KIGALI INDEPENDENT

UNIVERSITY ULK

SCHOOL OF ECONOMICS AND

BUSINESS STUDIES

DEPARTMENT OF ECONOMICS

Econometrics II

Prepared by

MWANGABWOBA JANVIER (MSC Applied statistics)

 Table of content

Chap1: Time Series Concepts

Chap2: Econometric Modelling: Model
specification and diagnostic testing
Chap3: Unit roots,Cointegration and Error
correction Model
Chap.4: Simultaneous equation models
 Chap1: Time Series Concepts

1.1 What is a time series?

 Chap1: Time Series Concepts

1.2 Uses of time series

Time series can be used for the following purpose:
- Forecasting;
- Trend removal;
- Seasonal adjustment;
- Detection of structural breaks;
- To answer questions of causality;
- Distinguish between short and long-run;
- Study of agent’s expectations ...
 Chap1: Time Series Concepts

1.3 A stochastic process

A stochastic process is “a collection of random variables ordered in
time”. The term stochastic comes from the Greek word stokhos,
which means target or bull’s eye.

Out of numerous arrow shots, you may be lucky to hit the bull’s eye
only a few times, most times the arrows will be spread randomly
around the bull’s eye (depending on your skill, of course)
 Chap1: Time Series Concepts
1.4 White noise
In time series terms, white noise is described as a series of
identically distributed random variables (IID). The Statistical
properties of a white nose series are as follow:
-A constant zero mean over time,
- A constant variance over time,
- A constant auto-covariance over time.
 Chap1: Time Series Concepts

1.5 Autoregressive processes

The simplest, purely statistical time series model is the First-Order
Autoregressive process - AR(1).

where εt is an uncorrelated random error term with zero mean and

constant variance, i.e. white noise.
The Second-Order Autoregressive process - AR(2) is given by,

In general, a pth-Order Autoregressive process – AR(p) is given

by,
Generating an autoregressive process in Eviews:
E.g: Xt = 0.5Xt-1 + εt, where εt ~ NIID(0,1)
Open Eviews ---- file ----- new----- workfile ------click undated or irregular
determine the end of observation (e.g. 10000) ----- OK.
Go back to: file ----- new ----- program -----write the following command

smpl @first @first

series x=nrnd
smpl @first+1 @last
series x=0.5*x(-1)+nrnd
smpl @first @last
---click run.
Go to quick estimate equation and type x ar(1) also view-graph –line-ok
 Chap1: Time Series Concepts

1.6 Moving average processes

-First-Order Moving Average process - MA(1):
Generating a Moving Average process in Eviews:
E.g. Yt = 0.3εt +0.4εt-1, where εt ~ NIID(0,1)
The command is:
smpl @first @last
series e=nrnd
smpl @first @first
series y=nrnd
smpl @first+1 @last
series y=0.3*e+0.4*e(-1)
smpl @first @last
Go to quick estimate equation type y e e(-1)
 Chap1: Time Series Concepts

1.7 Autoregressive & Moving Average (ARMA) processes

One autoregressive and one moving average term -
ARMA(1, 1) can be written as:
1.8 Differencing
To difference a variable is to subtract a specific lag from
it. For example, Yt = Yt-1 + εt in its differenced form is
1.9 Random walk
Consider the price of a stock traded on the Johannesburg Stock Exchange (JSE),
Pt. If you do not know what tomorrow’s price (Ptomorrow) is, then your best
guess is today’s price (Ptoday).

Of course, to assume this is unrealistic – trading in the stockmarket results in price

movements from day to day. To capture these daily price movements, one can
introduce a white noise term and such a process is called a random walk.
1.9 Random walk(con’t)
This random walk is also known as the drunkard’s walk. A
drunkard’s path from the bar to his home is erratic and
unpredictable, as is the path of a random walk series over time (yet
there is a definite destination).
The statistical properties of a random walk time series are:
- A constant mean over time,
- Changing variance over time,
- Changing auto covariance over time.

Therefore, a random walk is considered a non-stationary process.

