Documente Academic
Documente Profesional
Documente Cultură
to
understand
time-series
based
researches,
also
expect
that
some
of
you
will
be
is
series
data3
compiled,
is
long
believe
enough
for
that
you
the
Vietnam
time
to
conduct
such
Selected papers were compiled by Phung Thanh Binh & Vo Duc Hoang Vu (2009). You
The most important data sources for these studies can be World Banks World
highly
appreciated
by
international
journals
the
current
time,
EFA
becomes
the
most
stupid
of
economics,
HCMC.
They
blindly
imitate
topics
such
family
as
serial
models,
correlation,
impulse
ARIMA
response,
models,
variance
can
find
them
elsewhere
such
as
econometrics
of
the
ECM
using
the
ECMs
using
vector
error
have
previously
learned
in
the
basic
econometrics
Saying
information
differently,
as
possible
we
from
will
the
extract
past
as
history
much
of
the
two
forecasting
fundamental
and
dynamic
types,
namely,
modelling.
Pure
time
series
time
series
econometrics,
concern
much
in
with
univariate
analysis
we
building
structural
do
not
models,
able
to
forecast
well.
The
efficient
forecasting
as
significance
of
the
estimated
coefficients
correlogram,
Akaike
and
Schwarz
criteria,
and
(say,
ect).
including
analysis,
asset
On
the
bivariate
is
mostly
prices,
other
and
exchange
hand,
dynamic
multivariate
concerned
rates,
with
interest
modelling,
time
series
understanding
the
dynamic
modelling
has
become
increasingly
popular
series
with
common
trends,
or
cointegration)
and
volatility
or
ARCH)5.
Up
to
now,
dynamic
analysts
stochastic
should
process.
clearly
According
understand
to
Gujarati
the
term
(2003),
for
instance
the
GDP
of
$2873
billion
for
and
political
climate
then
prevailing.
http://nobelprize.org/nobel_prizes/economics/laureates/2003/
The
value
of
$2873
billion
as
the
mean
value
of
all
process
and
its
realization
in
time
series
to
draw
inferences
about
population;
in
time
reason
why
mention
this
term
before
examining
models
(population).
relate
Stock
&
to
the
Watson
stochastic
(2007)
say
process
that
the
from
the
past,
then
the
historical
process
that
has
received
great
deal
of
attention and scrutiny by time series analysts is the socalled stationary stochastic process. Broadly speaking,
a time series is said to be stationary if its mean and
variance are constant over time and the value of the
covariance6 between the two periods depends only on the
distance or gap or lag between the two time periods and
not the actual time at which the covariance is computed
(Gujarati, 2011). In the time series literature, such a
stochastic process is known as a weakly stationary or
covariance
stationary.
By
contrast,
time
series
is
are
invariant
process
is
over
normal,
time.
the
If,
however,
weakly
the
stationary
(a)
exhibits
mean
reversion
in
that
it
fluctuates
(c)
t); and
(c) Covariance: Cov(Yt,Yt+k) =
where
k,
= E[(Yt- )(Yt+k- )]
as
those
of
Yt.
In
short,
if
time
series
is
to
its
mean
(called
mean
reversion)
and
fluctuations
words,
nonstationary
time
series
will
have
behavior
only
consideration.
for
Each
the
set
of
time
time
period
series
under
data
will
possible
Therefore,
to
for
analysis,
such
generalize
the
it
purpose
to
of
(nonstationary)
other
time
forecasting
time
series
periods.
or
may
policy
be
of
is
According
process
variance
also
popular
to
Gujarati
purely
random
in
time
(2003),
if
it
we
has
series
call
zero
econometrics.
a
stochastic
mean,
constant
normal
linear
regression
model,
once
discussed
in
the
to
Stock
and
Watson
(2007),
time
series
time
series
data:
(1)
the
series
can
have
data:
deterministic
and
stochastic.
on)7.
For
example,
the
LEX
[the
logarithm
of
the
10
1000
1500
2000
.3
.2
.1
.0
-.1
-.2
500
1000
1500
2000
11
to
model
economic
time
series
as
having
stochastic
trends
unless
we
explicitly
say
otherwise.
The simplest model of a variable with a stochastic trend
is the random walk. There are two types of random walks:
(1)
random
walk
without
drift
(i.e.
no
constant
or
random
walk
without
drift
is
defined
as
follow.
(1)
12
Yt = Yt-1 + ut = Y0 + u1 + + ut
In general, if the process started at some time 0 with a
value Y0, we have
Yt
Y0
(2)
ut
therefore,
E(Yt)
E(Y0
ut)
Y0
E(Y0
ut
Y0)2
E(
ut)2
variance
increases
indefinitely,
thus
violating
(3)
time
series
data.
This
is
widely
known
13
as
the
8
4
0
-4
-8
-12
-16
-20
50
100
150
200
250
300
350
400
450
500
.03
.02
.01
.00
-.01
-.02
-.03
500
1000
1500
2000
14
+ Yt-1 + ut
(4)
comes
from
the
fact
that
if
we
write
the
preceding
equation as:
Yt Yt-1 = Yt =
+ ut
(5)
walk
with
drift
violates
both
conditions
of
stationarity:
E(Yt)
= Y0 + t.
Var(Yt) = t
distribution
depends
on
t,
that
is,
it
is
nonstationary.
