UNIVERSITY OF ABOMEY-CALAVI (UAC)

FACULTY OF ECONOMIC SCIENCES AND MANAGEMENT (FASEG)

PROGRAM: LICENSE IN ECONOMETRICS AND APPLIED STATISTICS (LESA)

YEAR OF STUDY: L3

COURSE: INTRODUCTION TO ADVANCED EVIEWS

In charge: Dr. ALAKONON B. Calixe

Mail:calixealakonon@[Link]

Cell: (+229) 67127207

ACADEMIC YEAR: 2022-2023

1
 Warning

This course material is in provisional version and should not be distributed without permission from
the author. Comments, suggestions, and remarks are welcome.

2
 Summary

Chapter 1: Getting Started with the Software

▪ Usage
▪ Creation
▪ Management
Chapter 2: Statistical Analysis
▪ Graphical representations
▪ Descriptive statistics
Chapter 3: Econometrics
▪ Estimations
▪ Tests
▪ Method

3
 1. Course Objectives
Econometrics refers to a set of statistical methods aimed at quantifying
economic phenomena. It is a valuable tool for analysis and assistance for its users.
in decision-making. This course material offers training in the practice of econometrics
on Eviews.
❖ General objective
This course aims to enable the learner to master the practice of econometrics.
❖ Specific objectives
At the end of the course, the learner should be able to:
To perform descriptive analyses on Eviews
To perform multiple linear regressions on Eviews
To carry out the different tests related to general linear model estimates
Etc...
2. Course Content
The course consists of an introduction and successive parts for the completion of a
estimation on Eviews.
3. Evaluation method
Two assessment methods are retained in the context of this course. This is the assessment
formative (continuous assessment) and summative evaluation (end-of-semester exam).
Formative assessment can also take the form of presentations or homework to be submitted.
on copy.
4. Teaching Method
The course will be projected. Some parts of the course will be subject to individual activities.
then in a small group.
5. Educational tools
It consists of the current course materials, board, chalk or markers, projector, internet,
etc.
6. Bibliographic References
Doucoure, F., (2005), "Econometric Methods: Lectures and Practical Work", University
Cheikh Anta Diop, Dakar.
Greene, W. (2000), "Econometric Analysis", Prentice Hall, 4th edition.
Hamilton, J., (1994), "Time Series Analysis", Princeton University Press.
Lardic, S., and Mignon, V., (2002), "Econometrics of macroeconomic time series and
financial" , Economica.

4
Régis Bourbonnais (2011), 'Econometrics: manual and solved exercises', Dunod, 8th ed.

Introduction

This introductory course in econometrics aims to provide students with

license 3/Economics of FASEG-UAC, the minimum knowledge required to perform
applied work in economics in which they may be required to use the methods
econometric (forecasting studies, applied microeconometrics work, modeling
macroeconomic;...)

In this course, the software Eviews 9 is used for the various applications proposed here.
The goal is to show how Eviews meets the needs of economic analysis.
It is also important to have prerequisites in econometric analysis and time series analysis.
temporal. Otherwise, the reading of the appropriate literature will be strong about it.

The developments mainly focus on the presentation of the Eviews software. It will be about
give students the basic elements to use this software. Then, will intervene
the analysis of economic series, notably statistical analysis through representations
graphics and descriptive statistics. Also the study of the general linear model. This
will allow us to present the estimation by the OLS and the various tests associated with the model
general linear.

5
Chapter 1: Introduction to Eviews

Eviews is a powerful software that is particularly useful for:

▪ Econometrics: time series, panel data; etc.
▪ Data handling: correction of seasonal variations; etc
▪ The forecast including the use of the Box-Jenkins method
This section aims to introduce the 'Minimum' on Eviews.
It will be about presenting the general environment of Eviews, knowing how to create a
workspace and import/export data. We will outline the process
importing the Excel file. Of course, Eviews can read other types of files in
other formats
1. General Environment of Eviews
▪ Main Menu: this bar is always present and accessible
▪ Command line: you can directly enter a command

6
 2. Create or open a work file (WORKFILE)
Having launched Eviews, you need to create your workspace.

