 1 + 1ht1 + Dt

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
TARCH and EGARCH

Glosten, Jaganathan and Runkle (1994) showed how to

allow the effects of good and bad news to have different
effects on volatility. Consider the threshold-GARCH
(TARCH) process

ht 0 1 t21 1dt 1 t21 1ht 1

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Figure 3.11: The leverage effect

Expected
Volatility (Etht+1)

0
t
New Information

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The EGARCH Model
ln(ht ) 0 1 ( t 1 / ht0.5
1 ) 1 | t 1 / ht 1 | 1 ln( ht 1 )
0.5

1. The equation for the conditional variance is in log-linear form.

Regardless of the magnitude of ln(ht), the implied value of ht can never
be negative. Hence, it is permissible for the coefficients to be negative.
2. Instead of using the value of , the EGARCH model uses the level of
standardized value of t1 [i.e., t1 divided by (ht1)0.5]. Nelson argues
that this standardization allows for a more natural interpretation of the
size and persistence of shocks. After all, the standardized value of t1 is
a unit-free measure.
3. The EGARCH model allows for leverage effects. If t1/(ht1)0.5 is
positive, the effect of the shock on the log of the conditional variance is
1 + 1. If t1/(ht1)0.5 negative, the effect of the shock on the log of the
conditional variance is 1 + 1.
4. Although the EGARCH model has some advantages over the
TARCH model, it is difficult to forecast the conditional variance of an
EGARCH model.Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 ...tttsaasas
201122

Testing for Leverage Effects

1. If there are no leverage effects, the squared errors should be
uncorrelated with the level of the error terms

st2 a0 a1st 1 a2 st 2 ...

2. The Sign Bias test uses the regression equation of the form
st2 a0 a1dt 1 1t
where dt1 is to 1 if t-1 < 0 and is equal to zero if t-1 0.

3. The more general test is

st2 a0 a1dt 1 a2 dt 1st 1 a3 (1 d t 1 ) st 1 1t
dt1st1 and (1 dt1)st1 indicate whether the effects of positive and negative
shocks also depend on their size. You can use an F-statistic to test the null
hypothesis a1 = a2 = a3 = 0.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 10

ESTIMATING THE NYSE U.S.

100 INDEX

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Figure 3.12: Comparison of the Normal and t Distribuitions
( 3 degrees of freedom)
0.3

0.2

0.1

0
-5 -4 -3 -2 -1 0 1 2 3 4 5
Normal distribution t-distribution

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The estimated model
rt = 0.043 + t 0.058rt1 0.038rt2 AIC = 9295.36, SBC = 9331.91
(2.82) (3.00) (1.91)

ht = 0.014 + 0.084 ()2 + 0.906 ht1

(4.91) (9.59) (98.31)

Instead, if we use a t-distribution, we obtain

rt = 0.061 + t 0.062rt1 0.045rt2 AIC = 9162.72, SBC = 9205.37

(5.24) (3.77) (2.64)

ht = 0.009 + 0.089()2 + 0.909 ht1

(3.21) (8.58) (95.24)

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
22.5

20.0

17.5

15.0

12.5

10.0

7.5

5.0

2.5

0.0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Figure 3.15: The Estimated Variance

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Section 11

MULTIVARIATE GARCH

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
11. MULTIVARIATE GARCH

If you have a data set with several variables, it often makes

sense to estimate the conditional volatilities of the variables
simultaneously.

Multivariate GARCH models take advantage of the fact that

the contemporaneous shocks to variables can be correlated
with each other.

Equation-by-equation estimation is not efficient

Multivariate GARCH models allow for volatility spillovers

in that volatility shocks to one variable might affect the
volatility of other related variables

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Suppose there are just two variables, x1t and x2t. For now, we
are not interested in the means of the series

Consider the two error processes

1t = v1t(h11t)0.5
2t = v2t(h22t)0.5

Assume var(v1t) = var(v2t) = 1, so that h11t and h22t are the

conditional variances of 1t and 2t, respectively.

We want to allow for the possibility that the shocks are

correlated, denote h12t as the conditional covariance between
the two shocks. Specifically, let h12t = Et-11t2t.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The VECH Model
A natural way to construct a multivariate GARCH(1, 1) is the vech
model

h11t = c10 + 11 (1t-1)2 + 121t-12t-1 + 13 (2t-1)2 + 11h11t1
+ 12h12t1 + 13h22t-1

h12t = c20 + 21 (1t-1)2 + 221t-12t-1 + 23 (2t-1)2 + 21h11t1
+ 22h12t1 + 23h22t-1

h22t = c30 + 31 (1t-1)2 + 321t-12t-1 + 33 (2t-1)2 + 31h11t1

+ 32h12t1 + 33h22t-1

The conditional variances (h11t and h22t) and covariance depend on
their own past, the conditional covariance between the two variables
(h12t), the lagged squared errors, and the product of lagged errors (1t-
12t-1). Clearly, there is a rich interaction between the variables. After
one period, a v1t shock affects h11t, h12t, and h22t.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
ESTIMATION
Multivariate GARCH models can be very difficult to
estimate. The number of parameters necessary can get
quite large.
In the 2-variable case above, there are 21 parameters.