Stock and Watson (2007) say that because the variance of
a
random
walk
autocorrelations
increases
are
without
not
bound,
defined
its
population
(the
first
Corr(Yt,Yt-1) =
Cov(Yt, Yt 1)
~
Var(Yt)Var(Yt 1)
15
30
25
20
15
10
5
0
-5
-10
50
100
150
200
250
300
350
400
450
500
10
5
0
-5
-10
-15
-20
-25
50
100
150
200
250
300
350
400
450
500
16
model
autoregressive
Yt-1 + ut (-1
resembles
model
the
[AR(1)],
1)
Markov
mentioned
(6)
first-order
in
the
basic
= 1,
is in fact
= 1.
Technically, if
= 1, we can
the
name
nonstationarity,
unit
random
root.
walk,
and
Thus,
unit
the
root
terms
can
be
treated as synonymous.
If, however, ||
17
Case 1:
= 0.67 is
= 1.26 is
root
and
is
non-stationary.
Graph
of
= 1 are presented in
Figure 2.9.
In order to reproduce the graphs and the series which are
stationary,
exploding
and
nonstationary,
we
type
the
1:
Open
new
workfile
(say,
undated
18
type),
smpl 1 200
Step 3: Plot X, Y, Z using the line plot type (Figures
2.7, 2.8, and 2.9).
plot X
plot Y
plot Z
5
4
3
2
1
0
-1
-2
-3
-4
25
50
75
100
19
125
150
175
200
1.6E+19
1.4E+19
1.2E+19
1.0E+19
8.0E+18
6.0E+18
4.0E+18
2.0E+18
0.0E+00
25
50
75
100
125
150
175
150
175
200
-5
-10
-15
-20
-25
25
50
75
100
20
125
200
most
cases
nonstationary
or
are
nonstationary.
trended
data
The
is
that
problem
the
with
standard
lead
to
incorrect
conclusions.
According
to
(2007)
states
that
many
economic
series
are
not
stationary
as
the
mean
is
continually
21
(7)
the
presence
obtained
from
spurious9
and
of
nonstationarity,
regression
these
of
regressions
this
then
the
results
kind
are
totally
are
called
spurious
regressions.
The intuition behind this is quite simple. Over time, we
expect
any
nonstationary
Figure
3.1),
so
over
series
any
to
wander
reasonably
long
around
sample
(see
the
then
performed
regression
of
one
series
on
the
This was first introduced by Yule (1926), and re-examined by Granger and Newbold
22
and
Newbold
(1974)
constructed
Monte
Carlo
(8)
Xt = Xt-1 + eXt
(9)
Yt
regression
and
Xt
are
between
independent
them
should
of
each
give
other,
any
insignificant
23
commands
in
Eviews
(after
opening
the
new
24
10
-10
-20
-30
-40
-50
-10
-5
10
15
20
25
or
if
R2
then
the
estimated
regression
must be spurious.
To understand the problem of spurious regression better,
it might be useful to use an example with real economic
data.
This
example
was
conducted
by
Asteriou
(2007).
results
obtained
from
such
following:
25
regression
are
the
Yt =
(4.743) (8.572)
Here we see very good t-ratios, with coefficients that
have
the
right
signs
and
more
or
less
plausible
0.945),
but
there
is
high
degree
of
is
totally
meaningless
because
the
money
supply data are for the UK economy and the GDP figures
are for the US economy. Therefore, although there should
not be any significant relationship, the regression seems
to fit the data very well, and this happens because the
variables used in the example are, simply, trended (i.e.
nonstationary).
So,
Asteriou
(2007)
recommends
that
then
in
general
any
linear
combination
of
26
distribution.
However,
when
the
variables
become
(although
not
always)
the
error
itself
be
get
selects
the
squared
errors
large.
But
parameters
as
OLS
so
small
as
as
regression
to
make
possible
the
will
because
sum
it
of
the
select
any
excluding
the
constant
(10)
(which
only
affects
ut
(11)
Y0
eY1
eY2
...
eYi
2(X0
eX1
eX2
...
eXi)
or
ut
t
i 1
eYi
27
t
2
i 1
eXi
(12)
Hence,
the
assumptions
of
the
CLRM
are
case,
the
regression
equations
are
meaningless.
Case 3: Yt and Xt are integrated of the same order [often
I(1)]and the ut sequence contains a stochastic
trend. In this case, we have spurious regression
and it is often recommended to re-estimate the
regression
equation
in
the
semi-difference
10
28
are
said
to
be
cointegrated.
The
concept
of
[of
both
the
level
(ACF)
and
the
first
is
the
starting
point
of
formal
tests
of
is
the
correlation
between
variable
function
is
graph
of
the
to
Hanke
(2005),
the
autocorrelation
This is not explained in this lecture. You can make references from either
Gujarati (2003: 808-813), Hanke (2005: 60-74), or Nguyen Trong Hoai et al (2009:
Chapter 3, 4, and 8).
29
(i.e.
the
autocorrelation
coefficient
is
time
(Figure
series
4.1).
are
In
not
other
related
words,
to
Yt
each
and
other
Yt-k
are
independent.
(2) If a series has a (stochastic) trend, successive
observations
are
autocorrelation
significantly
highly
correlated,
coefficients
different
from
are
zero
for
and
the
typically
the
first
as
the
number
of
lags
increases.
The
large
coefficient
(close
for
time
to
lag
1).
The
will
autocorrelation
also
be
large.
series
is
stationary,
the
autocorrelation
12
See Nguyen Trong Hoai et al, 2009 and my lecture about ARIMA models.