The first step is to create a workspace (WORKFILE) by

workfile>, that is to say a welcoming space for statistical series (monthly,
semi-annual, annual, without dates ...) and the calculation results. In the proposed window, it
convenient, either to choose the periodicity of the data (the start and end date for the series
temporal), or to indicate the number of observations (UNDATED series - for the series in
instantaneous cut)
Note: A WORKFILE can only contain series of the same periodicity.

After this instruction, obtain the following:

7
The creation of your workspace depends on the type of econometric analysis you want.
perform: Cross-sectional data econometrics, time series econometrics and
the econometrics of panel data.
You must therefore choose one of these options in 'Workfile structure type'
▪ For time series econometrics, you choose 'Dated-regular-frequency'
In this case, specify the nature of the data (annual, quarterly, etc.)
you will then specify the start and end dates.
▪ For the econometrics of cross-sectional data, you choose
Unstructured/Undated. In this case, you just specify the number of observations.
▪ For panel data econometrics, you choose 'Balanced Panel'; you
fill in the corresponding fields then.
Let's consider the three series summarized in Table 1 below from 1972 to 2004 except for GDP which
ends in 2001. In this table "Invest, GDP and Interest" respectively refer to
Private investment, Gross Domestic Product and the interest rate of the interbank market
BENI country.

Table 1: Series on a country BENI

Date Invest GDP Tinteret
1972 158 1090 3.5
1973 202 1255 5.5
1974 275 1414 5.5
1975 289 1906 8
1976 264 1933 8
1977 287 1979 8
1978 316 2209 8
1979 314 2751 8
1980 350 2987 10.5
1981 317 2479 10.5
1982 300 2583 12.5
1983 270 2480 10.5
1984 429 2337 10.5
1985 574 2579 10.5
1986 633 3763 8.5
1987 458 4600 8.5
1988 787 4980 9.5
1989 708 4626 11
1990 894 5698 11
1991 765 5500 11
1992 676 6027 12.5
1993 748 5431 10.5
1994 859 3642 10
1995 789 4476 7.5
1996 866 4651 6.5
1997 905 4387 6.5

8
 1998 867 4646 6.25
1999 925 4752 5.75
2000 4371 6.5
2001 4620 6.5
2002 4625 6.75
2003 4635 6.85
2004 4650 7.5

Since this involves time series, we can choose for the creation of our workfile.
Dated-regular-frequency. We filled in the fields as indicated in the diagram.
underneath.

Specify here the frequency of data observations: annual, semi-annual, monthly.

daily etc. Here the data is annual, it starts in 1972 and ends in 2004.
Clicking OK gives the following result:

9
This Workfile does not yet contain the variables. We will need to bring the data into the
software. Several possibilities exist to do it. We choose here to do
first entering the data into an Excel file and then importing said file
in Eviews.

3. Data imports

The data to be imported is located in Excel on your computer (the data from table 1).
The steps for importing are as follows. In the main menu, do:
File/Import/Import from file

You then select the Excel file on your computer. Once found, you click
on 'open' and it displays the following:

10
You then click on 'Next', 'Next' and finally 'Finish'. The data appears
now in your Workfile as follows:

You have just created the working file. The three series appear in the file. Save
this file named 'Econometric Practice FASEG' via the following instruction: File/Save As.
To view the observations of a series, double click on the series. To view the
four series, select them and click 'Open' then 'as group' and finally confirm. Or click
on 'View' then 'Show'; enter the names of the variables you want to open together
then validate.

