See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/335222720

NARDL: Implementation Using Eviews Add-in

Poster · August 2019

DOI: 10.13140/RG.2.2.30251.59687

CITATIONS READS

0 1,342

1 author:

Olaolu Olayeni
Obafemi Awolowo University
23 PUBLICATIONS   135 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Eviews Add-in for NARDL View project

All content following this page was uploaded by Olaolu Olayeni on 17 August 2019.

The user has requested enhancement of the downloaded file.

 NARDL: Implementation Using Eviews Add-in

Olayeni Olaolu Richard

Department of Economics
Obafemi Awolowo University, Ile-Ife, Nigeria

[email protected]/ [email protected]

The workfile containing the data used in this documentation can be downloaded
at:
https://drive.google.com/drive/u/1/folders/1vCxOzxRMVBIzvaxBla_se-koGq95bfdf

NB: If you find this material including the add-in useful, kindly cite as appropriate.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Features
o It is fully compatible with Eviews10 and should also work with Eviews 9.
o It allows for estimation of models with multiple asymmetric variables.
o It plots CUSUM and CUSUMSQ graphs.
o It plots multiplier graphs for all the specified threshold variables.
o It allows for the generation of the ARDL testable form through which
series of hypothesis tests (e.g., asymmetry tests) can be conducted.
o The Make Nonlinear ARDL routine no longer produces the NARDL##
equation objects automatically. Rather, the series to decompose must be
specified using the Asyvars keyword as explained below.

o Working with multiple decomposed variables

 Estimate the ARDL model as usual. In this example, we want to study
the asymmetry effects of lgdp lm2 and loip on lexhr. You are to ensure
that your model satisfies basic regression requirements.
 List the asymmetric variable(s) for the ARDL using Label. To do this,
click on View→Label. In the Attribute column, type Asyvars. This is a
keyword and should be specified without space. In the corresponding
Value column, list the asymmetric variable names. Note that other cells
are not editable! In the figure below, we have listed three variables in
the estimated ARDL model as asymmetric variables: lm2 lgdp loip.
(Note also that the number of variables are going to increase.
Therefore, be sure that you have enough sample size before engaging in
multiple asymmetric variable analysis.) The dependent variable is NOT
to be listed.
 You don’t need to be particularly concerned about the order of the
linear ARDL model since you are focused on the asymmetric model. The
generated the asymmetric ARDL model will determine its own lag
order, although using the same maximum lag order chosen for the
linear ARDL model. The generated asymmetric ARDL can be re-
estimated directly from its own equation output view and the new
order determined. Our linear ARDL model is given in the figure below:
 Having listed the asymmetric variables as explained above, click Add-
ins→Make Nonlinear ARDL to generate the nonlinear ARDL model. The
boxed Add-ins are the relevant toolkits needed to do most of the
analysis that you might want to carry out.

 After estimation, the workfile will be modified with additional series

objects generated for the asymmetric variables with _POS and _NEG
appended. The total number of new series objects generated will be
twice the number of asymmetric variables listed with the keyword
Asyvars.

 The nonlinear/asymmetric ARDL model is reported in the figure below:

 Generate the multiplier graphs for the asymmetric variables: While the
NARDL## is still in focus, click Add-ins→ NARDL Multiplier Graph. The
process will communicate the asymmetric variables to the NARDL
Multiplier Graph routine.
 Should you want to generate the multiplier graphs for a subset of
asymmetric variables, you can simply edit the list through View→Label
of the NARDL## equation object. The multiplier graphs will be
generated for all the asymmetric variables listed: Multiplier effects have
been computed as a unit change. Of course, if the variables were log-
transformed, the effects should be interpreted as a percentage change.
 .2

.1

.0

-.1

-.2

-.3

-.4

-.5
1 3 5 7 9 11 13 15

Multiplier for LM2(+)

Multiplier for LM2(-)
Asymmetry Plot (with C.I.)

2.0

1.5

1.0

0.5

0.0

-0.5
1 3 5 7 9 11 13 15

Multiplier for LGDP(+)

Multiplier for LGDP(-)
Asymmetry Plot (with C.I.)
 .6

.5

.4

.3

.2

.1

.0

-.1
1 3 5 7 9 11 13 15

Multiplier for LOIP(+)

Multiplier for LOIP(-)
Asymmetry Plot (with C.I.)

 While the NARDL## is still in focus, click Add-ins→ NARDL CUSUM and
CUSUMQ Graph. The CUSUM and CUSUMSQ graphs for the NARDL
model are generated. It can also be used to generate the CUSUM and
CUSUMQ graphs for linear model.

16

12

-4

-8

-12

-16
2007 2008 2009 2010 2011 2012 2013 2014

CUSUM 5% Significance
 1.4

1.2

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

-0.4
2007 2008 2009 2010 2011 2012 2013 2014

CUSUM of Squares 5% Significance

o Testing the short and the long-run asymmetry

Rather than do the summary for the hypothesis testing, I consider it more
appropriate to deliver the full menu so that each analyst can carry out the test of
her choice.
Thus the present NARDL add-in comes with a new routine called Make Testable
Form, which generates an independent equation object called LRFORM## for
which all the conventional hypothesis-testing tools are available. The equation
object LRFORM## reports the same results as does the upper panel of the inbuilt
routine Long Run Form and Bounds Test. It, however, differs from the latter
because it’s not hardwired and can be used for hypothesis testing. The Make
Testable Form routine also works with the linear ARDL model specification.
You can generate LRFORM## as follows:
 Estimate a linear ARDL model:
 While the equation object is still in focus, click on Add-ins→Make
Testable Form. The Make Testable Form routine generates LRFORM##,
which will be listed in the workfile with ## indicating the index
number.
Figure ?: Output using Make Testable Form
The output is the same as the hardwired output produced by the ARDL Long Run
Form. See the figure below. With this at hand, an analyst can go about hypothesis
testing as she would using the LS method.

Figure ?: Hardwired Eviews Output

More importantly, it can be used with the NARDL method. To do so, first make
the NARDL using Make Nonlinear ARDL from which the Make Testable Form
routine can be used. Thus asymmetry test can be tested from the generated LS
output. Using the above estimate, we make the nonlinear ARDL:
The hardwired from Eviews output is reported as:
Again, this hardwired output cannot be used for hypothesis testing. Thus, we
generate its equivalence using LS:
The two outputs are the same! However, for this particular example there is no
short-run asymmetry. To test for the long-run asymmetry, we carry out the Wald
test:
The null of no long-run asymmetry is not rejected.
 o Testing for Asymmetry in Model with Multiple Decomposed Series
From the analysis in Example 1, we generate the following model from which we
plan to test the asymmetry for the short and long run.

To help with proper assignment of the estimated coefficients to their

placeholders, it is recommended that Representations is generated for the model.
This is given in figure below:
  From the figure, the LR asymmetry test involving LOIP can be stated as
H0: C(3)=C(4)
and that involving LGDP as
H0: C(5)=C(6)
 The short run asymmetry test is given by
H0: C(11)+C(12)+C(13)=C(14)+C(15)+C(16)
for LOIP and
H0: C(17)+C(18)=C(19)+C(20)
for LGDP.
 Joint test is given by
H0: C(3)=C(4), C(11)+C(12)+C(13)=C(14)+C(15)+C(16)
for LOIP and by
H0: C(5)=C(6), C(17)+C(18)=C(19)+C(20)
for LGDP.

View publication stats