30
of
lags
increases
(Figure
4.3).
In
other
but
Yt
and
Yt-k
[as
increases]
are
completely independent.
(4) If a series has a seasonal pattern, a significant
autocorrelation
seasonal
time
coefficient
lag
or
will
multiples
occur
of
at
seasonal
the
lag
31
32
The
correlogram
becomes
very
useful
for
time
series
statistic,
Ljung-Box
(LB)
statistic,
or
is
equivalent
to
testing
for
the
See Nguyen Trong Hoai et al, 2009 and my lecture about ARIMA models to understand
33
Yt =
Yt-1 + ut
(13)
= 1 (unity and hence
= 1,
< 1 (why?).
where
= (
Yt-1 Yt-1 + ut
Yt
= (
Yt
- 1)Yt-1 + ut
Yt-1 + ut
(14)
< 0 (why?).
to
Asteriou
Yt-1 + ut
(2007),
(15)
this
is
an
extremely
second
case
is
also
allow,
34
non-stochastic
time
Yt =
T +
Yt-1 + ut
(16)
test
on
the
coefficient
of
the
lagged
and
16).
This
test
does
not,
however,
have
values
which
were
originally
calculated
by
tau
statistic
(Gujarati,
2003;
2011).
However,
(1991,1996)
tabulated
appropriate
critical
values for each of the three above models and these are
presented in Table 4.1.
Table 4.1: Critical values for DF test
Model
Yt =
Yt-1 + ut
1%
5%
10%
-2.56
-1.94
-1.62
Yt =
Yt-1 + ut
-3.43
-2.86
-2.57
Yt =
T +
-3.96
-3.41
-3.13
-2.33
-1.65
-1.28
Yt-1 + ut
35
= 0. The DF test
If
the
DF
statistical
value
is
smaller
in
in
order
to
eliminate
autocorrelation
in
the
test equation.
The lag length14 on these extra terms is either determined
by
Akaike
Information
Bayesian/Information
usefully
by
the
Criterion
(AIC)
or
Schwarz
Criterion
(SBC,
SIC),
or
more
length
necessary
to
whiten
the
lag
Yt
Yt
Yt
i 1
14
36
ut
(17)
Yt
Yt
Yt
ut
(18)
i 1
p
Yt
Yt
Yt
ut
(19)
i 1
of
the
deterministic
elements
and
T.
The
critical values for the ADF test are the same as those
given in Table 4.1 for the DF test.
According to Asteriou (2007), unless the econometrician
knows
the
actual
data-generating
question
concerning
estimate
(17),
whether
(18),
or
it
is
(19).
process,
most
there
is
appropriate
Daldado,
Jenkinson
a
to
and
then
answering
appropriateness
of
set
each
of
model
questions
and
moving
regarding
to
the
the
next
or
not
of
deterministic
regressors.
However,
the
Phillips-Perron
(PP)
37
tests.
Because
the
distribution
theory
that
supporting
the
Dickey-Fuller
have
(1988)
developed
procedure
constant
that
variance.
Phillips
generalization
allows
for
of
fairly
and
the
mild
Perron
ADF
test
assumptions
the
ADF
correlation
test
by
corrects
adding
lagged
Yt-1 + et
for
(20)
higher
differenced
order
terms
serial
on
the
from
of
the
error
process.
The
expressions
are
series
performing
the
PP
test
as
well.
The
the
ADF
statistic
and
therefore
the
MacKinnon
38
Yt
Yt
Yt
ut
i 1
STOP: Conclude
that there is no
unit root
NO
= 0?
NO
NO
= 0?
STOP: Conclude
that Yt has a
unit root
YES
YES
Estimate the model
NO
Yt
Yt
Yt
ut
i 1
is
STOP: Conclude
that there is no
unit root
= 0?
NO
= 0?
YES
NO
STOP: Conclude
that Yt has a
unit root
YES
NO
Yt
Yt
Yt
STOP: Conclude
that there is no
unit root
ut
i 1
is
= 0?
YES
39
STOP: Conclude
that Yt has a
unit root
stationarity
(H0).
An
alternative
(Kwiatkowski-Phillips-Schmidt-Shin).
Its
test
test
is
KPSS
procedure
es for all t.
2
s
S2t
, and
2
critical
values
are
routinely
produced
by
Open
the
file
ADF.wf1
by
File/Open/Workfile and then choosing
name from the appropriate path.
40
clicking
the file
Step 2
Step 3
Step 4
Step 5
Step 6
41
statistic together
regression.
with
the
estimated
test
Step 7
Step 8
42
43
Open
the
file
PP.wf1
by
File/Open/Workfile and then choosing
name from the appropriate path.
Step 2
Step 3
Step 4
44
clicking
the file
Step 5
Step 6
Step 7
45
log(EX)
46
log(EX)
case
of
bivariate
model,
you
have
once
known
the
variable.
the
remedies
serial
such
The
OLS
correlation,
as
regression
and
we
semi-difference
often
perform
methods
47
analysis,
we
consider
the
simple
autoregressive
can
analyse
both
short-run
and
long-run
(21)
effects
Yt
Xt
B0
(22)
B0
1
B1
A1
(23)
Proof:
Yt
Xt
B0
Yt 1
Xt
A1
Yt
Xt
Yt 2
Xt
A1
Yt 1
Xt
Yt 3
Xt
A1
Yt 2
Xt
B1 = A1.B0
B1
(why?)
A1(A1.B0
B1)
(why?)