11
To modify/edit the data of one or more series
Select the series or series then double click then <edit +/–>.
If the series is in graphical representation or on other functions (correlation, statistics,
etc.) select <View> then <Spreadsheat> then <edit +/–>

12
Chapter 2: Descriptive Analysis
2.1. Graphical representation of the series
In the database toolbar, click on <View> then <Graph>.
To modify the chart options, double click on the chart area and do
Graph option.
Open the investment and GDP series together, then follow the next instruction after opening.
the graphical representation: View/Graph/

❖ To add text to the graph

Click AddText
Enter
Bar graphs of the variables INV and GDP
Justification: Choose Center
Position: Choose Top (of choice)
Choose Text in Box
Click on Font
Font: Choose Arial
Font style: Choose Bold
Size: Choose 18
OK

You click on 'OK'. The two series are represented in the same graph as
indicated below:

13
The two series each show an upward trend over time.
❖ Create a multi-series graph
To create a series group
Open the group and then go to View/Graph... then OK
If you want to get multiple charts in one window, click on Options, then in
Multiple Series, click on Multiple Graphs
To change the Template, also click on Options (or double-click on the graph), then
Template & Objects then let yourself be guided
❖ Presentation of the graph
▪ Possibilities: add text, modify the scale of the axes, change the appearance
(template), save your own template in order to reuse it, etc...
▪ These aesthetic enhancements are very practical when the graphs need to respond to
the same graphic charter for example.
▪ Let's put these possibilities into practice through several examples.

2.2. To generate a series

To generate the series of logarithms associated with the three series. In the main menu, do
Quick/Generate Series, then type in the calculation formula or enter in the line of
order (at the top right)

❖ Clouds of points

Let's visualize the two series Invest and GDP and then go to View/Graph/Scatter.

14
The scatter plot suggests a linear fit, with a more or less strong slope. There is
therefore a correlation between Investment and the GDP of the country BENI. We expect then that
the coefficient associated with these variables is significantly different from zero.

❖ Scatter plot of the inv and pib variables and plotting the linear regression line

Quick → Graph → Enter GDP → OK

Specific: Choose Scatter
Fitlines: Choose Regression line → OK

900

800

700

600

500

400

300

200

100
1,000 2,000 3,000 4,000 5,000 6,000 7,000

gdp

15
2.3. Basic Statistics
❖ Click on a series, then:
▪ View/Descriptive Stats & Tests/Histogram and Stats
▪ View/Descriptive Stats & Tests/StatsTable

❖ For a Series Group:

▪ Quick/Group Statistics/Descriptive Statistics
▪ Missing observations are automatically excluded from the statistics.

16
❖ Histogram

Click on Quick → Graph → List of series: Enter salaries → OK Specific: Choose

Distribution → OK

gdp
12

10

0
1,000 2,000 3,000 4,000 5,000 6,000 7,000

17
❖ Frequency polygon

INV
7

0
100 200 300 400 500 600 700 800 900 1,000

Histogram Hist. Edge Polygon

❖ Cumulative increasing relative frequency curve

Click on Quick → Graph
•List of series: Saisirinv→ OK
•Specific: Choose Distribution → Choose Empirical CDF → OK

18
 inv

1.0

0.8

0.6

0.4

0.2

0.0
200 300 400 500 600 700 800 900

Cumulative decreasing relative frequency curve

Click on Quick → Graph

→ List of series : SaisirINV→ OK
Specific: Choose Distribution → Choose Empirical Survivor
OK
inv

1.0

0.8

0.6

0.4

0.2

0.0
200 300 400 500 600 700 800 900

19
 ❖ Graphs associated with a discrete quantitative character: Statistical table;
step curve

Example: Vaccine sales over the last 100 days are reported in the
table below:
0 1 2 3 4 5 6 0 1 3
0 1 2 3 4 5 6 1 1 3
0 1 2 3 4 5 0 2 2 1
0 1 2 3 4 5 0 3 3 2
0 1 2 3 4 1 2 3 2 3
0 1 2 3 4 1 2 1 1 2
0 1 2 3 4 1 2 2 1 1
0 1 2 3 4 1 2 0 1 2
0 1 2 3 4 1 2 1 1 2
0 1 2 3 3 1 2 2 2 3

➢ Statistical table
To do this, you need to transform the distribution into a column before importing it into
EVIEWS.
Then:
Click on Quick → Show →
Objects to display in a single window: Enter sales
OK
Click on View → One-Way Tabulation
OK