 Stationary process
Broadly speaking, 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
Mean: E(Yt) = μ
Variance: var (Yt) = E(Yt − μ)2 = σ2
Covariance: γk = E[(Yt − μ)(Yt+k − μ)]

If a time series is not stationary in the sense just defined, it is called a

nonstationary time series. In other words, a nonstationary time series will
have a time varying mean or a time-varying variance or both.
 TESTS OF STATIONARITY

In practice we face two important questions:

(1) How do we find out if a given time series is stationary?

(2) If we find that a given time series is not stationary, is there

a way that it can be made stationary?
 Data to be used
Notes:
GDP (Gross Domestic Product), billions of 1987 dollars, p. A-
96.
PDI (Personal disposable income), billions of 1987 dollars,
PCE (Personal consumption expenditure), billions of 1987
dollars,
Profits (corporate profits after tax), billions of dollars,
Dividends (net corporate dividend payments), billions of
dollars, p. A-110.
Source: U.S. Department of Commerce, Bureau of Economic
Analysis, Business Statistics, 1963–1991, June 1992.
 1. Graphical Analysis

before one pursues formal tests, it is always advisable to plot the time
series under study, Such an intuitive feel is the starting point of more
formal tests of stationarity.
 2. Autocorrelation Function (ACF) and Correlogram

One simple test of stationarity is based on the so-called autocorrelation

function (ACF). The ACF at lag k, denoted by ρk, is defined as

ρk = γk/γ0 = covariance at lag k/variance

ACF and PACF
 ACF and PACF
The solid vertical line in this diagram represents the zero axis;
observations above the line are positive values and those
below the line are negative values. As is very clear from this
diagram, for a purely white noise process the autocorrelations
at various lags hover around zero. This is the picture of a
correlogram of a stationary time series. Thus, if the
correlogram of an actual (economic) time series resembles the
correlogram of a white noise time series, we can say that time
series is probably stationary.
 GDP of USA
Now let us take a concrete economic example. Let us examine the
correlogram of the GDP time series. The correlogram up to 25 lags is
shown in Figure below. The GDP correlogram up to 25 lags also shows a
pattern similar to the correlogram of the random walk model.
The autocorrelation coefficient starts at a very high value at lag 1 (0.969)
and declines very slowly. Thus it seems that the GDP time series is
nonstationary.
If you plot the correlograms of the other U.S. economic time series
shown in Figures above, you will also see a similar pattern, leading to the
conclusion that all these time series are nonstationary; they may be
nonstationary in mean or variance or both.
Two practical questions may be posed here.
First, how do we choose the lag length to compute the
ACF?
Second, how do you decide whether a correlation
coefficient at a certain lag is statistically significant?
The answer follows.
 The Choice of Lag Length

This is basically an empirical question. A rule of thumb is to

compute ACF up to one-third to one-quarter the length of the
time series. Since for our economic data we have 88 quarterly
observations, by this rule lags of 22 to 29 quarters will do. The
best practical advice is to start with sufficiently large lags and
then reduce them by some statistical criterion, such as the
Akaike or Schwarz information criterion
 Statistical Significance of Autocorrelation
Coefficients

Consider, for instance, the correlogram of the GDP time series given in
. How do we decide whether the correlation coefficient of 0.638
at lag 10 (quarters) is statistically significant? The statistical significance of any
ρˆk can be judged by its standard error.
in large samples the sample autocorrelation coefficients are normally
distributed with zero mean and variance equal to one over the sample
size. Since we have 88 observations, the variance is 1/88 = 0.01136 and the
standard error is √0.01136 = 0.1066. Then following the properties of the
standard normal distribution, the 95% confidence interval for any (population)
ρk is:
ρˆk ± 1.96(0.1066)
In other words,
Prob (ρˆk − 0.2089 ≤ ρk ≤ ˆρk + 0.2089) = 0.95
If the interval includes the value of zero, we do not reject the
hypothesis that the true ρk is zero, but if this interval does not
include 0, we reject the hypothesis that the true ρk is zero.

