Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 51

Time Series Econometrics:

Basic Concept

Chapter 21

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 1


Introduction

 Regression analysis based on time series


data implicitly assumes that the underlying
time series are stationary. The classical t
tests, F tests, etc. are based on this
assumption.

 In practice most economic time series are


nonstationary.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 2


Introduction

 Sometimes autocorrelation results because


the underlying time series is nonstationary.

 Sometimes in regressing time series data


obtain a very high R2 even though there is no
meaningful relationship between two
variables  spurious regression.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 3


Selected US Economic Time Series
5000

4000
“trending” upward

3000

2000

1000
70 72 74 76 78 80 82 84 86 88 90

GDP PDI PCE

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 4


Selected US Economic Time Series
250

200
“trending” upward
150

100

50

0
70 75 80 85 90

PROFITS DIVIDENDS

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 5


Selected INA Economic Time Series
500000

400000
“trending” upward
300000

200000

100000

0
84 86 88 90 92 94 96 98 00 02
KONS Y YD

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 6


Selected INA Economic Time Series

200000

150000 “trending” upward

100000

50000

0
84 86 88 90 92 94 96 98 00 02
X M

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 7


Key concepts

1. Stochastic processes
2. Stationary stochastic processes
3. Purely random processes
4. Non stationary processes
5. Integrated variables
6. Random walk models
7. Cointegration
8. Deterministic and stochastic trends
9. Unit root tests
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 8
Stochastic processes

 A random or stochastic process is a


collection of random variables ordered in
time.
 If Y denote a random variable, and if it is
continuous, we denote it as Y(t), but if is is
discrete, we denoted it as Yt

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 9


Stationary stochastic processes(1)

 A stochastic process is said to be stationary if its


mean and variance are constant over time and
the value of the covariance between the two
time 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.
 Known as: weakly stationary, covariance
stationary, second-order stationary, wide sense

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 10


Stationary stochastic processes(2)

 Let Yt be a stochastic time series with these


properties
Mean E  Yt   
var  Yt   E  Yt     
2 2
Variance
Covariance  k  E  Yt     Yt  k    
 If a time series is stationary, its mean, variance, and
autocovariance (at various lags) remain the same
no matter at what point we measure them; they are
time invariant

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 11


Stationary stochastic processes(3)

 A nonstationary time series will have a time-


varying mean or a time-varying variance or both.
 Why are stationary time series so important?
 Because if a time series is nonstationary, we can
study its behavior only for the time period under
consideration. It is not possible to generalize it to
other time periods. For the purpose of
forecasting, a nonstationary time series may be
of little practical value.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 12


Purely random processes

 A special type of stochastic process: a


stochastic process purely random or white
noise  if has zero mean, constant variance
2, and serially uncorrelated. If also
independent  strictly white noise

 
ui ~ IIDN 0,  2 
that is ut is independently and
identically distributed as a normal distribution
with zero mean and constant variance.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 13


Nonstationarity processes

 Example: Random Walk Model (RWM)


 Two types of random walk
1. Random walk without drift (i.e. no constant
or intercept term)
2. Random walk with drift (i.e. a constant term
is present)

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 14


Random walk without drift

 Suppose ut is a white noise error term with


mean 0 and variance 2, then the series Yt is
said to be a random walk if
Yt  Yt 1  ut
 We can write:
Y1  Y0  u1
Y2  Y1  u2  Y0  u1  u2
Y3  Y2  u3  Y0  u1  u2  u3

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 15


In general: Yt  Y0   ut
E  Yt   E  Y0   ut   Y0
var  Yt   t 2

 As t increase, its variance increases


indefinitely, thus violating a condition of
stationary.
 The RWM without drift is a nonstationary
stochastic process

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 16


Yt  Yt 1  ut
Yt  Yt 1  Yt  ut

 It is easy to show that, while Yt is nonstationary,


its first difference is stationary.
 The first difference of a random walk series are
stationary