Tabulation of SALES
Sample: 1 100
Included observations: 100
Number of categories: 7

Value Count
0 14
1 27
2 26
3 18
4 9
5 4
6 2
Total 100

20
 ❖ staircase curve

Click on Quick → Graph → List of series: Enter sales → OK

Specific: Choose Distribution → Choose Empirical CDF → OK
Sales

1.0

0.8

0.6

0.4

0.2

0.0
0 1 2 3 4 5 6 7

21
Chapter 3: Econometric Estimation

3.1. OLS Estimation of Time Series Data

We are now looking to estimate the following linear model:

( ) = 0+ 1 ( ) + 2 +

This model indicates, according to Keynesian theory, that investment is a function

decreasing interest rate. We would like to test this theory for the case of
country BENI. Consequently, we expect the coefficient to be negative. Before doing
the estimation we are successively testing the linear correlation between the variables and the
normality for each of the variables.

Note: As this concerns time series, we should normally perform the test of
stationarity for each variable.

1. Determination of the degree of integration of the model variables: Stationarity

The assumptions of the Phillips-Perron test are:

H0: the process X is non-stationary
H1: the process X is stationary
The decision rule of the test is:
✓ If the probability value is greater than the alpha threshold, the null hypothesis of no
Stationarity is not rejected. The variable X is non-stationary.
✓ If the probability value is less than or equal to the alpha threshold, the null hypothesis of
Non-stationarity is rejected. The variable X is stationary.
❖ The test is based on three types of model.
▪ a model with constant
▪ A model with constant and trend
▪ A model without constant or trend
❖ It is essential to note that the Philips-Perron stationarity test is being performed.
on only one of the three models.
❖ In practice, we adopt a sequential strategy in three main steps. We start
to estimate the model with constant and trend. If the trend is not significant,
the model is estimated with a constant. Finally, if the constant is not significant, we
proceed with the estimation of the model without constant or trend.

22
 ❖ To do this, Click Quick → Series statistics → Unit Root Test → Enter X → OK
❖ Test type: Choose Phillips-Perron
❖ Test for unit root in: Choose level
❖ Include in test equation: Choose Trend and intercept
❖ Choose Newey-West Bandwidth
❖ OK
We first test the significance of the coefficient of the time variable.
@TREND("1972"), by the Student's method.
If the critical probability associated with the test is less than 5%, we reject the null hypothesis.
Thus, the coefficient associated with the time variable is significantly different from 0.
We keep the model with constant and trend. (to be done during the course)

2. Test of linear correlation between variables

The null hypothesis of the test is 'The variables are not correlated'. It will therefore be rejected if the
the associated critical probability is less than 5% (Conventional threshold). One can choose
other conventional thresholds (1% or 10% generally) taking into account the precision
sought after.
Instructions on Eviews:
▪ Click on 'Quick' then on 'Show' and enter the variables 'Invest GDP'
Click on 'Ok'.
▪ Then click on 'Views' and then on 'Covariance Analysis'

➢ Click on 'Ok' to finish.

23
 ➢ We obtain the following result:

Only the probability associated with the correlation between Investment and GDP is zero. Therefore, we can reject.

the null hypothesis that investment and GDP are not correlated. Therefore,

▪ The variables Investment and GDP are correlated.

▪ The variables Invest and Tinterest are not correlated.
▪ The GDP and interest rate variables are not correlated.

3. Jarque-Bera normality test

The null hypothesis of the test is that the variable follows a normal distribution.

Instructions on Eviews:
▪ Click 'Quick' then 'Group Statistics' then 'Descriptive Statistics' then
Common Sample
▪ Enter the names of the three series in the 'Series List' window: Invest GDP Interest
▪ Finally click on 'OK'

The result obtained is as follows after the instructions:

24
All the critical probabilities associated with the Jarque-Bera statistic are greater than
conventional thresholds (1%, 5%, and 10%). Therefore, one cannot reject the hypothesis of normality.
for each of these variables.