Applying this to the estimated value of ρˆ10 = 0.638, the reader can
verify that the 95% confidence interval for true ρ10 is (0.638 ±
0.2089) or (0.4291, 0.8469).
Obviously, this interval does not include the value of zero,
suggesting that we are 95% confident that the true ρ10 is significantly
different from zero.
As you can check, even at lag 20 the estimated ρ20 is statistically
significant at the 5% level.
 Q statistic

Instead of testing the statistical significance of any individual

autocorrelation coefficient, we can test the joint hypothesis
that all the ρk up to certain lags are simultaneously equal to
zero. This can be done by using the
Q statistic developed by Box and Pierce, which is defined as

where n = sample size and m = lag length. The Q statistic is

often used as a test of whether a time series is white noise
In large samples, it is approximately
distributed as the chi-square distribution with m df. In an
application, if the computed Q exceeds the critical Q value
from the chi-square distribution at the chosen level of
significance, one can reject the null hypothesis that all the
(true) ρk are zero; at least some of them must be nonzero.
 Ljung–Box (LB) statistic
A variant of the Box–Pierce Q statistic is the Ljung–Box
(LB) statistic,
which is defined as

Although in large samples both Q and LB statistics follow

the chi-square distribution with m df, the LB statistic has
been found to have better (more powerful, in the statistical
sense) small-sample properties than the Q statistic.
If Q or LB are lower than chi-square, then the null hypothesis cannot
be rejected .
Examples:
Suppose that a researcher had estimated the first five autocorrelation
coefficients using a series of length 100 observations, and found
them toLagbe. 1 2 3 4 5

Autocorrelation 0.207 -0.013 0.086 0.005 -0.022

coefficient

Source: Chris Brooks, 2008, page 210

Test each of the individual correlation coefficients for significance, and test
all five jointly using the Box--Pierce and Ljung--Box tests.

Solution
A 95% confidence interval can be constructed for each coefficient using the
Confidence Interval: CI=±1.96 × 1 √T=(−0.196, +0.196), because T = 100.
The null hypothesis that a given coefficient is zero is rejected if the
coefficient lies outside the confidence interval.
Using autocorrelation coefficient in the table above, the null hypothesis is
rejected at the 5% level only for the first autocorrelation coefficient is
significantly different from zero, because other autocorrelation coefficients
lies inside the confidence interval.
 Exercise 2
Let us consider the consumer price index (P) in Rwanda and use EVIEWS
to test significance of autocorrelation coefficients.
 ans
A 95% confidence interval is this case is CI=±1.96 × 1/ √T=(−0.26, +0.26),
because T = 56.

Lag 1 2 3 4 5
Autocorre 0.953 0.901 0.849 0.799 0.748
lation
coefficient

The null hypothesis that a given coefficient is zero is rejected if the coefficient
lies outside the confidence interval. Using autocorrelation coefficient in the
table above, the null hypothesis is rejected at the 5% level for all
autocorrelation coefficients. In the first part of the correlogram of the
consumer Price Index (P), we can see that each autocorrelation coefficient lies
out the confidence interval.
Returning to the GDP example given in Figure , the value of the LB
statistic up to lag 25 is about 891.25. The probability of obtaining such an
LB value under the null hypothesis that the sum of 25 squared estimated
autocorrelation coefficients is zero is practically zero, as the last column of
that figures shows. Therefore, the conclusion is that the GDP time series is
nonstationary, therefore reinforcing that the GDP series may be nonstationary.

In exercise you are asked to confirm that the other four U.S. economic time
series are also nonstationary.
 THE UNIT ROOT TEST
A test of stationarity (or non stationarity) that has become widely
popular over the past several years is the unit root test. We will
first explain it, then illustrate it and then consider some limitations
of this test.
We start with Yt = ρYt−1 + ut −1≤ρ≤1
where ut is a white noise error term.
Deduct both side by Yt−1 ,this will be
Yt − Yt−1 = ρYt−1 − Yt−1 + ut = (ρ − 1)Yt−1 + ut
which can be alternatively written as:
where δ = (ρ − 1) and
d is the first-difference operator
We test the (null) hypothesis that δ = 0. If δ = 0, then ρ = 1, that is we
have a unit root, meaning the time series under consideration is
nonstationary.
eq1)
eq2)
eq3)

where t is the time or trend variable.