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 17


Random walk with drift
Yt    Yt 1  ut
Yt  Yt 1  Yt    ut
where  is known as the drift parameter.
E  Yt   Y0  t
var  Yt   t 2

the variance increases over time, violating the


condition of (weak) stationarity
 In RWM, with or without drift, is a nonstationary
stochastic process.
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 18
Unit root stochastic process

 Write the RWM as


Yt   Yt 1  ut 1    1
 If =1  the RWM (without drift)  unit root
problem, that is, a situation of stationarity.
The term nonstationarity, random walk and unit
root can be treated as a synonymous.
 If ||1  the time series Yt is stationary

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 19


Trend Stationary (TS) and Difference
Stationary (DS) Stochastic Process

 If the trend in a time series is completely


predictable and not variable  a deterministic
trend
 If it is not predictable  a stochastic trend
 The model of the time series Yt
Yt  1   2t  3Yt 1  ut
ut is a white noise error term; t is time

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 20


Trend Stationary (TS) and Difference
Stationary (DS) Stochastic Process
 Pure random walk:
1  0,  2  0,  3  1
Yt  Yt 1  ut nonstationary
Yt   Yt  Yt 1   ut stationary

a RWM without drift is a difference stationary


process (DSP)

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 21


 Random walk with drift
1  0,  2  0, 3  1
Yt  1  Yt 1  ut nonstationary
Yt   Yt  Yt 1   1  ut

a positive or negative trend  a stochastic trend


The model is a DSP because the nonstationary in
Yt can be eliminated by taking first difference of
the time series.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 22


 Deterministic trend
if 1  0,  2  0,  3  0
we obtain :
Yt  1   2t  ut nonstation ary
Yt   1   2t   ut trend stationary

is called a trend stationary process (TSP). The


mean is not constant, its variance is.
The procedure of removing the deterministic
trend is called detrending.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 23


 Random walk with drift and deterministic trend
if 1  0,  2  0,  3  0
we obtain : Yt  1   2t  Yt 1  ut
Yt  Yt 1  1   2t  ut
Yt  1   2t  ut Yt is nonstation ary
 Deterministic trend with stationary AR(1)
component
if 1  0,  2  0,  3  0
we obtain :
Yt  1   2t   3Yt 1  ut
which is stationary around the deterministic trend
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 24
Integrated Stochastic Processes
 The random walk model is but a specific case of
a more general class of stochastic processes
known as integrated processes.
 RWM without drift is nonstationary, but its first
difference is stationary.
 RWM without drift  integrated of order 1 ~ I(1).
 If a nonstationary time series has to be
differenced d times to make it stationary, that
time series is said to be integrated of order d ~
I(d)

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 25


Properties of Integrated Series

Let Xt, Yt, and Zt be three time series


1. If Xt ~ I(0) and Yt ~ I(1), then Zt=(Xt+Yt) ~ I(1)

2. If Xt ~ I(d), then Zt=(a+bXt) ~ I(d), where a


and b constants.
3. If Xt ~ I(d1) and Yt ~ I(d2), then Zt=(aXt+bYt) ~
I(d2), where d1<d2
4. If Xt ~ I(d) and Yt ~ I(d), then Zt=(aXt+bYt) ~
I(d*); d* is generally equal to d, but in some
case d*<d.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 26


The Phenomenon of Spurious
Regression
 Phenomenon of spurious or nonsense
regression
 Yule (1926): spurious correlation could
persist in nonstationary time series even if
the sample is very large
 Granger & Newbold (1974): an R2 > d is a
good rule of thumb to suspect that the
estimated regression is spurious

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 27


Tests of Stationarity:
Two important questions

1. How do we find out if a given time


series is stationairy?
2. If we find that a given time series is not
stationarity, is there a way that it can be
made stationarity?

