Professor Bob Reed

Second Semester 2009

University of Canterbury
ECON 324

WEEK #3 ASSIGNMENT
DUE: Monday, July 27th
Do the following exercises:
1A.

Review the Week #1 and Week #2 Assignments.

1B.

Answer Questions #1-2.

2A.

Read the following material

This section lays out a suggested general procedure for determining which
FGLS estimator to use in a given situation. It is meant only as a rough
beginners guide to get you started in your EViews estimation. You will want
to amend this to be more sophisticated as you learn more econometrics.
However, I believe that the following procedure, which is based primarily on
my own research experience, is better than none.
NOTE: Dont worry if you find this confusing the first time through. This
weeks assignment will subsequently walk you through two empirical
examples to help you gain confidence.
STEP ONE: Estimate the model using OLS
STEP TWO: Add fixed effects
If N > T, add Period fixed effects first. If T > N, add Cross-section fixed
effect first.
After you add your first set of fixed effects, compare the Schwarz criterion
with fixed effects to the Schwarz criterion without fixed effects. If the
criterion is lower with fixed effects, then keep them. If higher, do not keep
them.
After you check for the one kind of fixed effects (say, Period), check for the
other.
NOTE #1: Fixed effects are dummy variables. Cross-section fixed effects
are a set of dummy variables where each cross-sectional unit (minus one of
course) gets its own dummy variable. Period fixed effects are a set of dummy
variables where each time period (again minus one) gets its own dummy
variable. We will talk more about fixed effects later in the course.

NOTE #2: You can have BOTH cross-section and period fixed effects.
NOTE #3: When you include Cross-section fixed effects, you cannot have
any variables that are the same over time for each cross-sectional unit. For
example, race and gender variables do not change over time for a given
person. Therefore, they could not be included in an equation with Crosssection fixed effects.
NOTE #4: When you include Period fixed effects, you cannot have any
variables that are the same for all cross-sectional units for each time unit. For
example, the variable Year would be the same for all cross-sectional units
for a given year and thus could not be included in an equation with Period
fixed effects.
STEP THREE: Determine which GLS Weights option to use
If T > 2N, use Cross-section SUR
If T > N, use Cross-section weights
If T < N, use Period weights
1
If T < N, use Period SUR
2
NOTE #1: The rationale for this rule has to do with reliability. When T is
larger than N, the Cross-sectional parameters are estimated with greater
reliability than the Period parameters, ceteris paribus; and vice versa.
NOTE #2: It is possible that some of these options may not be estimable in
EViews, depending on the particular set of fixed effects included in your
model. If the first best option is unavailable, keep going backwards in terms
of increased restrictions. For example, if you cant use Cross-section SUR,
use Cross-section weights. If you cant use Cross-section weights, use No
weights, etc.
STEP FOUR: Determine which Coef covariance method option to use
If T > 2N, then use the following: (i) Coef covariance method = Crosssection SUR, and (ii) Coef covariance method = White cross-section
If T > N, then use the following: (i) Coef covariance method = Crosssection weights, and (ii) Coef covariance method = White diagonal
If T < N, then use the following: (i) Coef covariance method = Period
weights, and (ii) Coef covariance method = White diagonal

1
N, then use the following: (i) Coef covariance method = Period
2
SUR, and (ii) Coef covariance method = White period
If T <

STEP FIVE: Add a lagged dependent variable to the specification

Add a lagged dependent variable to the list of variables in the Equation
Estimation/Specification dialog box. For example, if the dependent variable
is Y, add the variable Y(-1). Keep the respective term in the specification
if the associated p-value is less than 0.05.
NOTE #1: The rationale for doing this is threefold: (i) it allows explanatory
variables to have effects that extend beyond the current period, (ii) it can serve
as a proxy for omitted variables associated with the cross-sectional unit, and
(iii) it can serve as a control for serial correlation.
NOTE #2: Adding a lagged dependent variable causes you to lose one
observation per cross-sectional unit.
NOTE #3: The preceding five-step procedure is designed to produce two
final equation estimates. Usually, the coefficient estimates will be the same
in both 1 , but the coefficient standard errors (and t-values, p-values, hypothesis
test results, etc.) will be different. I suggest that you reject the null hypothesis
only if both sets of estimates indicate rejection at the 5% significance level.
2B.