In each case, the null hypothesis is that δ = 0(or ρ = 1); that is,
there is a unit root—the time series is nonstationary. The
alternative hypothesis is that δ is less than zero(ρ < 1); that is, the
time series is stationary.
If the null hypothesis is rejected, it means that Yt is a stationary
time series with zero mean in the case of eq1
that Yt is stationary with a nonzero mean [= β1/(1 − ρ)] in the
case of eq2 and that Yt is stationary around a deterministic trend
in eq3.
 1.Identification in Eviews
Open the data—view—graph—line
View—correlogram—
View---unit root—level
When
p< α(0.05): we reject the null hypothesis

For ADF(Augmented Dickey Fuller) and PP(Phillips Perron) the H0 is that the series
have a unit root(not stationary)
while
for KPSS(Kwiatkowski-Phillips-Schmidt-Shin) the H0 is that the series is stationary

Before differencing we have to use level

we show the correlogram and partial correlogram of theGDP series. From
this figure, two facts stand out:
First, the ACF declines very slowly; as shown in Figure , ACF up to 23 lags
are individually statistically significantly different from zero, for they all are
outside the 95%confidence bounds.
Second, after the first lag, the PACF drops dramatically, and all PACFs after
lag 1 are statistically insignificant

After first difference , Now we have a much different pattern of ACF and
PACF. The ACFs at lags 1, 8, and 12 seem statistically different from zero
One way of accomplishing this is to consider the ACF and PACF and the
associated correlograms of a selected number of ARMA processes, such as
AR(1), AR(2), MA(1), MA(2),ARMA(1, 1), ARIMA(2, 2), and so on. Since
each of these stochastic processes exhibits typical patterns of ACF and
PACF, if the time series under study fits one of these patterns we can
identify the time series with that process. Of course, we will have to apply
diagnostic tests to find out if the chosen ARMA model is reasonably
accurate.
 FORECASTING

The values of δ, α1, α8, and α12 are already known from the estimated
regression . Therefore,we can easily obtain the forecast value of Y1992−I. The
numerical estimate
of this forecast value is
 2.ESTIMATION OF THE ARIMA MODEL in
Eviews
The partial autocorrelations with spikes at lag 1, 8, and 12 seem statistically
significant but the rest are not; if the partial correlation coefficient were significant
only at lag 1, we could have identified this as an AR(1) model. Let us therefore
assume that the process that generated the (first-differenced) GDP is at the most an
AR(12) process. Of course, we do not have to include all the AR terms up to 12,
for from the partial correlogram we know that only the AR terms at lag 1, 8, and12
are significant
 Let Y*t denote the first differences of U.S. GDP. Then our tentatively
identified AR model is

go to Quick ---estimate equation

d(gdp) c ar(1) ar(8) ar(12)

 4.FORECASTING in Eviews
Remember that the GDP data are for the period 1970–I to 1991–IV. Suppose,on
the basis of model , we want to forecast GDP for the first four quarters of 1992.
But in the dependent variable is change in the GDP over
the previous quarter. Therefore, if we use the model, what we can obtain are the
forecasts of GDP changes between the first quarter of 1992 and the fourth quarter
of 1991, second quarter of 1992 over the first quarter of 1992, etc.
To obtain the forecast of GDP level rather than its changes, we can
“undo” the first-difference transformation that we had used to obtain the
changes. (More technically, we integrate the first-differenced series.) Thus,
to obtain the forecast value of GDP (not GDP) for 1992–I, we rewrite model as
 3.DIAGNOSTIC CHECKING in Eviews

How do we know that the model is a reasonable fit to the data?

One simple diagnostic is to obtain residuals from and obtain the ACF
and PACF of these residuals, say, up to lag 25.
The estimated AC and PACF are shown in Figure As this figure shows,
none of the autocorrelations and partial autocorrelations is individually
statistically significant.
In otherwords, the correlograms of both autocorrelation and partial
autocorrelationgive the impression that the residuals estimated from are
purely random.
Hence, there may not be any need to look for another ARIMA model
Thus the forecast value of GDP for 1992–I is about $4877 billion (1987
dollars).Incidentally, the actual value of real GDP for 1992–I was $4873.7
billion;
the forecast error was an overestimate of $3 billion.Note that if you were to
use (22.5.2) to compute the forecast change of
GDP from 1991–IV to 1992–I, you would obtain this figure as −$4.25
billion.
 Chap 2: Econometric Modelling: Model
specification and diagnostic testing

The 9th assumption of the Classical Linear Regression Model

(CLRM) specifies that the regression model used in the regression
analysis is ‘correctly’ specified’. If this assumption is violated we
encounter the problem of model specification error or model
specification bias.
 Examine the following questions

- How do we find the ‘correct’ model? In other words, what are the criteria in
choosing a model for empirical analysis?