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 28


Tests of Stationarity: Graphical
Analysis
5000
 if a given time series
over the period have
4000
been increasing
(upward trend) or 3000
decreasing
(downward trend), 2000
suggesting that a
given time series is 1000
70 72 74 76 78 80 82 84 86 88 90
not stationary.
GDP PDI PCE

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 29


Tests of Stationarity: Autocorrelation
Function (ACF) and Correlogram
 Population Autocorrelation Function (PACF)
 k covariance at lag k
k   1  k  1
0 variance
 Sample Autocorrelation Function (SACF)

ˆ k 
ˆk
ˆk 
  Y  Y Y t t k Y 
ˆ0 n

ˆ0 
  Y Y  t
2

n
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 30
 The choice of lag length
 Rule of thumb: one-third to one-quarter the length of
the time series
 Akaike / Schwarz Information Criterion

 Statistical significance
ˆ k ~ N  0,1 n 
The 95% confidence interval : ˆ k  1,961 n 
in other words,
Prob  ˆ k  1,961 n    k  ˆ k  1,961 n    0,95

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 31


 If the interval includes the value of zero, we do
not reject the hypothesis that the true  k is zero,
but if this interval does not include 0, we reject
the hypothesis that the true  k is zero

 Joint hypothesis: all the  k up to certain lags


are simultaneously equal to zero.
Q-statistic (Box-Pierce, 1970):
m
Q  n ˆ ~   df  m 
2
k
2

k 1

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 32


 Ljung-Box (LB) statistic  a variant of Box-
Pierce Q-Statistic

 ˆ k2 
m
LB  n n  2    
 ~  2
 df  m 
k 1  n  k 
In small sample, the LB statistic has been
found to have better (more powerfull) than the
Q statistic

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 33


Tests of Stationarity: The Unit Root
Test
 We start with
Yt  Yt 1  ut ut is a white noise error
 If =1 random walk model without drift
(nonstationary stochastic process)
 For, theoretical reason, we manipulate
Yt  Yt 1  Yt 1  Yt 1  ut
Yt     1Yt 1  ut
Yt  Yt 1  ut
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 34
 If  = 0, then  = 1, that is we have a unit root
 the time series is nonstationary
 Since  = ( - 1), for stationary  must be less
than one. For this to happen  must be
negative.
 The t value of the estimated coefficient of Yt-1
does not follow the t distribution even in large
samples.
 Dickey-Fuller test   (tau) statistic

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 35


The Dickey-Fuller (DF) Test
Yt is a random walk : Yt  Yt 1  ut
Yt is a random walk with drift : Yt  1  Yt 1  ut
Yt is a random walk with drift
arround a stochastic trend : Yt  1   2t  Yt 1  ut
assume : ut is uncorrelat ed
In each case,
H0 :   0 the time series is nonstationary
Ha :   0 the time series is stationary

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 36


 Important: the critical value of the tau test to
test the hypothesis that =0, are different for
each of three specifications of the DF test.

 If || > critical tau value  reject Ho, the time


series is stationary
 If || ≤ critical tau value  do not reject Ho,
the time series is nonstationary

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 37


The Augmented Dickey Fuller Test

 Assume ut is correlated.
 Adding the lagged values of the dependent
variables, ΔYt
m
Yt  1   2t  Yt 1   i  Yt i   t
i 1

t is a pure white noise error term

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 38


The Phillips-Perron (PP) Unit Root
Tests
 The DF test: ut are independently and
identically distributed
 The ADF test: take care of possible serial
correlation in the error terms by adding the
lagged difference terms of regressand
 The PP test: use nonparametric statistical
methods to take care of the serial correlation
in the error terms without adding lagged
difference terms
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 39
Why are there so many unit root
tests?
 Size of test  the level of significance (the
probability of committing a Type I error
 Power of test  the probability of rejecting
the null hypothesis when it is a false.
Calculating by subtracting the probability of a
Type II error from 1.
Type II error: the probability of accepting a
false null hypothesis
The maximum power is 1.
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 40
A Critique of the unit root tests

