Features of Financial Time-Series Data

Univariate ARCH/GARCH models

Multivariate GARCH models
References

Chapter 5: Modelling Volatility and

Correlation of Financial Time Series

Advanced Financial Empirical Research

1/50
 Features of Financial Time-Series Data
Univariate ARCH/GARCH models
Multivariate GARCH models
References

Features of Time-Series Financial Data I

1. Leptokurtosis - That is, the tendency for financial asset

returns to have distributions that exhibit fat tails and excess
peakedness at the mean.
I A distribution with positive excess kurtosis is called leptokurtic.
In a letokurtic distribution, risk is coming from outlier events
and extreme observations are much more likely to occur.

2/50
 Features of Financial Time-Series Data
Univariate ARCH/GARCH models
Multivariate GARCH models
References

Features of Time-Series Financial Data II

2. Volatility clustering or volatility autocorrelation - The

variance of the returns exhibits time-variation
(heteroskedasticity). There is a tendency for volatility in
financial markets to appear in bunches. Thus, large returns
(of either sign) are expected to follow large returns, and small
returns (of either sign) to follow small returns.
I A plausible explanation for this phenomenon, which seems to
be an almost universal feature of asset return series in finance,
is that the information arrivals which drive price changes occur
in bunches rather than being evenly spaced over time.

3/50
 Features of Financial Time-Series Data
Univariate ARCH/GARCH models
Multivariate GARCH models
References

Features of Time-Series Financial Data III

3. Leverage effects - Volatility tends to rise more following a

large price fall than following a price rise of the same
magnitude.
I The intuition is that a fall in the value of a firm’s stock causes
the firm’s debt to equity ratio to rise.

4/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

ARCH
To model ”volatility clustering”, Engle (1982) proposed the
AutoRegressive Conditional Heteroskedasticity (ARCH) model.
K
X
yt = bk xkt−1 + ut (1)
k=1

ut |Φt−1 ∼ N(0, σt2 ) (2)

σt2 = ω + αut−1
2
(3)

I Φt−1 is the information available at time t − 1.

I σt2 is the conditional variance of the residuals (conditioning on
Φt−1 ).
I σt2 ≡ Var[ut |φt−1 ] = E (ut |Φt−1 − E[ut |Φt−1 ])2 = E[ut2 |Φt−1 ]
5/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

ARCH

The model is known as an ARCH(1), since the conditional variance

depends on one lagged squared error.
Eq(1) is named as the mean equation, as it specifies the process
for the conditional mean of yt . Eq (3) is named as the conditional
variance equation.

6/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

ARCH

ARCH(1) can be extended to the general case, an ARCH(q)

model, where the error variance depends on q lags of squared errors

σt2 = ω + α1 ut−1
2 2
+ α2 ut−2 2
+ . . . + αq ut−q (4)

7/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Non-Negativity Constraints

Since σt2 is a conditional variance, its value must always be strictly

positive. To ensure that the model produces positive values for the
variance at all times, all of the coefficients ω and α:s are
usually required to be non-negative.

8/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Unconditional variance
If
(α1 + α2 + . . . + αq ) < 1 (5)
the unconditional variance for an ARCH(q) is:
σ 2 = E[σt2 ] (6)
2 2 2
= E[ω + α1 ut−1 + α2 ut−2 + . . . + αq ut−q ]
2 2 2
= ω + α1 E[ut−1 ] + α2 E[ut−2 ] + . . . + αq E[ut−q ] (7)
2 ] = E[u 2 ] = . . . = E[u 2 ] = σ 2 ,
Since E[ut−1 t−2 t−q

σ 2 = ω + α1 σ 2 + α2 σ 2 + . . . + αq σ 2 (8)
= ω + (α1 + α2 + . . . + αq )σ 2
ω
σ2 =
1 − α1 − α2 − . . . − αq
9/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Testing for ARCH effect I

The test is a joint null hypothesis that all q lags of the squared
residuals have coefficient values that are not significantly different
from zero.
1. Run the linear regression in the mean specification, e.g.
K
X
yt = bk xkt−1 + ut , (9)
k=1

saving the residuals, ût .

10/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Testing for ARCH effect II

2. Square the residuals, and regress them on q own lags to test

for ARCH of order q , i.e. run the regression

ût2 = γ0 + γ1 ût−1
2 2
+ γ2 ût−2 2
+ · · · + γq ût−q + vt . (10)

Obtain R 2 from this regression.

Hint: Recall that σt2 = E[ut2 |Φt−1 ] in ARCH(1).

11/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Testing for ARCH effect III

3. The test statistic is

TR 2 ∼ χ2 (q) (11)

where T is the number of observations.