4. Actual estimation
Before the estimation, it seems logical to generate log(Investment) and log(GDP) respectively. For this
enter the following instructions directly into the 'command' bar:
generate lpib = log (gdp)

generate linvest = log(Invest)

Validate each time. New variables are therefore created, namely 'lpib' and 'linvest'. We
we can therefore estimate our model by following the instruction below:
➢ Click on 'Quick' and then on 'Estimate Equation'
➢ In the window that opens "Equation specification" we write the model as follows in
starting with the explained variable 'linvest'. 'C' denotes the constant.
Invest with interest
➢ You choose at the level of 'Method' 'Least Squares (NLS/ARMA)'
➢ Then you click on "ok". The estimated result is as follows:

25
Table 2: First Estimate

5. Tests on residues

Before any interpretation of the results, we will check the OLS assumptions on the
residuals to know: normality, homoscedasticity, and no autocorrelation.

5.1. Normality of errors

From the estimation menu, go to view/residual tests/histogram-normality test. We have the

next result:

26
The Jarque-Bera probability is greater than 5%. Therefore, we cannot reject the hypothesis.
according to which the residuals follow a normal distribution.

5.2. Test for heteroscedasticity of residuals (White test)

Perform view/residual tests/white heteroscedasticity (non cross term)

The two critically marked probabilities in yellow are greater than 5%: There is
homoscedasticity.

You can also perform the ARCH test to check for homoscedasticity.

5.3. Residual autocorrelation test

Several tests exist to verify autocorrelation.

5.3.1. Durbin-Watson Test

Application conditions: time series, more than 15 observations and estimation with
constant.

It reads on the results of the OLS estimation of the production function that DW=1.149.
But Eviews does not provide the probability associated with this statistic.

What the correlogram confirms (perform view/residual tests/correlogram Q statistics).

27
5.3.2. Breusch-Godfrey Test

This test is based on a Fisher test of the nullity of the coefficients and allows testing a
autocorrelation of order greater than or equal to 1.

Let's test a first-order autocorrelation. The instruction is as follows:

➢ Make view/residual tests/serial correlation LM test
➢ Enter 1 in "lag to include"
The result obtained is as follows:

The probability values are less than 5%, so we reject the null hypothesis.
correlation of errors. The errors are autocorrelated to order 1.
When we repeat the test with lag 2, the residuals become uncorrelated. This would mean
that the residuals are correlated with order 1 and not with order 2.

One can create the correlogram of the residuals to observe:

28
We only look at "Partial correlation". Upon reading the correlogram (Partial
correlation) only the first bar comes out of the confidence interval frame. The second one does not.
sort past. Therefore, there is first-order autocorrelation.

6. Estimation of the model by the Cochrane-Orcutt method

If all the tests on the residues were conclusive, then we will be able to interpret the estimated model.
at point 3. But given that autocorrelation is not verified, we cannot therefore use the
estimated model at point 3. Therefore, it will be necessary to correct this observed first-order autocorrelation.

Instructions on Eviews
➢ Click on "Quick" then on "Estimate Equation"
➢ Enter the following model:
Invest with ar(1) interest
➢ Always choose 'Least Squares' then 'Ok'.
In this case, we obtain:

29
Table 3: Estimation 2

We can visualize the residuals of the new estimated model. The first-order autocreation is therefore
corrected.

[Link] of the specification of the estimate 2

There are several specification tests (Ramsey, etc.). Here, we retain the Ramsey test.
Reset tests the following specification errors:
➢ Omission of explanatory variables
➢ Incorrect functional form
➢ Correlation between explanatory variables and error terms due among others to
measurement errors, lagged terms of the endogenous correlated with the errors.
The estimation by OLS becomes biased and inconsistent if the test is inconclusive.

Make view/stability diagnostics/Ramsey Reset test/number of fitted term (taper 3)

30
The model estimated in the second position by the Cochrane-Orcutt method is well specified.
because the probabilities marked in yellow are below 5%
8. Chow test on the second estimation

This is a Stability test. A natural question is that of the homogeneity of the parameters.
on two subpopulations. The null hypothesis is 'the model is stable.'