Once lagged values of {x1t} and {x2t} and/or explanatory

variables are added to the mean equation, the estimation
problem is complicated.

As in the univariate case, there is not an analytic solution

to the maximization problem. As such, it is necessary to
use numerical methods to find that parameter values that
maximize the function L.

Since conditional variances are necessarily positive, the

restrictions for the multivariate case are far more
complicated than for the univariate case.
The results of the maximization problem must be such
that every one of the conditional variances is always
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The diagonal vech

One set of restrictions that became popular in the early literature is the
so-called diagonal vech model. The idea is to diagonalize the system such
that hijt contains only lags of itself and the cross products of itjt. For
example, the diagonalized version of (3.42) (3.44) is

h11t = c10 + 11(1t-1)2 + 11h11t1

h12t = c20 + 221t-12t-1 + 22h12t-1
h22t = c30 + 33(2t-1)2 + 33h22t1

Given the large number of restrictions, model is relatively easy to

estimate.
Each conditional variance is equivalent to that of a univariate GARCH
process and the conditional covariance is quite parsimonous as well.
The problem is that setting all ij = ij = 0 (for i j) means that there are
no interactions among the variances. A 1t-1 shock, for example, affects
h11t and h12t, but does not affect the conditional variance h2t.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
THE BEKK
Engle and Kroner (1995) popularized what is now called the BEK (or
BEKK) model that ensures that the conditional variances are positive.
The idea is to force all of the parameters to enter the model via
quadratic forms ensuring that all the variances are positive. Although
there are several different variants of the model, consider the
specification
Ht = C'C + A't-1t-1'A + B'Ht-1B

where for the 2-variable case

c11 c12 11 12 11 12
C ; A ; B
c12 c22
21 22 21 22

If you perform the indicated matrix multiplications you will find

h11t ( c112 c122 ) (112 12t 1 211 211t 1 2 t 1 212 22t 1 )
( 112 h11t 1 2 11 21h12 t 1 212 h22 t 1 )

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
THE BEK II
In general, hijt will depend on the squared
residuals, cross-products of the residuals, and
the conditional variances and covariances of all
variables in the system.
The model allows for shocks to the variance of one of
the variables to spill-over to the others.
The problem is that the BEK formulation can be quite
difficult to estimate. The model has a large number of
parameters that are not globally identified. Changing
the signs of all elements of A, B or C will have effects
on the value of the likelihood function. As such,
convergence can be quite difficult to achieve.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 The BEKK (and Vech) as a VAR
h11t ( c112 c122 ) (112 12t 1 211 211t 1 2 t 1 212 22t 1 ) ( 112 h11t 1 2 11 21h12 t 1 212 h22 t 1 )

h12t c12 ( c11 c22 ) 121112t 1 (11 22 12 21 )1t 1 2 t 1 21 22 22t 1

11 12 h11t 1 ( 11 22 12 21 )h12 t 1 21 22 h22 t 1

h22 t ( c22
2
c122 ) (122 12t 1 212 221t 1 2 t 1 22
2 2
2 t 1 )
( 122 h11t 1 2 12 22 h12 t 1 222 h22 t 1 )

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Constant Conditional Correlations (CCC)
As the name suggests, the (CCC)model restricts the
correlation coefficients to be constant. As such, for each i
j, the CCC model assumes hijt = ij(hiithjjt)0.5.

In a sense, the CCC model is a compromise in that the

variance terms need not be diagonalized, but the
covariance terms are always proportional to (hiithjjt)0.5. For
example, a CCC model could consist of (3.42), (3.44) and

h12t = 12(h11th22t)0.5

Hence, the covariance equation entails only one

parameter instead of the 7 parameters appearing in
(3.43).

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
EXAMPLE OF THE CCC MODEL
Bollerslev (1990) examines the weekly values of the
nominal exchange rates for five different countries--
the German mark (DM), the French franc (FF), the
Italian lira(IL), the Swiss franc (SF), and the British
pound (BP)--relative to the U.S. dollar.
A five-equation system would be too unwieldy to estimate
in an unrestricted form.
For the model of the mean, the log of each exchange rate
series was modeled as a random walk plus a drift
yit = i + it (3.45)
where yit is the percentage change in the nominal
exchange rate for country i,
Ljung-Box tests indicated each series of residuals did
not contain any serial correlation

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Next, he tested the squared residuals for serial dependence. For example, for the
British pound, the Q(20)-statistic has a value of 113.020; this is significant at
any conventional level.
Each series was estimated as a GARCH(1, 1) process.The specification has the
form of (3.45) plus

hiit = ci0 + ii (it-1)2 + iihiit1 (i = 1, , 5)

hijt = ij(hiithjjt)0.5 (i j)

The model requires that only 30 parameters be estimated (five

values of i, the five equations for hiit each have three
parameters, and ten values of the ij).
As in a seemingly unrelated regression framework, the system-
wide estimation provided by the CCC model captures the
contemporaneous correlation between the various error terms.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
RESULTS
The estimated correlations for the period during which the European
Monetary System (EMS) prevailed are
DM FF IL SF
FF 0.932
IL 0.886 0.876
SW 0.917 0.866 0.816
BP 0.674 0.678 0.622 0.635