I H0 : γi = 0 for i > 0.

12/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Limitations of ARCH models

ARCH models are rarely used nowadays due to its limitations:

I It can be problematic to decide the value of q, the number of
lags of the squared residual in the model.
I The value of q that is required to capture all of the
dependence in the conditional variance, might be very large.
This would result in a model that is not parsimonious.

13/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

GARCH models I

Generalized ARCH or GARCH model is developed independently by

Bollerslev (1986) and Taylor (1986). Usually, it is found that a
GARCH(1,1) model is sufficient to capture the volatility dynamics:

σt2 = ω + αut−1
2 2
+ βσt−1 (12)

where one lagged squared error and one lagged variance is needed.

14/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

GARCH models II

Why is GARCH better and far more widely used than ARCH?
I GARCH is more parsimonious, and avoids overfitting.

15/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

GARCH models III

GARCH(1,1) model can be written as a restricted infinite
order ARCH model:

σt2 = ω + αut−1
2 2
+ βσt−1 (13)
2 2 2
= ω+ αut−1+ β(ω + αut−2 + βσt−2 )
2 2 2 2
= ω + βω + αut−1 + αβut−2 + β σt−2
2 2
= ω + βω + αut−1 + αβut−2 + β 2 (ω + αut−3
2
+ βσt−32
)
= ω + βω + β 2 ω + αut−1 2 2
+ αβut−2 + αβ 2 ut−3
2
+ β 3 σt−3
2

= ω(1 + β + β 2 ) + α(1 + βL + β 2 L2 )ut−12

+ β 3 σt−3
2

= ω(1 + β + β 2 + · · · ) + α(1 + βL + β 2 L2 + · · · )ut−1

2
+ β ∞ σ02

where L is the lag operator.

16/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

GARCH models IV

The GARCH(1,1) is a parsimonious model. It contains only three

parameters in the conditional variance equation, but allows an
infinite number of past squared errors to influence the current
conditional variance.

17/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Non-Negativity Constraints and Unconditional Variance I

Non-Negativity Constraints
To ensure that the GARCH produces positive values for the
variance at all times, all of the coefficients in the conditional
variance equation are usually required to be non-negative.

18/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Non-Negativity Constraints and Unconditional Variance II

Unconditional Variance
If α + β < 1, the unconditional variance under GARCH(1,1) is

σ 2 = E[σt2 ] = E[ω + αut−1

2 2
+ βσt−1 ] (14)
2 2
=ω+ αE[ut−1 ]+ βE[σt−1 ]
2 2
= ω + ασ + βσ
ω
σ2 = (15)
1−α−β

19/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Use of GARCH Models in Volatility Forecasting

In practice, the conditional mean of financial time series is

usually found to be an AR process. So, it is possible to show
that

Var(yt |yt−1 , yt−2 , . . .) = Var(ut |ut−1 , ut−2 , . . .) (16)

Hence, modelling σt2 will give models and forecasts for the
variance of yt as well. Forecasts of σt2 will be forecasts of the
future variance of yt .

20/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Use of GARCH Models in Volatility Forecasting

Making an s-step ahead forecast

The s-step forecast volatility can be used to price an option with s
steps to expiration using our volatility model, to find optimal hedge
ratio s steps ahead, etc.

21/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Use of GARCH Models in Volatility Forecasting

From the GARCH(1,1) equation

2
σt+1 = ω + αut2 + βσt2 (17)

At time t, the values of all the terms on the RHS of (17) are
2
known. Let σ̂t+1 be the one-step-ahead forecast at time t. Taking
conditional expectation of (17)
2 2
σ̂t+1 = E[σt+1 |Φt ] = ω + αut2 + βσt2 (18)

where Φt denotes all information available up to and including

observation t.

22/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Use of GARCH Models in Volatility Forecasting

How is the two-step-ahead forecast calculated? Given GARCH(1,1)
2 2 2
σt+2 = ω + αut+1 + βσt+1 (19)
Let 2
σ̂t+2be the two-step-ahead forecast at time T . The terms on
2
the RHS of eq (19) are unkonwn at t. We calculate σ̂t+2 by taking
conditional expectation of eq(19)
2 2 2 2
σ̂t+2 = E[σt+2 |Φt ] = ω + αE[ut+1 |Φt ] + βE[σt+1 |Φt ] (20)
2 2
=ω+ ασt+1 + βE[σt+1 |Φt ]
2 2
=ω+ ασt+1 + β σ̂t+1
2
σt+1 is unknown at t, so it is replaced with the forecast for it,
2 , so that (20) becomes
σ̂t+1
2 2 2
σ̂t+2 = ω + ασ̂t+1 + β σ̂t+1 (21)
2
= ω + (α + β)σ̂t+1
23/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Use of GARCH Models in Volatility Forecasting