One can question the existence of a temporal break in behaviors. One can,
also, to ask oneself for example if the estimated investment behavior on series
Temporal is homogeneous over time or if the production technologies, estimated
on a panel of companies are homogeneous across sectors.

Obtaining test statistics.

➢ Implementation: view/stability tests/Chow Breakpoint test...
➢ Enter 'Breakpoint dates' 1994 then Ok

31
The result obtained is as follows:

The two probabilities are greater than 5%. The model estimated in second position by the
the Cochrane-Orcutt method is therefore stable

In other words, the empirical Fisher admits a probability exceeding the threshold: we reject
the null hypothesis of coefficient stability for the two concerned periods.

9. Interpretation of the coefficients of the model estimated in the second position by the method
Cochrane-Orcutt

Upon analysis, it is noted that:

➢ R²=0.94: 94% of the variance of 'linvest' is explained by the model. It is therefore

of good quality fit.
➢ The empirical Fisher p-value (F-statistic) is zero (F-statistic) is zero (<0.05): the
The model is satisfactory.
➢ The Student's t (t-statistic) of 'GDP' is greater than the theoretical thresholds (see table) and
the p-value is lower than the 1% threshold: the coefficient of 'lPIB' is significant.
➢ On the other hand, the P-value of 'Tinterest' is greater than 10%. The interest rate does not explain
not the investment in the country BENI during the study period.
3.2. Estimation of Panel Data

We are now looking to estimate the following model:

= 0+ 1 deped + 2 depinv + 2 depsante +

This model indicates, in accordance with Keynesian theory, that the growth rate
economic is a function of education expensesdeped); investments (depinv)and
health expendituresdepsante).

32
For panel data, after importing, it is necessary to specify the model.
Hausman specification test
H0 = Presence of random effects
Presence of fixed effects
First, we estimate the random model
➢ Click on 'Quick' and then on 'Estimate Equation'
➢ Enter the following model: tcpib c deped depinv depsante
➢ Panel options→Cross-Section/ Random→OK
TCRPIB
Method: Panel EGLS (Cross-section random effects)
2003 2017
Periods included: 15
Cross-sections included: 5
Total panel (balanced) observations: 75
Swamy and Arora estimator of component variances

Variable Coefficient Std. Error t-Statistic Prob.

C 2.490287 1.367752 1.820716 0.0729

Department of Education
-0.077169 0.061614 -1.252455 0.2145
DEPINV 0.073917 0.039795 1.857435 0.0674
HEALTH SERVICES 0.438397 0.413242 1.060871 0.2923

Effects Specification
S.D. Rho

Cross-section random 0.000000 0.0000

Idiosyncratic random 2.066954 1.0000

Weighted Statistics

R-squared 0.096093 Mean dependent var 4.915382

Adjusted R-squared 0.057900 S.D. dependent var 2.142107
Standard Error of regression 2.079169 Sum squared resid 306.9289
F-statistic 2.515979 Durbin-Watson stat 2.343525
Prob(F-statistic) 0.065123

Unweighted Statistics

R-squared 0.096093 Mean dependent var 4.915382

Sum squared resid 306.9289 Durbin-Watson stat 2.343525

33
 ❖ View→Fixed/Random effects testing→Correlated Random effects-Hausman test

Correlated Random Effects - Hausman Test

Equation: EQ01
Test cross-section random effects

Chi-Square.
Test Summary Statistic Chi-Sq. d.f. Prob.

Cross-section random 4.667298 3 0.1978

** WARNING: estimated cross-section random effects variance is zero.

Cross-section random effects test comparisons:

Variable Fixed Random Var(Diff.) Prob.

DEPED -0.074766 -0.077169 0.000779 0.9314

DEPINV 0.097055 0.073917 0.001183 0.5012
HEALTHCARE -0.195497 0.438397 0.236144 0.1921

Cross-section random effects test equation:

TCRPIB
Method: Panel Least Squares
02/07/23
2003 2017
Periods included: 15
Cross-sections included: 5
Total panel (balanced) observations: 75

Variable Coefficient Standard Error t-Statistic Prob.