Answer Question #3.

3A.

Read the following material

This section walks you through the steps involved in (i) downloading an
EXCEL spreadsheet from the course website and (ii) importing that data set
into EViews to make an EViews workfile.
NOTE: All data sets and subsequent empirical work should be saved on your
P: drive.
STEP ONE: Create a folder on your P: drive called ECON324. Now
create a folder within that folder called WEEK3
STEP TWO: Download the Excel spreadsheet file USAirlineData from
the website: http://www.econ.canterbury.ac.nz/personal_pages/bob_reed/
Courses/ECON_324/Week3/index.html into your WEEK3 folder.
The data set USAirlineData is a balanced panel data set consisting of a total
of 90 observations on 6 U.S. Airlines for the years 1970-1984. The data are
listed on the Data sheet, and defined on the Variable definitions sheet.

The two sets of estimates will produce different coefficient estimates only if one includes a lagged
dependent variable term, and the other does not.

After you download the EXCEL spreadsheet, take a look at the data to get a
feel for the data set.
STEP THREE: Read the EXCEL data set into EViews
First, start EVIEWS.
Once EViews has opened up, select File/New/Workfile from the Main Menu.
A dialog box should appear. Under the Workfile structure type, choose
Balanced Panel from the drop down menu. Under Panel specification,
enter (i) 1970 for Start date:, (ii) 1984 for End date:, and (iii) 6 for
Number of cross sections: Then click OK (see below).

The following Workfile window should appear:

Next, we will import the Excel data set USAirlineData.xls. From the
toolbar of the Workfile window, click on the Proc button. From the dropdown menu, choose Import/Read Text-Lotus-Excel. An Open window
will appear.
On the Files of type: line of the Open window, choose Excel(*.xls). On
the Look in: line of the Open window, choose the folder where your
USAirlineData.xls file resides. Then select the USAirlineData file so that
its name appears on the File name line. Your Open window should look
like the one below. Then click on the Open button.
4

A new window should open up: the Excel Spreadsheet Import window.
Under Upper-left data cell, type A2. Under Names for series or Number
if named in file, type 6. (NOTE: To confirm that there are 6 variables in
USAirlineData, you should check the original EXCEL data sheet.) Your
window should now look like the one below. Click OK.

If you successfully imported the data from USAirlineData.xls, your workfile

should now look like the one below:

Save this workfile by choosing File/Save As from the EViews menu bar.
When the Save As window opens up, name the workfile USAirlineData
and save it into your WEEK3 folder as an EViews Workfile (see below).
Choose the double precision option when given the choice.

Exit Eviews by clicking on the X in the upper, right-hand corner of your

EViews display.
3B.

Answer Question #4.

4A.

Read the following material (EXAMPLE #1)

This section walks you through the general procedure outlined in Exercise #2
above, applying it to the EViews data set you created in Exercise #3 above.
Your task is to estimate the following model relating airline costs to output,
fuel price, and capacity:

COSTit = 0 + 1 OUTPUTit + 2 PRICEit + 3 CAPACITYit + it,

where i 1,2,...,6 ; t 1970,1971,...,1984 ; and the explanatory variables
are independent; and the error terms are likely characterized by nonspherical
behaviour.

NOTE #1: I will walk you through the steps of estimating your model. You
should repeat everything I do on your own and see if you get the same results.
NOTE #2: The goal in estimation is not perfection. The goal is to get
close. There is a truth out there. As econometricians, we know the world
is far more complicated than our simple models. Our goal is to get close to
that truth.