2
Similarly, let σ̂t+3 be the three-step-ahead forecast at time t.
2 2 2 2
σ̂t+3 = E[σt+3 |Φt ] = ω + αE[ut+2 |Φt ] + βE[σt+2 |Φt ] (22)
2
= ω + (α + β)σ̂t+2
 
2
= ω + (α + β) ω + (α + β)σ̂t+1 (23)
= ω + (α + β)ω + (α + β)2 σ̂t+1
2
(24)

Any s-step-ahead forecast for s ≥ 2 will be

s−1
X
2
σ̂t+s =ω (α + β)i−1 + (α + β)s−1 σ̂t+1
2
(25)
i=1

24/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Asymmetric GARCH Models

GARCH models enforce a symmetric response of volatility to

positive and negative shocks. This arises since the conditional
variance is a function of the magnitudes of the lagged residuals
and not their signs (in other words, by squaring the lagged error,
the sign is lost).
In order to model leverage effects in financial time-series, we need
asymmetric formulations of GARCH.

25/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Asymmetric GARCH Models

I GJR-GARCH by Glosten et al. (1993)
The conditional variance is now given by

σt2 = ω + αut−1
2 2
+ γut−1 2
It−1 + βσt−1 (26)
2 2
= ω + (α + γIt−1 )ut−1 + βσt−1
(
1 if ut−1 > 0
where It−1 =
0 if ut−1 ≤ 0

According to (26), the response of σt2 to ut−1

2
(
α+γ if ut−1 > 0
=
α if ut−1 ≤ 0
Therefore, For leverage effect, we would see γ < 0.

26/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Asymmetric GARCH Models

The condition for non-negativity in GJR-GARCH is ω > 0, α > 0,

β ≥ 0, and α + γ ≥ 0. That is, the model is still admissible, even
if γ < 0, provided that α + γ ≥ 0.

27/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Asymmetric GARCH Models

I The Exponential GARCH (EGARCH) by Nelson (1991)
One specification of exponential GARCH is
 s 
|ut−1 | 2 ut−1
ln σt2 = ω + α  q − + γq 2
+ β ln σt−1
2
σt−1 π σ 2
t−1
(27)
(
α+γ if ut−1 > 0
I Asymmetric response =
α−γ if ut−1 ≤ 0
I If the relationship between volatility and returns is negative, γ
will be negative.
I Since ln σt2 is modelled, then even if the parameters in eq(27)
are negative, σt2 will be positive. There is thus no need to
artificially impose non-negativity constraints on the model
parameters.
28/50
 Features of Financial Time-Series Data
ARCH
Univariate ARCH/GARCH models
GARCH models
Multivariate GARCH models
Maximum Likelihood Estimation
References

Maximum Likelihood Estimation

The log likelihood function for (G)ARCH models is

T T
T 1X 1X ut2
LLF = − ln(2π) − ln σt2 − (28)
2 2 t=1 2 t=1 σt2

where σt2 follows the process specified by the (G)ARCH model.

29/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Multivariate GARCH models: modelling

covariances/correlations

Covariances/correlations between asset returns also vary over time.

Multivariate GARCH models are in spirit very similar to their
univariate counterparts, except that the former also specifies
equations for how the covariances move over time.

30/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Multivariate GARCH models: modelling

covariances/correlations
Consider yt an N × 1 vector of asset returns at t

yt = xt−1 b + ut , (29)
ut ∼ NID(0, Ht ), (30)

I xt−1 is an N × K matrix of explanatory variables.

I b is an K × 1 vector of coefficients for the explanatory
variables.
I Ht ≡ Vart−1 (ut ) is the N × N conditional variance-covariance
matrix.
I Ht have different formulations in different multivariate
GARCH models.
31/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Positive Semi-Definiteness of Conditional Covariance

Matrix

Since Ht is a variance-covariance matrix, it must always be

positive semi-definite.
I Positive semi-definiteness of Ht ensures that
I the matrix is symmetric,
I all the variances (diagonal elements) are non-negative,
I the covariance between two series is the same irrespective of
which of the two series is taken first (i,e, σij = σji ).
I Ht being symmetric means that only N(N + 1)/2 elements in
Ht are unique. Specifically, N variances on the diagonal and
N(N − 1)/2 off-diagonal covariances.