A12(A1.B0
B1)]
(why?)
Yt
Xt
A1
Yt
Xt
A1(A1.B0
48
B1)] (why?)
B0
[A1B0
B1]
A1[A1B0
B1]
A12(A1.B0
B1)]
...
A1(A1.B0
B1)]
(24)
A1B0
A1[A1B0
B1]
A12(A1.B0
B1)]
...
A1(A1.B0
(25)
B1)]
can
B0
1
=
also
B1
= equation (23)
A1
take
expectations
to
derive
the
long-run
E(Yt)
A0
1 A1
(B0
(1
B1)
E(Xt)
A1
= + E(Xt)
or simply to write:
Y* = + X*
Here,
(B0+B1)/(1-A1)
is
the
(26)
long-run
effect
of
49
Yt
Xt
B0
(27)
B0
YT
Xt
B1
A1
B2...
Bq
A2...
Ap
(28)
= A0 (1A1)Yt-1 + B0 Xt + (B0+B1)Xt-1 + ut
= B0 Xt (1A1) Yt
= B0 Xt (1A1) Yt
= B0 Xt
= B0 Xt
Yt
A0
(1 A1)
Xt
Xt
(B0
(1
1
B1)
Xt
A1)
+ ut
+ ut
+ ut
ECTt-1 + ut
(29a)
(29b)
equilibrium),
there
adjustment is given by
is
correction.
Speed
of
to
Asteriou
(2007),
the
concept
of
(1987),
Watson
Phillips
(1988),
and
Phillips
Ouliaris
(1986
and
(1990),
1987),
Stock
and
and
Johansen
is
known
that
trended
time
series
can
potentially
and
then
use
the
stationary
series
for
however,
is
not
ideal
because
it
not
only
error
processes.
called
as
These
stochastic
combination
cumulated
trends
of
error
and
two
cumulated
processes
normally
we
are
could
error
often
expect
51
them
which
eliminates
the
nonstationarity.
In
this
2007).
Cointegration
becomes
an
overriding
that,
if
there
really
is
genuine
long-run
common
trend
that
links
them
together.
For
an
of
Yt
and
Xt
can
be
directly
taken
from
(30)
t
u
Yt
52
2 t
(31)
gives
quarterly
expenditure
(PCE)
data
and
on
personal
personal
disposible
(i.e. after-tax) income (PDI) for the USA for the period
1970-2008 (Gujarati, 2011: pp.226). Both graph (Figure
6.1) and ADF tests (Table 6.1) indicate that these two
series are not stationary. They are I(1), that is, they
have stochastic trends. In addition, the regression of
log(PCE) on log(PDI) seems to be spurious (Table 6.3).
Since both series are trending, let us see what happens
if we add a trend variable to the model. The elasticity
coefficient is now changed, but the regression is still
spurious
(Table
6.4).
However,
after
estimating
the
the
obtained
[I(0)].
This
log(PCE)
residuals
implies
b1
that
b2log(PDI)
is
linear
b3T)
stationary
series
combination
cancels
(et
out
the
53
9.2
9.0
8.8
8.6
8.4
8.2
8.0
7.8
7.6
25
50
75
LOG(PCE)
100
125
150
LOG(PDI)
54
55
they
between
tells
have
them.
us
long-run,
In
that
the
there
or
equilibrium,
present
is
context,
strong
relationship
economic
relationship
theory
between
the
language
log(PCE)
B1
of
+
cointegration
B2log(PDI)
theory,
the
B3T
known
is
equation
as
single
equation,
the
simple
tests
of
residuals
regression.
Granger
estimated
These
(EG)
and
from
modified
Augmented
the
tests
cointegrating
by
the
Engle-Granger
Engle(AEG)
of
variables,
each
having
unit
root
(Gujarati, 2011).
6.4 Interpretation of the ECM
According
to
Asteriou
(2007),
the
concepts
of
56
the
general
linear
autoregressive
distributed
lag
(21)
that
he
(26)
long-run
effect
(slope
oe
elasticity)
Yt
Xt
+ ut
(29a)
or
Yt = B1 Xt
ECTt-1 + ut
(29b)
15
We can easily expand this model to a more general case for large numbers of
57
dynamics
captured
by
the
differenced
term.
us
with
information
about
the
= (1-A1)
speed
of
the
distance
the
system
is
away
from
58
The
coefficient
in
equation
(29a,b)
is
the
error-
In
fact,
tells
us
how
much
of
the
it
is
convenient
model
measuring
the
in
terms
typically
eliminate
involved;
they
of
first
trends
resolve
the
difference,
from
the
problem
of
variables
spurious
regressions.
16
59
which
(1981)
nonstationary
introduced
processes
and
remarkable
the
link
between
of
long-run
concept
Test
the
variables
for
their
order
of
test
each
variable
to
integration.
The
first
step
is
to
60
Fuller
and
the
augmented
Dickey-Fuller
tests
series
methods
apply
to
stationary
variables.
b) If the variables are integrated of different
orders, it is possible to conclude that they
are not cointegrated.
c) If both variables are integrated of the same
order, we proceed with step two.
Step 2
Yt
are
integrated
of
the
same
order
estimate
the
long-run
equilibrium
X
t
Yt
t
u
there
obtained
will
variables
yields
is
are
no
cointegration,
be
spurious.
However,
cointegrated,
consistent
61
the
estimators
OLS
for
results
if
the
regression
the
co-
integrating parameter .