C 3.543794 2.096977 1.689954 0.0957

Department of Education
-0.074766 0.067638 -1.105378 0.2729
DEPINV 0.097055 0.052601 1.845112 0.0694
DEPSANTE -0.195497 0.637898 -0.306471 0.7602

Effects Specification

Cross-section fixed (dummy variables)

R-squared 0.157010 Mean dependent var 4.915382

Adjusted R-squared 0.068937 S.D. dependent var 2.142107
Standard Error of regression 2.066954 Akaike info criterion 4.390567
Sum squared resid 286.2441 Schwarz criterion 4.637766
Log likelihood -156.6463 Hannan-Quinn criterion. 4.489271
F-statistic 1.782715 Durbin-Watson stat 2.489784
Prob(F-statistic) 0.105260

The probability associated with the Hausman statistic is 0.1978, which is above the 5% threshold. Therefore H0
cannot be rejected. So there is the presence of random effects.

34
 ❖ Estimation of the selected model
According to the Hausman test, the model to be estimated is random effects.

TCRPIB
Method: Panel EGLS (Cross-section random effects)
Sample: 2003 2017
Periods included: 15
Cross-sections included: 5
Total panel (balanced) observations: 75
Swamy and Arora estimator of component variances

Variable Coefficient Standard Error t-Statistic Prob.

C 2.490287 1.367752 1.820716 0.0729

Department of Education
-0.077169 0.061614 -1.252455 0.2145
DEPINV 0.073917 0.039795 1.857435 0.0674
DEPHEALTH 0.438397 0.413242 1.060871 0.2923

Effects Specification
S.D. Rho

Cross-section random 0.000000 0.0000

Idiosyncratic random 2.066954 1.0000

Weighted Statistics

R-squared 0.096093 Mean dependent var 4.915382

Adjusted R-squared 0.057900 S.D. dependent variable 2.142107
Standard Error of regression 2.079169 Sum squared resid 306.9289
F-statistic 2.515979 Durbin-Watson stat 2.343525
Prob(F-statistic) 0.065123

Unweighted Statistics

R-squared 0.096093 Mean dependent var 4.915382

Sum of squared residuals 306.9289 Durbin-Watson stat 2.343525

❖ Validity of the model: Normality test of the residuals and the autocorrelation test
of independence )

➢ Normality test of the residuals

H0 = The errors follow a Normal distribution

H1= The errors do not follow a Normal distribution

After the estimation of the chosen model, we do:

View→Residual diagnostic/ Histogram- Normality test

35
 12
Series: Standardized Residuals
Sample 2003 2017
10
Observations 75

8 Mean 8.19e-16
Median 0.226679
6 Maximum 6.101348
Minimum -6.588122
Std. Dev. 2.036587
4
Skewness -0.214878
Kurtosis 3.877159
2
Jarque-Bera 2.981556
0 Probability 0.225197
-6 -4 -2 0 2 4 6

The probability associated with the Jarque-Bera statistic is 0.225197, which is above the 5% threshold.
So we cannot dismiss the hypothesis that the residuals follow a normal distribution.

❖ Error autocorrelation test (Independence test)

H0 = independent
dependents

After estimating the chosen model, we do:

View → Residual diagnostic / Cross-Section Dependence Test

Residual Cross-Section Dependence Test

Null hypothesis: No cross-section dependence (correlation) in residuals
Equation: EQ01
Periods included: 15
Cross-sections included: 5
Total panel observations: 75
Note: non-zero cross-section means detected in data
Cross-section means were removed during computation of correlations.

Test Statistics d.f. Prob.

Breusch-Pagan LM 8.407690 10 0.5891

Pesaran scaled LM -0.356051 0.7218
Pesaran CD 0.904741 0.3656

The probability associated with the Breusch-Pagan statistic is 0.5891, which is greater than
5% threshold. Therefore, we cannot reject the hypothesis that the residuals are
independent.
Conclusion: The model is valid and we can interpret the estimated coefficients.

36