STEP ONE: Estimate the model using OLS

Open up the workfile you created in Exercise #3 above. Get into your
WEEK3 folder and double-click on the workfile USAirlineData.
To estimate a regression equation, go to the EViews menu bar and select
Quick/Estimate Equation. An Equation Estimation window should
open.
In the Equation Estimation/Specification work area, enter the following
specification:
cost c output price capacity

NOTE: You have to specify the constant term c if you want to run the
regression with an intercept (see below).

Next, click on the Panel Options tab. Make sure (i) that None is selected
for both Cross-section and Period Effects specifications, (ii) GLS Weights
= No weights, and (iii) Coef covariance method = Ordinary.

Click OK. An Equation window should appear with estimation results.

These are your OLS Results (see below).

Save these results by clicking on the Freeze button. When the new window
opens up, click on the Name button and call your results STEP_ONE (see
below): These results will be permanently saved in your USAirlinesData
workfile for future reference.

STEP TWO: Add fixed effects

Since T > N, we add Cross-section fixed effects first. Go back to the

Equation EQ01 window, and click on the Estimate button, then click on
the Panel Options tab. Select Fixed from the drop-down menu associated
with Cross-section under Effects specification (see below):

Then click OK. You will obtain the results below:

Note that the Schwarz criterion value with Cross-section fixed effects is
27.38525. This is larger than the Schwarz criterion value without these fixed
effects (Schwarz criterion = 27.20208 from previous OLS results). Big is Bad
when it comes to the Schwarz criterion! Therefore, do not include Crosssection fixed effects in your model.

10

Repeat the previous analysis, only now include Period fixed effects (but not
Cross-section fixed effects see below). 2

You will obtain the following estimation results:

NOTE: The Schwarz criterion value is now 22.17080, compared to

27.20208 without these fixed effects. Small is beautiful when it comes to the
2

If the inclusion of cross-sectional fixed effects had lowered your Schwarz criterion, you would have
kept the cross-section fixed effects. These would then be included in your model when you
subsequently added period fixed effects.

11

Schwarz criterion! Therefore, you decide to keep the Period fixed effects in
your equation specification.
Save these results by clicking on the Freeze button. When the new window
opens up, click on the Name button and call your results STEP_TWO.
STEP THREE: Determine which GLS Weights option to use

T = 15 and N = 6, so our data fits into the category: T > 2N. Accordingly, we
will attempt to estimate the model (which now includes Period fixed effects)
using the GLS Weights = Cross-section SUR option. To do that, return to
the Equation EQ01 window, click on the Estimate button, then click on
the Panel Options tab. Select GLS Weights = Cross-section SUR (see
below);

Then click on OK. You receive the following error message:

12

Sadly, this FGLS estimator is not estimable in EViews. In this case, your
notes instruct you to go backwards and estimate the next most restrictive
model. That would be GLS Weights = Cross-section weights. However, if
you try and estimate this model, you will find that this model is also not
estimable in EViews. Thus, you are left with GLS Weights = No weights.
This is the same model you estimated in STEP TWO. Nevertheless, estimate
this model again and save it in the usual fashion as STEP_THREE.
STEP FOUR: Determine which Coef covariance method option to use

Since T > 2N, we estimate two equations, one with Coef covariance method
= Cross-section SUR, the other with Coef covariance method = White
cross-section.
Starting with Coef covariance method = Cross-section SUR, you should fill
out the Panel Options tab in your Equation Estimation window as follows
(see below).

Click the OK button to produce the corresponding estimates (see next page).
Save your results using the Freeze button and name it STEP_FOURA.
Repeat the process, only this time set Coef covariance method = White
cross-section. Save these latter results under the name STEP_FOURB.
Compare your results with the output reported below.

13

14

STEP FIVE: Add a lagged dependent variable to the specification.

Return to the Equation EQ01 window and click on the Estimate button.
With the Specification tab selected, add the variable COST(-1) to the list
of variables (see below):
cost c output price capacity cost(-1)

Click OK. The results are reported below.