32/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

A formulation of Ht , named VECH, was proposed by Bollerslev
et al. (1988). The VECH model parametrizes directly the process
of the N(N + 1)/2 variances and covariances in Ht .

ht = C + Aηt−1 + Ght−1 (31)

where
ht = vech(Ht ), (32)
ηt = vech(ut ut0 ). (33)
vech(·) denotes the column-stacking operator applied to the upper
portion of the symmetric matrix. C is a (N + 1)N/2 × 1 column
vector of conditional variance and covariance intercepts, and A and
G are square parameter matrices of order (N + 1)N/2.
33/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

Consider a bivariate case (N = 2). Define

   
" # c1 a11 a12 a13
σ11,t σ12,t
Ht = , C = c2  A = a21 a22 a23  ,
   
σ21,t σ22,t
c3 a31 a32 a33
 
b11 b12 b13
G = b21 b22 b23  (34)
 
b31 b32 b33

34/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

The VECH model in the bivariate case is given by

      
2
u1,t−1
σ11,t c1 a11 a12 a13
2
σ22,t  = c2  + a21 a22 a23   u2,t−1 (35)
      

σ12,t c3 a31 a32 a33 u1,t−1 u2,t−1
  
b11 b12 b13 σ11,t−1
+ b21 b22 b23  σ22,t−1 
  
b31 b32 b33 σ12,t−1

35/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

The processes for the conditional variances and the conditional

covariance can be written as
2 2
σ11,t = c1 + a11 u1,t−1 + a12 u2,t−1 + a13 u1,t−1 u2,t−1 + b11 σ11,t−1
+ b12 σ22,t−1 + b13 σ12,t−1 (36)
2 2
σ22,t = c2 + a21 u1,t−1 + a22 u2,t−1 + a23 u1,t−1 u2,t−1 + b21 σ11,t−1
+ b22 σ22,t−1 + b23 σ12,t−1 (37)
2 2
σ12,t = c3 + a31 u1,t−1 + a32 u2,t−1 + a33 u1,t−1 u2,t−1 + b31 σ11,t−1
+ b32 σ22,t−1 + b33 σ12,t−1 (38)

36/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

Advantage and Disadvantage of VECH
I Advantages
I The model (35) is simple to interpret.
I The conditional variance of one series is dependent on the
lagged conditional variance and the lagged squared shock of
the other series. This allows for modelling volatility spillover
from one series to another, which is a widespread phenomenon
for financial assets.
I Disadvantage
I This model is highly parametrized. For N = 2 there are 21
parameters, while for N = 3 there are 78, and N = 4 implies
210 parameters!
I The VECH processes do not guarantee that Ht is a positive
semi-definite matrix.

37/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The VECH model

Diagonal VECH
Bollerslev et al. (1988) suggest the diagonal VECH (DVECH)
model in which A and G are assumed to be diagonal: every
element σijt depends only on its own lag and on the previous value
of ui,t−1 uj,t−1 .

σij,t = ωij + αij ui,t−1 uj,t−1 + βij σij,t−1 for i, j = 1, 2, . . . N (39)

The DVECH is much more parsimonious than the VECH, but it

cannot capture spillover of volatility from one variable to another
because every element σijt follows a univariate process.

38/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The BEKK model I

Engle and Kroner (1995) propose a new parametrization for Ht ,

namely BEKK, which ensures that Ht is always positive-definite.

Ht = C 0 C + A0 ut−1 ut−1
0
A + G 0 Ht−1 G (40)

where C, A and G are N × N matrices and C is an upper

triangular matrix. The positive definiteness of Ht is ensured owing
to the quadratic nature of the terms on the equations’ RHS.

39/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The BEKK model II

I C, A, and G are not unique, e.g.

A0 ut−1 ut−1
0 A = (−A)0 ut−1 ut−1
0 (−A).
I To get unique estimates of the parameters, Engle and Kroner
(1995) restrict a11 , g11 and the diagonal elements of C to be
positive.

40/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The BEKK model III

Diagonal BEKK
The number of parameters in the BEKK(1,1) model is
N(5N + 1)/2. To reduce the number of parameters, one can
impose a diagonal BEKK model, i.e. A and G are diagonal
matrices.

41/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The BEKK model IV

In the bivariate case, the diagonal BEKK model is:
! ! !
σ11,t σ12,t c11 0 c11 c12
Ht = =
σ12,t σ22,t c12 c22 0 c22
! ! !
a 0 u1,t−1   a 0
+ 11 u1,t−1 u2,t−1 11
0 a22 u2,t−1 0 a22
! ! !
g 0 σ11,t−1 σ12,t−1 g11 0
+ 11
0 g22 σ12,t−1 σ22,t−1 0 g22
! !
2 2 u2
a11 a11 a22 u1,t−1 u2,t−1
c11 c11 c12 1,t−1
= 2 2 + 2 u2
c11 c12 c12 + c22 a11 a22 u1,t−1 u2,t−1 a22 2,t−1
!
2 σ
g11 g11 g22 σ12,t−1
11,t−1
+ 2 σ (41)
g11 g22 σ12,t−1 g22 22,t−1

42/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The BEKK model V

Th conditional variances have respective GARCH formulations.