Step 3
Check
for
(cointegration)
the
order
of
variables
are
order
to
determine
if
the
actually
cointegrated,
denote
residual
sequence
the
t
u
is
residuals
of
Thus,
from
the
the
series
estimated
equation
of
long-run
the
the
by
t .
u
estimated
relationship.
If
to
be
stationary,
the
Xt
and
Yt
are
cointegrated.
Step 4
the
variables
are
cointegrated,
the
to
analyse
the
long-run
and
short-run
coefficient,
which
is
the
end,
we
always
have
to
check
for
the
62
the
variables.
When
estimating
the
long-run
can
be
used
Consider,
for
example,
as
regressors
the
case
of
and
just
why.
two
on
is
the
residuals
equivalent
of
(i.e.
those
there
is
two
no
feature
obviously
of
the
becomes
EG
far
approach.
more
The
complicated
variables
there
may
be
more
than
one
63
can
not
treat
this
possibility.
So,
the
most
residual
series
and
the
second
step
is
to
are
needed
to
determine
the
order
of
the
second
step
involves
estimating
the
long-run
64
genr res1=resid
The third step (the actual test for cointegration) is a
unit root test on the residuals, the command for which
is:
adf res1
for no lags, or
adf(4) res1
for 4 lags in the augmentation term, and so on.
All
coeeficients
statistically
in
the
significant
at
65
table
are
6%
lower
or
individually
level.
The
ln(PCEt/PCEt-1).
This
is
the
short-run
consumption-
and
short-term
(quarterly
PCE
data),
is
corrected
suggesting
within
slow
rate
of
common
in
adjustment to equilibrium.
to
Asteriou
(2007),
it
is
quite
explanatory
variables
for
given
dependent
of
simultaneous
equations,
in
which
it
is
decision
regarding
such
differentiation
among
17
distinction
between
endogenous
66
and
exogenous
Yt = A1 + B1Xt +
CjYt
j 1
p
+ u1t
(32)
FjXt
+ u2t
(33)
EjYt
Xt = A2 + B2Yt +
DjXt
j 1
j 1
j 1
error
terms,
called
impulses
or
innovations
or
18
Gujarati (2011, pp.266) said that [from the point of view of forecasting] each
equation in VAR contains only its own lagged values and the lagged values of the
other variables in the system. Similarly, Wooldridge (2003, pp.620-621) said that
whether the contemporaneous (current) value is included or not depends partly on
the purpose of the equation. In forecasting, it is rarely included.
67
bivariate
VAR
often
has
the
following
features
the
number
of
lagged
values
of
each
one
lagged
value
of
each
variable
on
the
VAR
system
can
be
extended
to
several
variables.
(4) In the two-variable system, there can be at most
one
cointegrating,
between
them.
If
system,
there
can
or
we
be
equilibrium,
have
at
a
most
relationship
three-variable
two
VAR
cointegrating
68
variables
VAR
[that
using
are
first
I(0)]
differences
to
remove
of
common
trends.
The
variables
are
cointegrated:
the
error
The
model
becomes
vector
error
variables
are
not
cointegrated:
the
VAR
models
are
in
most
cases
better
than
those
(see
Mahmoud,
1984;
69
McNees,
1986).
Besides
the
following
aspects.
First,
they
are
a-theoretic
there
are
no
restrictions
on
any
of
the
models
that
might
have
an
underlying
consistent
to
the
loss
of
degrees
of
freedom.
Thus,
if
the
creating
problems
in
estimation.
Third,
the
since
they
totally
lack
any
theoretical
background.
this
section,
we
extend
the
single-equation
error
70
(34)
1Zt-1
2Zt-2
+ +
p-1Zt-p+1
Zt-1 + ut
include
(35)
the
speed
of
adjustment
where
to
equilibrium
the
Zt-1
term
is
equivalent
to
the
error
except
that
now
Zt-1
contains
up
to
(p
1)
two
lagged
terms,
and
the
model
is
then
the
following:
Yt
Xt
Wt
Yt
Xt
Wt
1
1
1
Yt
Xt
Wt
1
1
1
or
71
ut
(33)
Yt
Xt
Wt
Yt
Xt
Wt
11
21
Yt
Xt
Wt
12
11
21
31
12
22
32
22
31
32
ut
(34)
= ([
1111
1212][
1131
1121
1232])
1222]
Yt
Xt
Wt
(35)
11(11Yt-1
+ 21Xt-1 + 31Wt-1) +
12(12Yt-1
+ 22Xt-1 + 32Wt-1)
(36)
11
and
12.
11
differing
21
speeds
31).
72
of
adjustment
coefficients
21
relationship
31
exists,
can
we
then
say
that
the
21
31
weakly exogenous.
Suppose that we have k variables in a VECM, the
matrix
contains
the
error
correction
terms
k k
(linear
Implications
Rank of
r = 0
0 < r < k
r = k
73
we
can
have
only
up
to
k-1
co-integrating
Testing the
variables.
Step 2
Setting
model.
the
order
of
integration
appropriate
lag
length
of
all
of
the
most
common
procedure
in
choosing
the
74
1Zt-1
+ +
k-1Zt-p+1
(Z t
1 t) Zt-1
2t
+ ut
(37)
In
general
five
distinct
models
can
be
considered. Although the first and the fifth
75
1:
No
intercept
or
trend
2:
Intercept
(no
trend)
in
CE
=
in
=
(co=
CE,
no
= 0).
= 0).