15

Note that the Prob value (i.e. p-value) associated with COST(-1) is
0.0689. This variable is not significant at the 5% level, thus we do not keep
the COST(-1) term. Go back and reestimate this model without this variable
and save the set of estimates as STEP_FIVEA (your results should be
identical to your STEP_FOURA results.
Repeat the preceding analysis, only this time add the COST(-1) term to your
STEP_FOURB specification (the one where Coef covariance method =
White cross-section). You will once again find that the COST(-1) coefficient
has a p-value greater than 5% (the p-value is 0.0931). Thus, you again drop
the variable COST(-1) from your specification, reestimate the model, and save
the results as STEP_FIVEB (which will be identical to your
STEP_FOURB results).
The STEP_FIVEA and STEP_FIVEB results are your final equation
estimates. These are the estimates that you should use in your analysis of the
determinants of U.S. Airline costs (e.g. interpretation of coefficients,
hypothesis tests, etc.).
Exit EViews, being sure to save your results for future reference.

4B.

Answer Questions #5-7.

5A.

Read the following material (EXAMPLE #2)

Download the Excel spreadsheet file Wages from the website:
http://www.econ.canterbury.ac.nz/personal_pages/bob_reed/
Courses/ECON_324/Week3/index.html into your WEEK3 folder.
The data set Wages is a balanced panel data set consisting of a total of 4360
observations on 545 workers for the years 1980-1987. The data are listed on
the Data sheet, and defined on the Variable definitions sheet. After you
download the EXCEL spreadsheet, take a look at the data to get a feel for the
data set.
Repeat the procedure of Exercise #3 above, saving the associated EViews
workfile in your WEEK3 folder under the name Wages.

5B.

Answer Question #8.

16

6A.

Read the following material (EXAMPLE #2 -- continued)

Your task is to estimate the following model relating a workers wage to their
labor market experience and employment in the manufacturing sector:
LNWAGEit = 0 + 1EXPERIENCEit + 2EXPERIENCE2it + 3MANUit + it,
where i 1,2,...,545 ; t 1980,1981,...,1987 ; and the explanatory variables
are independent; and the error terms are likely characterized by nonspherical
behaviour.
Repeat the procedure of Exercise #4 above, saving the updated EViews
workfile in your WEEK3 folder under the name Wages.
NOTE: I have posted my output for each of the five steps on the course
webpage:
http://www.econ.canterbury.ac.nz/personal_pages/bob_reed/Courses/
ECON_324/Week3/index.html
Check your results with mine at each step of the procedure to see that we
agree. Make note of any discrepancies.

6B.

Answer Questions #9-10.

Be prepared to answer the following questions in class:

1.

Consider the general specification of the error variance-covariance matrix,

Cov E
11,11

12 ,11

1,11
21,11

22 ,11

2,11

1,11

2 ,11

,11

11,12

11,1

11,21

11,22

12 ,12

1,12
21,12

12 ,1

1,1T
21,1

12 ,21

1,21
21,21

12 ,22

1,22
21,22

22 ,12

2,12

1,12

22 ,1

2,1T

1,1

22 ,21

2,21

1,21

2 ,12

,12

2 ,1

,1

2 ,21

,21

11,2

11, 1

11, 2

12 ,2

1,2
21,2

12 , 1

1,N 1
21, 1

12 , 2

1, 2
21, 2

22 ,22

2,22

1,22

22 ,2

2,2

1,2

22 , 1

2,N 1

1, 1

22 , 2

2, 2

1, 2

2 ,22

,22

2 ,2

,2

2 , 1

, 1

2 , 2

, 2

11,NT
12 ,NT

1,
21,NT

22 ,NT

2,

1,NT

2 ,NT

HINT: An enlarged version of this error variance-covariance matrix is attached to

the end of this assignment. You may find it helpful to make copies of this page and
use them to answer the questions below.
17

1A.

1B.
1C.

1D.

1E.
1F.

1G.
1H.