 Size of test: the DF test is sensitive to the


way it is conducted.
Example: if the true model is a pure random
walk but we estimate a random walk with
drift, and conclude that on the 5 percent level,
the time series is stationary. This conclusion
may be wrong because the true level of
significance is much larger than 5 percent.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 41


 Power of tests. The DF test have low power,
they tend to accept the null of unit root.
There are several reasons:
1. The power depends on the span of the data more
than size of the sample.
2. If  ≈ 1 but not exactly 1, the unit root test may
declare such a time series nonstationary.
3. These type of tests assume a single unit root. If a
time series is integrated of order higher than 1, there
will be more than one unit root  use the Dickey-
Pantula test
4. If there are structural breaks in a time series, the unit
root tests may not catch them.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 42


 There are have been modifications of these
tests:
1. Perron and Ng (PN)
2. Elliot, Rothenberg, and Stock (ERS)
3. Fuller
4. Leybounre
 Maddala & Kim (1998): the tradisional DF,
ADF, and PP tests should be discarded.
 But there is no uniformly powerful test of
unit root hypothesis

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 43


Transforming Nonstationary
Time Series
 Difference-Stationary Processes
If a time series has a unit root, the first
difference of time series are stationary.
 Trend-Stationary Processes
If a time series has a unit root around the
trend line, the simplest way to make a time
series stationary is to regress it on time and
the residuals from this regression will be
stationary.
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 44
In other words,
run Yt  1   2t  ut

uˆt  Yt  ˆ1  ˆ2t  stationary
uˆt is a linearity detrended time series

The trend may be nonlinier: Yt  1   2t   3t  ut


2

 Pointed out:
 If a time series is DSP, but we treat it as TSP 
underdifferencing.
 If a time series is TSP, but we treat it as DSP 
overdifferencing

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 45


Cointegration:
Regression of a Unit Root Time Series on another
Unit Root Time Series
 The regression of a nonstationary time series on
another nonstationary time series may produce a
spurious regression.
 Suppose Xt~I(1) and Yt~I(1), the linear combination
is I(0)
Yt  1   2 X t  ut
ut  Yt  1   2 X t ut ~ I  0 

X and Y are cointegrated


01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 46
 Economically speaking:
“two variables will be cointegrated if they
have a long-term, or equilibrium, relationship
between them”
 The valuable contribution of the concepts of
unit root, cointegration, etc is to force us find
out if the regression residuals are stationary.
 Granger (1986): a test for cointegration can
be thought of as a pre-test to avoid spurious
regression situations.

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 47


Testing for Cointegration
 Engle-Granger(EG) or Augmented Engle-
Granger (AEG) Test
 Run the cointegrating regression and obtain the
residuals
 Use DF and ADF test on the residuals
 Cointegrating Regession Durbin-Watson
(CRDW) Test (Sargan & Bhargava, 1983)
 Use the d value obtained from cointegrating
regression, but the null hypothesis is d=0 rather
than d=2

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 48


Cointegration and Error
Correction Mechanism (ECM)
 If X and Y are cointegrated  there is a long-
term, or equilibrium, relationship between
them.
 In short-run there may be disequilibrium 
equilibrium error. We can use this error term
to tie the short-run behavior of Y to its long-
run value.
 The error correction mechanism  Sargan 
Engle-Granger
01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 49
 Engle-Granger Representation Theorem
“if two variables Y and X are cointegrated, then the
relationship between the two can be expressed as
ECM”

Cointegtrating regression : Yt  1   2 X t  ut
ut  Yt  1   2 X t ut ~ I  0 
ut 1  Yt 1  1   2 X t 1
ECM : ΔYt  α0  α1 ΔX t  α2ut 1  et
2  0
2  speed of adjustment: how quickly the
equilibrium is restored?

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 50


 www.addebook.com
 www.ebookee.com
 www.ebook3000.com
 http:/ideas.repec.org
 www.ssrn.com
 www.imf.org
 www.irti.org

01/06/22 Gujarati, 4e (2003) prepared by Sri Yani K 51

You might also like