= 0).
integrating vectors.
There are two methods for determining the
number of co-integrating relations, and both
involve estimation of matrix
(1) One
method
tests
the
.
null
hypothesis,
76
on the
called
>
>
> >
n.
If the variables
rank
of
is
zero
and
will
all
equal
the
characteristic
roots
zero.
Therefore, ( 1
) will be equal to 1
i
1)
T ln(1
r 1)
(38)
max).
(2) The
second
method
is
based
on
a
likelihood ratio test about the trace of
the matrix (and because of that it is
called the trace statistic). The trace
statistic considers whether the trace is
increased by adding more eigenvalues
beyond the rth eigenvalue. The null
hypothesis in this case is that the
number of co-integrating vectors is less
77
ln(1
i r 1
r 1)
(39)
statistic
is
greater
than
the
critical
values
specified.
Table 8.3: Trace test
H0
H1
Statistic
95% Critical
Decision
r = 0
r = 1
62.18
47.21
Reject H0
r = 2
19.55
29.68
Accept H0
r = 3
8.62
15.41
Accept H0
r = 4
2.41
3.76
Accept H0
78
We
conclude
that
this
data
exhibits
one
cointegrating
vector.
8.2.3 The Johansen approach in Eviews
Eviews
has
cointegration
statistics.
specific
using
Consider
command
Johansen
the
file
for
approach
testing
under
JOHANSEN.wf1,
for
group
which
has
1:
Determine
the
order
of
integration
for
the
variables.
To do this, we apply the unit-root tests on all three
variables. We apply the Doldado, Jenkinson and SosvillaRivero
(1990)
procedure
for
choosing
the
appropriate
79
80
according
to
the
data
interval
(year,
quarter,
81
the
VAR
estimates,
select
View/Lag
test
Eviews
each
by
one
opening
of
the
models
Quick/Group
for
cointegration
in
Statistics/Cointegration
82
X Y Z
then press <OK>. The five alternative models explained in
Step 3 above are given under labels 1, 2, 3, 4, and 5.
There
is
another
option
(option
in
Eviews)
that
has
the
(default
by
Eviews)
numbers
for
83
84
85
From
this
summary,
we
see
that
Model
(Linear,
determining
optimal
lag
length,
and
number
of
VAR
Correction
Specification,
in
VAR
type,
choose
add
all
Vector
Error
variables
in
86
(3) We
then
enter
number
Cointegration:
Figure 8.4: Cointegration
87
of
cointegrating
in
9. CAUSALITY TESTS
According to Asteriou (2007), one of the good features of
VAR models is that they allow us to test the direction of
causality.
different
Causality
to
the
in
econometrics
concept
in
everyday
is
somewhat
use
(take
other
with
distributed
88
lags.
The
relationship
it
is
possible
to
have
that
(a)
Yt
causes
Xt
The
problem
is
to
find
an
appropriate
Xt,
if
Xt
can
be
predicted
with
greater
not
using
such
past
values,
all
other
terms
Granger
causality
variables, say,
test
Yt and
for
the
of
two
stationary
Xt = A2 +
p
j 1
p
j 1
Cj Yt
Ej Yt
p
j 1
p
j 1
Dj Xt
Fj Xt
+ u1t
+ u2t
(40)
(41)
89
order
1.
In
this
model,
we
can
terms
in
have
the
following
different cases:
Case 1
The
lagged
equation
(40)
are
lagged
terms
in
equation
(41)
are
not
The
lagged
terms
in
equation
(41)
are
lagged
terms
in
equation
(40)
are
not
X and lagged
Y terms are
(40)
and
(41),
so
that
we
have
bi-
X and lagged
Y terms are
First,
estimate
the
VAR
model
given
by
90
direction
of
causality
based
upon
the
four
cases
mentioned above.
More analytically, and for the case of one equation (i.e.
we
will
examine
equation
(40)),
it
is
intuitive
to
Regress
Yt
on
lagged
terms
as
in
the
following model:
Yt
j 1
Cj Yt
u1t
(42)
Regress
Yt on lagged
Yt
A1
j 1
Cj Yt
p
j
j 1
Dj Xt
u1t
(40)
Set
the
null
and
alternative
hypotheses
as
below:
H0 :
H1 :
Step 4
p
j 1
p
j 1
Dj
Dj
0 or Xt does cause Yt
91
(RSSR
RSSU)/ p
RSSu /(N k)
20
However, this is not a good way of conducting Granger causality test (why?)
92
Note
that
this
lag
specification
93
is
not
highly
Therefore,
when
we
variable
Yt
Xt,
Sims
causes
want
to
check
suggests
whether
estimating
the
Yt
A1
Xt
A2
p
j 1
Cj Yt
p
j 1
Ej Xt
p
j
j 1
Dj Xt
p
j
j 1
Fj Yt
m
j
i 1
Xt
Yt
m
j
i 1
u1t
(43)
u2t
(44)
94
the
testing
m
i 1
leading
for
the
values
lagged
of
X.
values
Therefore,
of
Xt,
we
instead
of
test
for
is
unclear
which
version
of
the
two
tests
is
95
necessary,
which
in
turn
introduces
additional
an
information
criterion.
One
such
SIC(p)
where
RSS(p)
is
the
ln
RSS(p)
T
sum
of
(p
1)
squared
ln T
T
(45)
residuals
of
the
pmax,
where
pmax
is
the
largest
value
of
considered.