- What types of model specification errors will we encounter in practice?

- What are the consequences of specification errors?

- How do we detect specification errors? In other words, what are some of the
diagnostic tools that one can use?

- What are the remedies of the specification errors once detected? What are the
benefits?

- How do we evaluate the performance of competing models?

 2. 1 Model Selection criteria

According to Hendry and Richard (1983), a model chosen for empirical analysis
should satisfy the following criteria:
- Data admissibility
- Be consistent with theory; that is, it must make economic sense.
- Have weakly exogenous explanatory variables/ regressors; that is, the regressors
must be uncorrelated with the error term.
- The model must exhibit parameter constancy; that is, the value of parameters
should be stable. In the absence of parameter constancy, predictions will not be
reliable.
- The model must exhibit data coherency; that is, the residual estimated from the
model must be purely random (technically, white noise).
-The model must include all rival models in the sense that it is capable of explaining
their results.
In short, other models cannot be an improvement over the chosen model
2.2 Types of specification errors
By summary, in developing an empirical model, one is likely to
commit one or more of the following specification errors:
1. Omission of relevant variable(s).
2. Inclusion of unnecessary variable(s)
3. Adopting the wrong functional form.
4. Errors of measurement
5. Incorrect specification of the stochastic error term.
It is fruitful to distinguish between the model specification errors and
model mis-specification errors.
The first four types of error discussed above are essentially in the nature
of the model specification errors in that we have in mind a “true” model
but somehow we do not estimate the correct model. The
last type of error is in the nature of mis-specification errors.
 2.3 Consequences of model specification errors

2.3.1 Underfitting a model (Omitting relevant variable(s))

The consequences of omitting variable X3 model are as follows:

2.3.2 Overfitting a model (Including irrelevant variable(s))

The following apply on this specification error:

 2.4 Tests of specification errors

2.4.1 Detecting the presence of unnecessary variables

(Overfitting a Model)
 2.4 Tests of specification errors(con’t)

2.4.2 Test for omitted variables and incorrect functional form

In determining model adequacy, we look at some broad features

of the results such as:
-The R^2 value;
-The estimated t ratios;
- The signs of estimated coefficients in relation to their prior
expectation;
- The Durbin-Watson statistic…
If these diagnostics are reasonably good, we proclaim that the chosen
model is fair representation of reality.

If the results do not look encouraging, then we begin to worry about

model adequacy and look for remedies.

To aid us in determining whether model inadequacy is on account of one

or more of these problems, we can use some of the following methods.
- Examination of the residuals;
- Ramsey’s RESET Test;
- Langrage Multiplier (LM) Test for adding variables.
 Chap. 3: Unit roots, Cointegration and Error
correction model
3.1 Stochastic processes
3.1.1 Stationary stochastic processes
A type of stochastic process that has received a great deal of attention and
scrutiny by time series analysts is the so-called stationary stochastic process.
Broadly speaking, 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. In
time series literature, such a stochastic process is known as a weakly
stationary process.
3.1.2 Non stationary stochastic process
In time series models in econometrics, a linear stochastic
process has a unit root if 1 is a root of the process's characteristic
equation.
The process will be non-stationary with the following properties:
Keeping in mind that the terms non stationarity, random walk,

and unit root can be treated as synonymous, we distinguish three

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) and
random walk with drift and

(3)deterministic trend (i.e. constant term and trend are present).

3.1.3 Unit root stochastic process
Let us write the random walk model (RWM) as:
In practice, then, it is important to find out if a time series
posses a unit root.