The parameters governing the dynamics of the covariance equation
are the products of the corresponding parameters of the two
variances equations.

43/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Maximum Likelihood Estimation

The log likelihood function for multivariate GARCH models is

T T
NT 1X 1X
LLF = − ln(2π)− ln |Ht |− (yt −xt−1 b)0 Ht−1 (yt −xt−1 b)
2 2 t=1 2 t=1
(42)

44/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

The Estimation of Time-Varying Betas

Multivariate GARCH models provide a simple method for

estimating time-varying betas.
σim,t
βi,t = 2 (43)
σm,t

where βi,t is the time-varying beta estimate at time t for stock i,

σim,t is the conditional covariance between market returns and
2
returns to stock i at time t and σm,t is the conditional variance of
the market return.

45/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Dynamic Hedge Ratios

A hedge can be achieved by taking opposite positions in spot and

futures markets simultaneously, so that any loss sustained from an
adverse price movement in one market should to some degree be
offset by a favourable price movement in the other.
I The ratio of the number of units of the futures asset that are
purchased relative to the number of units of the spot asset is
the hedge ratio.
I A usual strategy is to choose the optimal hedge ratio that
minimises the variance of the returns of a portfolio containing
the spot and futures position; this is known as the
minimum-variance hedge ratio.

46/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Dynamic Hedge Ratios

Let ∆St and ∆Ft represent the logarithmic returns of the spot and
futures prices, respectively, from time t − 1 to t.
I For a short hedge (i.e. long in the asset and short in the
futures contract), the expected return at time t − 1 of the
hedger’s portfolio is

Et−1 (Rt ) = Et−1 (∆St ) − ht−1 Et−1 (∆Ft ) (44)

where ht−1 is the hedge ratio determined at time t − 1 for

hedging position in period t.
I For a long hedge, the expected return at time t − 1 of the
hedger’s portfolio is

Et−1 (Rt ) = ht−1 Et−1 (∆Ft ) − Et−1 (∆St ) (45)

47/50
 Positive Semi-Definiteness of Conditional Covariance Matrix
Features of Financial Time-Series Data
The VECH model
Univariate ARCH/GARCH models
The BEKK model
Multivariate GARCH models
Maximum Likelihood Estimation
References
Use of Multivariate GARCH in Finance: Examples

Dynamic Hedge Ratios

2 , of the
For both cases, the variance of the expected return, σP,t
portfolio is
2 2 2
σP,t = σS,t + ht−1 σF2 ,t − 2ht−1 σSF ,t (46)

Minimizing (46) with respect to ht−1 , we get the optimal hedge

∗ ,
ratio, ht−1

∗ σSF ,t
ht−1 = (47)
σF2 ,t

The time-varying σSF ,t and σF2 ,t can be obtained from the

conditional covariance matrix estimated using multivariate GARCH
models.

48/50
 Features of Financial Time-Series Data
Univariate ARCH/GARCH models
Multivariate GARCH models
References

References I

Bollerslev, T. (1986). Generalized autoregressive conditional

heteroskedasticity. Journal of econometrics, 31(3):307–327.
Bollerslev, T., Engle, R., and Wooldridge, J. (1988). A capital
asset pricing model with time-varying covariances. The Journal
of Political Economy, pages 116–131.
Engle, R. and Kroner, K. (1995). Multivariate simultaneous
generalized arch. Econometric theory, 11(01):122–150.
Engle, R. F. (1982). Autoregressive conditional heteroscedasticity
with estimates of the variance of united kingdom inflation.
Econometrica: Journal of the Econometric Society, pages
987–1007.

49/50
 Features of Financial Time-Series Data
Univariate ARCH/GARCH models
Multivariate GARCH models
References

References II

Glosten, L. R., Jagannathan, R., and Runkle, D. E. (1993). On the

relation between the expected value and the volatility of the
nominal excess return on stocks. The journal of finance,
48(5):1779–1801.
Nelson, D. B. (1991). Conditional heteroskedasticity in asset
returns: A new approach. Econometrica: Journal of the
Econometric Society, pages 347–370.
Taylor, S. J. (1986). Modelling Financial Time Series. World
Scientific Publishing.

50/50