It is interesting that correlations among continental

European currencies were all far greater than those for the
pound. Moreover, the correlations were much greater than
those of the preEMS period. Clearly, EMS acted to keep the
exchange rates of Germany, France, Italy and Switzerland
tightly in line prior to the introduction of the Euro.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 For the second step, you should check the squared residuals for the
presence of GARCH errors. Since we are using daily data (with a five-
day week), it seems reasonable to begin using a model of the form
The sample values of the F-statistics for the null hypothesis that 1 =
= 5 = 0 are 43.36, 89.74, and 20.96 for the Euro, BP and SW,
respectively. Since all of these values are highly significant, it is possible
to conclude that all three series exhibit GARCH errors.

5
0 it25
t
2

i 1

The sample values of the F-statistics for the null hypothesis that 1 =
= 5 = 0 are 43.36, 89.74, and 20.96 for the Euro, BP and SW,
respectively. Since all of these values are highly significant, it is
possible to conclude that all three series exhibit GARCH errors.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
1.0

0.8

0.6

0.4

0.2

0.0

-0.2
2000 2002 2004 2006 2008 2010 2012
Figure 3.16: Pound/Franc Correlation from the Diagonal vech

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Panel a: Volatility Response of the Euro
0.4

0.3

0.2

0.1

0.0
November December January February March April May June July
2009

Panel b: Response of the Covariance

0.4

0.3

0.2

0.1

0.0
November December January February March April May June July
2009

Panel c: Volatility Response of the Pound

0.4

0.3

0.2

0.1

0.0
November December January February March April May June July
2009

Figure 3.17 Variance Impulse Responses from Oct. 29, 2008

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Appendix: The Log Likelihood Function
1 1 12t 22t 2 121t 2 t
Lt exp 2
0.5
2 h11h22 (1 12 )
2
2(1 )
12 h11 h22 ( h h
11 22 )

where 12 is the correlation coefficient between 1t and 2t;

12 = h12/(h11h22)0.5.

Now define
h11 h12
H
h12 h22
1 1
Lt 1/ 2
exp tH 1 t
2 H 2

where t = ( 1t, 2t )', and | H | is the determinant of H.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
 Now, suppose that the realizations of {t} are independent, so
that the likelihood of the joint realizations of 1, 2, T is the
product in the individual likelihoods. Hence, if all have the same
variance, the likelihood of the joint realizations is
T
1 1
L 1/ 2
exp tH 1 t
t 1 2 H 2

T T 1 T
ln L = ln (2 ) ln | H | tH 1 t
2 2 2 t 1

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
MULTIVARIATE GARCH MODELS
For the 2-variable
T
1 1 1
L 1/ 2
exp
2 t t t
H
t 1 2 H t

h11t h12t
Ht
h12t h22 t

T 1 T
ln L ln(2 ) (ln | H t | tH t1 t )
2 2 t 1

The form of the likelihood function is identical for models with k

variables. In such circumstances, H is a symmetric k x k matrix, t is a
k x 1 column vector, and the constant term (2) is raised to the power
k.
Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
The vech Operator
The vech operator transforms the upper (lower) triangle of
a symmetric matrix into a column vector. Consider the
symmetric covariance matrix
h11t h12t
Ht
h22 t
vech(Ht) = [ h11t, h12t, h22t ]
h12t
Now consider t = [1t, 2t]. The product tt = [1t, 2t][1t, 2t] is

12t 1t 2 t

1t 2t 12t

2

vech( t t ) = 1t , 1t 2t , 2 t
2

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
If we now let C = [ c1, c2, c3 ], A = the 3 x 3
matrix with elements ij, and B = the 3 x 3
matrix with elements ij, we can write

vech(Ht) = C + A vech(t-1t-1) + Bvech(Ht-1)

it should be clear that this is precisely the

system represented by (3.42) (3.44). The
diagonal vech uses only the diagonal
elements of A and B and sets all values of ij
= ij = 0 for i j.

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Constant Conditional Correlations
h11t 12 (h11t h22t )0.5
Ht
( h h
12 11t 22t ) 0.5
h22t

Now, if h11t and h22t are both GARCH(1, 1) processes, there are
seven parameters to estimate (the six values of ci, ii and ii
and 12).

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.
Dynamic Conditional Correlations
STEP 1: Use Bollerslevs CCC model to obtain the GARCH
estimates of the variances and the standardized residuals
STEP 2: Use the standardized residuals to estimate the
conditional covariances.
Create the correlations by smoothing the series of standardized
residuals obtained from the first step.
Engle examines several smoothing methods. The simplest is the
exponential smoother qijt = (1 )sitsjt + qijt-1 for < 1.
Hence, each {qiit} series is an exponentially weighted moving
average of the cross-products of the standardized residuals.
The dynamic conditional correlations are created from the qijt as
ijt = qijj/(qijtqjjt)0.5

Copyright 2015 John, Wiley & Sons, Inc. All rights reserved.