In this course we will discuss the most famous tests, namely

the Augmented Dickey-Fuller test(ADF) and the Phillips-
Perron test(PP). Both these tests use the existence of a unit root
as the null hypothesis.
 3.2 The unit root test

A unit root test tests whether a time series variable is non-stationary

using an autoregressive model. Why unit roots testing?
- Much conventional asymptotic theory for least squares assume
stationarity of the explanatory variables;
- Not all economic time series are stationary;
- In many cases economic theory posit causal relationships between
economic series that are non-stationary. E.g. aggregate consumption,
national income, exchange rates etc;
- If variables are non-stationary we may have non-sense regressions
named so by Yule (1926) or spurious regression in the terminology
of Granger and Newbold (1974).
3.2.1 The Dickey-Fuller tests
In statistics, the Dickey-Fuller test tests whether a unit root is
present in an autoregressive model AR(1). It is named after the
statisticians D. A. Dickey and W. A. Fuller, who developed the
test in the 1970s.

Consider the simple AR(1) model

3.2.2 Phillips Perron (PP): a non-parametric test of
unit root

While the DF-procedure aims to retain the validity of the tests

based on white-noise errors in the regression model by ensuring
that the errors are indeed white noise,
the PP acts instead to modify the statistics after the estimation in
order to take into account the effect that autocorrelated errors
will have on the results.
3.2.3 Order of integration
What does this mean? A time series is said to be integrated of
order d, written I(d), if after being difference d times it becomes
stationary.
- If a variable is stationary, that variable is said to be integrated
of order zero, written as I(0).
- If the first difference of a non-stationary variable is stationary,
that variable is said to be integrated of order one, written as I(1).
- If second differences are required to achieve stationarity, then
the variable is integrated of order two, written as I(2).
For example, suppose a stock price is 5 on Monday, 6 on
Tuesday, 7 on Wednesday, and 8 again on Thursday.

One differences that series by turning it into a series of

daily price increments. In this case, if we difference just
once we get 1 ... 1 ...1. We obtained a stationary series by
differencing it just once, which means our original series
is integrated of order one and thus difference stationary.
Why are we concerned with the order of integration
of a series?
A direct bearing on the appropriateness and statistical validity of
regression results is of utmost importance. For example if we
wish to regress t Y on t X , when
- Yt and X t are stationary, classical OLS is valid.
- Yt and Xt are integrated of the different orders, regression is
meaningless.
- Yt and Xt are integrated of the same order and residuals are
nonstationary, regression may be spurious.
- Yt and Xt are integrated of the same order and residuals
are stationary, regression may indicate a cointegrating relationship
According to Granger and Newbold, an R2 > Durbin-Watson
statistics (d) is a good rule of thumb to suspect that the
estimated regression is spurious.
Exercises
having the root z = 1, which lies on, not outside, the unit circle.
In fact,
the particular AR(p) model given by (5.58) is a non-stationary
process known as a random walk
 Example 3
is the following process for yt stationary?
Thus the expected or mean value of an autoregressive process of
order one is given by the intercept parameter divided by one minus
the autoregressive coefficient.
(iii) Turning now to the calculation of the autocorrelation
function, the autocovariances must first be calculated. This is
achieved by following similar algebraic manipulations as for
the variance above, starting with the definition of the
autocovariances for a random variable. The
autocovariances for lags 1, 2, 3, . . . , s, will be denoted by γ1,
γ2, γ3, . . . , γs , as previously.
 Exercise 5

5. You have estimated the following ARMA(1,1) model for

some time series data
yt = 0.036 + 0.69yt−1 + 0.42ut−1 + ut
Suppose that you have data for time to t−1, i.e. you know that
yt−1 = 3.4, and ˆut−1 = −1.3
Obtain forecasts for the series y for times t, t +1, and t +2
using
the estimated ARMA model
Exercise 6
 3.2 Cointegration
3.2.1 The concept of cointegration
Cointegration is an econometric property of time series
variables. If two or more series are themselves non-stationary,
but a linear combination of them is stationary, then the series are
said to be cointegrated.
It is often said that cointegration is a means for correctly testing
hypotheses concerning the relationship between two variables
having unit roots.
Time series Xt and Yt are said to be cointegrated of order d, b,
denoted by
3.2.2 The econometrician’s rationale behind cointegration
Why should we be concerned with cointegration?
The answer is that cointegration is a possible solution to the problem
of nonstationarity found in many economic time series.
Recall if we wish to regress Yt on Xt , when
- Yt and Xt are stationary, the application of Ordinary Least
Squares Estimation is statistically acceptable;
- Yt and Xt are non-stationary, the assumptions upon which
Ordinary Least Squares Estimation rest are violated,
rendering its application inappropriate.
3.2.3 Precondition for cointegration
Given the following linear relationship