The formula for SIC might look a bit mysterious at first,
but it has an intuitive appeal. Consider the first term
in equation (45). Because the regression coefficients are
estimated
necessarily
21
by
OLS,
decreases
the
(or
sum
at
of
least
squared
does
not
Others including FPE, HQ, and LR are also used in empirical studies.
96
residuals
increase)
AIC(p)
ln
RSS(p)
T
(p
1)
2
T
(46)
The difference between the AIC and the SIC is that the
term lnT in the SIC is replaced by 2 in the AIC, so
the second term in the AIC is smaller (why?). Stock and
Watson (2007) state that the second term in the AIC is
not large enough to ensure that the correct lag length is
chosen, even in large samples, so the AIC estimator of p
is not consistent.
Despite this theoretical shortcoming, the AIC is widely
used in practice. If you are concerned that the SIC might
yield
model
with
two
few
lags,
the
AIC
reasonable alternative.
97
provides
98
99
determine
(and
fix)
the
optimal
lags
for
the
variables.
If
the
regression
model
has
SIC(K)
ln
RSS(K)
T
ln T
T
(47)
AIC(K)
ln
RSS(K)
T
2
T
(48)
100
There
are
two
important
using
an
information
practical
criterion
considerations
to
estimate
when
the
lag
candidate
models
must
be
estimated
over
the
same
different
models.
According
to
Stock
and
Watson
This
model
was
developed
by
Pesaran
et
al.
of
regressors
are
cointegrated).
the
variables
purely
This
(irrespective
I(0),
is
purely
specially
I(1)
linked
of
or
whether
mutually
with
the
ECM
estimating
the
ARDL
modelling
approach
involves
Yt
0y
1yYt
Xt
0x
1xYt 1
2y Xt
m
1
2xXt 1
i 1
Yt
Xt
m
i 1
101
m
i
j 1
m
i
j 1
Xt
u1t
(49)
Yt
u2t
(50)
for
investigating
relationships.
In
the
one
case
of
or
one
more
or
long-run
more
long-run
1y
2y
On
the
= 0
1y
0.
other
dependent
hand,
variable,
cointegration is H0:
in
equation
the
1x
(50),
null
2x
when
hypothesis
is
the
of
no
1x
2x
0.
102
previous
studies,
would
like
to
suggest
the
analysis
of
the
interaction
among
savings,
relationship
between
domestic
savings
and
relationship
between
savings
and
growth:
foreign
direct
investment
and
gross
domestic
direct
An
investment
empirical
study
and
of
economic
growth
causality
and
in
error
correction mechanisms
The
interactions
among
foreign
direct
investment,
and
economic
growth
in
Vietnam:
Granger
causality analysis
Export
expansion
and
economic
growth:
Testing
for
between
financial
development
and
economic
104
long-run
relationship
between
stock
returns
and
inflation in Vietnam
The
relationship
between
financial
deepening
and
adjustment
in
the
long-run
relationship
relationship
between
gas
consumption
and
relationship
between
energy
105
consumption
and
consumption
Evidence
from
and
economic
cointegration
growth
and
in
error
Vietnam:
correction
model
The causality between energy consumption and economic
growth in Vietnam
The
relationship
macroeconomic
between
performance:
the
price
Empirical
of
oil
and
evidence
for
Vietnam
growth
and
government
expenditure:
Evidence
from Vietnam
Government
revenue,
government
expenditure,
106
and
The
relationship
between
budget
deficits
and
money
causality
between
money
and
income
for
the
Vietnam economy
Money, inflation and casuality: Evidence from Vietnam
Money-output Granger causality: An empirical analysis
for Vietnam
Time-varying
parameter
error
correction
models:
The
transmission
mechanism
in
Vietnam:
VAR
analysis
tourism
and
economic
development
in
107
of
rice
prices
and
agricultural
wages
in
Vietnam
Macroeconomic
factors
and
agricultural
production
Others
Hypotheses
testing
concerning
relationships
between
relationship
between
macroeconomic
variables
and
adjustment
and
private
manufacturing
108
rates
and
inflation
using
cointegration
techniques
The
long-run
relationship
between
house
prices
and
It
is
noted
nonstationary
that
panels,
many
empirical
typically
studies
characterised
use
by
the
panel
109
REFERENCES
Asteriou, D. and Hall, S.G. (2007) Applied Econometrics:
A Modern Approach Using Eviews and Microfit, Revised
Edition. Palgrave Macmillan.
Cheung, Y.W. and Lai, K.S. (1995) Lag Order and Critical
Values of the Augmented Dickey-Fuller Test, Journal of
Business & Economic Statistics, Vol.13, No.3, pp.277280.
Dickey, D.A. and Fuller, W.A. (1979) Distribution of the
Estimators for Autoregressive Time Series with a Unit
Root, Journal of the American Statistical Association,
Vol.74, No.366, pp.427- 431.
Dickey, D.A. and Fuller, W.A. (1981) Likelihood Ratio
Statistics for Autoregressive Time Series with a Unit
Root, Econometrica, Vol.49, p.1063.
Diebold, F.X. (2004) Elements of Forecasting, 3rd Edition,
Thomson.
Dolado,
J.,
T.Jenkinson
Cointegration
and
Unit
and
S.Sosvilla-Rivero.
Roots,
Journal
of
(1990)
Economic
J.
(1970)
Testing
for
Serial
Correlation
in
Dependent
Variables,
Econometrica,
Vol.38,
pp.410-421.