What specific restrictions on are implied when one chooses the EViews
options: GLS Weights = No weights and Coef covariance method =
Ordinary?
What specific restrictions on are implied when one chooses the EViews
option Coef covariance method = White diagonal?
What specific restrictions on are implied when one chooses the EViews
option Cross-section Weights for either GLS Weights or Coef covariance
method?
What specific restrictions on are implied when one chooses the EViews
option Cross-section SUR for either GLS Weights or Coef covariance
method?
What specific restrictions on are implied when one chooses the EViews
option Coef covariance method = White cross-section?
What specific restrictions on are implied when one chooses the EViews
option Period Weights for either GLS Weights or Coef covariance
method?
What specific restrictions on are implied when one chooses the EViews
option Period SUR for either GLS Weights or Coef covariance method?
What specific restrictions on are implied when one chooses the EViews
option Coef covariance method = White period?

2.

The less restrictive the assumptions are on , the more parameters one needs
to estimate. What is the main cost of having to estimate more parameters in
the error variance-covariance matrix?
HINT: Refer to your answer for Questions #11A and 11B on the Week #2
Assignment.

3.

See if you can provide a rationale, in plain English, for the logic underlying
each of the five steps of this general procedure.

4.

With your workfile window active, click on the Quick button from the
EViews main menu, then choose Show A Show dialog box will open
up. Type in the name of one of the variables and see if the EViews data
matches up with the data from the EXCEL spreadsheet. Play around with
some of the options in the Series window to get a feel for what EViews can
do.

5.

Interpret your final estimation equations. What do these results have to say
about the relationship between costs for U.S. airlines and (i) output, (ii) fuel
price, and (iii) capacity? Does anything seem unusual to you?

6.

The estimated coefficients for the variables Output and Capacity are
insignificant at the 5% level. Interpret these results. On the basis of these
estimates, what is your best guess about the effect of Output and
Capacity on Cost?

18

7.

Compare your output from STEP_ONE with your final equation

estimates. The results from STEP_ONE are stronger: every coefficient is
significant at well below the 5% level. Which set of results is better? Be
prepared to explain your reasoning.

8.

With the Wages Eviews workfile open, use the Quick/Show button and
double-check your variables to make sure you correctly imported the data.
Take some time to play around with your data. Do you find anything
noteworthy? In particular, is there a time trend in workers wages? Is this
what you would expect?

9.

Interpret your final estimation equations.

(HINT: You may find the handout A Note on Functional Forms to be
helpful in this regard. It is posted on the course webpage.)
Interpret the coefficient on the variable MANU. In plain English, explain
what a coefficient of 0.059 means.
The estimated coefficients for the variables Experience and Experience2
are insignificant at the 5% level. Interpret these results. On the basis of these
estimates, what is your best guess about the effect of an additional year of
experience on a workers wages, ceteris paribus?
HINT: This is a tricky question. Give it a good try, but dont spend too much
time on it.

9A.
9B.

10.

Compare your output from STEP_ONE with your final equation

estimates. The results from STEP_ONE are stronger: every coefficient is
significant at well below the 5% level. Which set of results is better? Be
prepared to explain your reasoning.

19

 Cov E

11,11

12 ,11

1,11
21,11

22 ,11

2,11

1,11

2 ,11

,11

11,12

11,1

11,21

11,22

12 ,12

1,12

12 ,21

1,21

12 ,22

1,22

21,12

12 ,1

1,1T
21,1

21,21

22 ,12

22 ,1

2,12

1,12

2,1T

1,1
2 ,1

2 ,12

,12

,1

11,2

11, 1

11, 2

12 , 1

1,N 1
21, 1

12 , 2

1, 2

21,22

12 ,2

1,2
21,2

21, 2

22 ,21

22 ,22

22 ,2

22 , 1

22 , 2

2,21

1,21

2,22

1,22

2 ,22

2,N 1

1, 1
2 , 1

2, 2

1, 2

2 ,21

2,2

1,2
2 ,2

,21

,22

,2

20

, 1

2 , 2

, 2

11,NT
12 ,NT

1,
21,NT

22 ,NT

2,

1,NT

2 ,NT