Consequently, in the relationship between two variables, both

must be integrated of the same order (and generally, I(1)) for the
residuals to be I(0).
If the number of variables involved in the relationship increases,
the problem becomes more complicated but it is still possible that
variables of different order of cointegration can be cointegrated.
By conclusion, the cointegration technique is
extended beyond two variables, and occasionally to
variables integrated at different orders.
3.2.4 Testing for cointegration

There are two most popular approaches to testing for

cointegration; the Engle-Granger (EG) two-step method and
the Johansen procedure.
The first is an analysis of the stationarity of the residuals
from the levels regression (the long-run regression), and it is
that approach that we pursue here.
3.2.4.1 Engle-Granger (EG) or Augmented Engle-Granger
(AEG) test
We already know how to apply the DF or ADF unit root tests.
All we have to do is estimate a regression, obtain the residuals,
and use the ADF tests. We also know that the residuals will
have a zero mean and no trend by construction. Rather we can
proceed directly to the ADF test without a constant or a trend
(noted “None” in Eviews) or use the DW-based test (Mukherjee
et al., 1998: 399)7.
 The Engle–Granger 2-step method
This is a single equation technique, which is conducted as follows:
Step 1
Make sure that all the individual variables are I(1). Then estimate
the cointegrating regression using OLS. Note that it is not possible
to perform any inferences on the coefficient estimates in this
regression -- all that can be done is to estimate the parameter
values. Save the residuals of the cointegrating regression, ˆut .
Test these residuals to ensure that they are I(0). If they are I(0),
proceed to
Step 2; if they are I(1), estimate a model containing only first
differences
Practical part
Cointegration in Eviews
Application
Step2:lag length criteria
Step3:cointegration test
Cointegration specification
Step4 :Run VEC model in Eviews
VEC output
P-value
 Coefficient diagnostics
View-------Coeff Diagnostics-----walid test

For those coeficients with p-value greater than 0.05,

Ho: c(x)=c(n)=0
 Residual diagnostics
Ho: There is no serial autocorrelation

View---residual diagnostic---serial LM test---

lag(that we have been using along the example)

When there is no serial autocorrelation that means

we are happy with the outcome(model looks good)
 Stability diagnostic
View---stability diagnostic---recursive estimates—
CUSUM Test.

The model is said to be dynamically stable, when

the blue lies in the interval(we are happy with this
model)
 Chap4: Simultaneous Equation
Previously we considered single equation models, with a single
dependent variable and one or more explanatory variables. The
cause-and-effect relationship ran from X to Y.
In many situations, the one-way cause and effect relationship is not
meaningful. This occurs if Y is determined by X and some of the X’s
are determined by Y. In these models there is more than one equation
where each of the dependent variables is endogenous. That is, one
may not estimate the parameter of a single equation without taking
into account information provided by other equations in the system.
Consider the following system:
 4.1 Examples of simultaneous equation models

4.1.1 Demand and supply model

The price (P) of a commodity (Q) and the quantity sold are determined
by the intersection of the demand and supply curves for the
commodity.
 4.1 Examples of simultaneous equation models