Engle, R.F. and Granger, C.W.J. (1987) Co-integration
and
Error
Correction
Estimates:
Representation,
110
C.W.J.
(2004)
Time
Series
Analysis,
C.W.J.
and
Newbold,
P.
(1977)
Spurious
W.E.,
R.C.Hill
and
G.C.Lim.
rd
(2008)
Using
Edition, John
D.N.
(2003)
Basic
4th
Econometrics,
Edition,
McGraw-Hill.
Gujarati,
D.N.
(2011)
Econometrics
by
Example,
1st
J.E.
and
Wichern,
D.W.
(2005)
Business
and
Inference
111
on
Cointegration,
with
Tests,
Journal
of
Business
&
Economic
X.
(2001)
Expenditure,
Government
and
Temporal
Revenue,
Causality:
Government
Evidence
from
J.G.
Distribution
(1994)
Approximate
Functions
Cointegration,
Journal
for
of
Asymptotic
Unit-Root
Business
&
and
Economic
Unit
Root
and
Cointegration
Tests,
Journal
of
for
Cointegration,
Journal
of
Applied
Comparison
112
of
US
Macroeconomic
Forecasts,
Journal
of
Business
and
Economic
M.
(2007)
Energy-GDP
Relationship
for
Oil-
P.C.B.
Spurious
(1998)
New
Regressions,
Tools
for
Econometrica,
Understanding
Vol.66,
No.6,
pp.1299-1325.
Phillips, P.C.B. and Perron, P. (1988) Testing for a
Unit
Root
in
Time
Series
Regression,
Biometrica,
D.L.
Econometric
R.S.
and
Rubinfeld,
R.
(2002)
Applications, 5
Sims,
C.A.
th
th
(1998)
Edition, McGraw-Hill.
Introductory
Econometrics
with
(1980)
Macroeconomics
and
Reality,
J.H.
and
Watson,
M.W.
(2007)
Introduction
to
Vector
Autoregressive
113
with
Possibly
Integrated
Processes,
Journal
of
Econometrics,
Vol.66,
No.1,
pp.225-250.
Vogelvang,
B.
(2005)
Econometrics:
Theory
and
J.M.
(2003)
Introductory
114
Econometrics:
APPENDIX
STATA COMMANDS
Source: Practical exercises, Advanced Econometrics course
2012, Wageningen University, The Netherlands.
EXAMPLE 1
Use the data set WHEATOIL.dta, which contains (nominal)
prices of wheat (pwht), nominal oil prices (poil) and a
time indicator (t). The data are monthly and available
for the period Jan 1990 till December 2008 (19 years*12
months = 228 obs.). In this example, we will investigate
whether there is a long-run relationship between wheat
prices and oil prices. There may be all kinds of reasons
for such a relationship: oil is an important input in
fertilizer production, is used for applying machinery,
drives transportation costs, etc.
Please declare the data to be time-series data using the
following command:
tsset t
1. Create one graph with line plots of both pwht and poil
against t. Considering the line plot for pwht, do you
think this variable is stationary? Motivate your
answer.
115
20
100
40
60
poil
200
150
pwht
250
80
300
100
ptbinh@ueh.edu.vn
50
100
150
200
250
t
pwht
poil
The graphs for both pwht and poil indicate that there are
stochastic trends (means are not constant) and their
variances are also not constant. For the pwht, it first
increases and highly fluctuates (from observation 1 to
about
70),
followed
by
a
declining
period
(from
observation about 70 to about 120) with less fluctuation,
then it tends to increase and especially decline very
quickly in the last months. Therefore, we might say that
pwht is not stationary.
2. Use appropriate tests to find out
integration of both pwht and poil?
116
the
orders
of
117
118
119
KPSS test
120
121
122
123
124
ADF test
125
126
127
-50
50
Residuals
100
150
ptbinh@ueh.edu.vn
5.
50
100
150
200
250
128
5.1 Residual-based
Granger approach)
test
for
no
cointegration
(Engle-
the
residuals
are
not
stationary
(no
129
130
131
132
133
134
We recall that:
Null hypothesis in Augmented
series is not stationary
Null Hypothesis
stationary
in
KPSS
Dickey
test:
135
the
Fuller
time
test:
series
time
is
AUSTIN
136
137
138
139
DALLAS
140
Both ADF and KPSS tests indicate that the housing price
in Dallas is non-stationary. We now test the stationarity
of its first-differenced data.
141
142
143
HOUSTON
144
Both ADF and KPSS tests indicate that the housing price
in
Houston
is
non-stationary.
We
now
test
the
stationarity of its first-differenced data.
145
146
SAN ANTONIO
147
148
149
Both ADF and KPSS tests indicate that the firstdifferenced data of housing price in San Antonio is
stationary. Therefore, the housing price in San Antonio
is integrated of order on [I(1)].
150
151
152
153
154
Adj.Parameter
D_austin
D_dallas
in
the
VECM
model
can
be
Coef.
P-value
Significance
11
-0.148
0.012
Yes
12
-0.040
0.742
No
21
0.073
0.128
No
155
D_houston
D_sa
22
-0.309
0.002
Yes
31
0.190
0.000
Yes
32
0.604
0.000
Yes
41
0.283
0.000
Yes
41
-0.178
0.185
No
156
If the lags are 4, we can see that there are two long-run
cointegrating relationships between/among house prices
of: (i) Austin, Houston and San Antonio; and (ii) Dallas,
Houston, and San Antonio.
157