4.1.2 Keynesian model of income determination

 4.1 Examples of simultaneous equation models

4.1.3 The IS model of Macroeconomics (Good

market equilibrium)
 4.1 Examples of simultaneous equation models

4.2.4 The LM model of Macroeconomics (Money market

equilibrium)
 4.2 The identification problem
4.2.1 Notation and definitions
4.2.2 The identification problem
By the identification problem we mean whether numerical estimates of
the parameters of the structural equation can be obtained from the
estimated reduced-form coefficients. This can be done if the particular
equation is identified. It cannot be done when the equation under
consideration is underidentified or unidentified.
An identified equation may be either exactly (fully or just) identified or
overidentified
Exactly identified – if a unique numerical value of the structural
parameters can be obtained.
Overidentified – if more than one numerical value of
the structural parameters can be obtained.
There is underidentification when the model contains a
number of structural coefficients that is greater than the
number of equations.
For example, there is no way of obtaining four structural
unknowns from only two reduced-form coefficients.
4.2.3 Rules for identification
If the exclusion of some of the variables (coefficient = 0) is
the only restriction on the structural coefficients, the
necessary condition of identifiability is as follow:
 4.3 Simultaneous equation methods

In practice single equation methods are used:

• Ordinary least squares;
• Indirect least squares;
• Two stage least squares.
4.3.1 Recursive models and ordinary least squares
We saw that because of interdependence between the stochastic
disturbance term and the endogenous explanatory variables OLS is
inappropriate for the estimation of an equation in a system of
simultaneous equations. Estimators are biased and inconsistent.
The alternative name triangular stems from the fact that if we
form the matrix of coefficients of the endogenous variables we
obtain the following matrix:
4.3.2 Estimation of a just identified equation: The method of
Indirect least squares (ILS)

For just or exactly identified structural equations the

method of obtaining estimates of the structural
coefficients from the OLS estimates of the reduced-
form coefficient is known as the method of indirect
least squares and the estimates obtained are known as
indirect least squares estimates.
ILS involves the following three steps:
 ILS
Step 1: We first obtain the reduced-form equations.
Step 2: We apply OLS to the reduced-form equations
individually. This is allowed since the explanatory variables are
predetermined and uncorrelated with the stochastic disturbances.
Therefore, we obtain consistent estimates.
Step 3: We obtain estimates of the original structural coefficients
from the estimated reduced-form coefficients. When an equation
is exactly identified, there is a one-to-one correspondence
between the structural and reduced-form coefficients.
4.3.3 Estimation of an overidentified equation: The method of
two-stage least squares (2SLS)

Stage 1: Regress each endogenous variable on all the

predetermined variables in the whole system, not just
the equation

Stage 2: Replace the endogenous variables appearing

on the right hand side in the original equations by
their estimated values and run the OLS regressions.
A practical example
The data in the following table are the deviations from the mean
(the mean of a series is subtracted from every observation).

Date( Questions:
t) Y1t Y2t X1t Y1t-1
1. Determine the identifiability
1992 -30 -14 -26 20
1993 -4 10 5 -30 conditions of the model.
1994 -19 -19 -36 -4 2. Estimate the equations using OLS
1995 -6 -11 -6 -19 and show how the estimates are not
1996 -9 6 -13 -6 Best Linear Unbiased Estimators
1997 11 -12 25 -9 (BLUE).
1998 9 10 5 11 3. Estimate the coefficients using the
1999 19 11 32 9
2000 29 19 14 19
2SLS.
 answer
1. The identifiability conditions of the
model
 2. Estimation using OLS

a. Equation 1:
Go to quick –estimate equation –y1t=c(1)*(y2t+x1t)

Y1t = 0.5166666667*(Y2t+X1t)
b. Equation 2:

Y2t = 0.518739709*Y1t + 0.04025986485*Y1t_1

Conclusion:
aˆ = 0.516,bˆ =0.518,
cˆ = 0 ,It is not significantly different from zero.
c. Are the above estimates BLUE?

To show that the estimates are not BLUE we have to

prove that exogenous variable in a give equation is
correlated with the error term. That is, we perform the
reduced form equations:
 3. Estimation using the 2SLS

Stage 1: Regress each endogenous variable on all the

predetermined variables in the whole system (X1 and Yt-1), not just
the equation.
y2t=c(1)*x1t+c(2)*y1t_1
Stage 2: Replace the endogenous variables appearing on the
right hand side in the original equations by their estimated
values and run the OLS regressions.
Go to quick –generate series--
y2estimated=0.355358*x1t+0.139683*y1t_1
From the above output
Then go to quick ---estimate equation
y1=c(1)*(y2estimated+x1t)
God BLESS YOU ALL!!!!!

I am because you are.