8:49 a.m.
Hi all,
I'm a student at University of Bergamo and i'm writing my dissertation on the
asymmetric eur/usd exchange rate effects on trade relations. I need some help to figure
out how to perform the Wald test on Gretl in order to test the asymmetry hypotesis.
Thanks in advance
F.Marchini
9:12 a.m.
It can be found in Function Package on waldTest by Oleh
Komashko.
On Tue, 18 Aug 2020 at 6:19 PM, <f.marchini(a)studenti.unibg.it> wrote:
Hi all,
I'm a student at University of Bergamo and i'm writing my dissertation on
the asymmetric eur/usd exchange rate effects on trade relations. I need
some help to figure out how to perform the Wald test on Gretl in order to
test the asymmetry hypotesis.
Thanks in advance
F.Marchini
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
9:39 a.m.
Am 18.08.2020 um 15:12 schrieb Dhanasekaran Kuppusamy:
It can be found in Function Package on waldTest by
Oleh
Komashko.
Indeed Oleh has a package which covers a lot of cases, but for 95% of
applications the native/built-in restriction Wald tests are just fine
and should be used.
On Tue, 18 Aug 2020 at 6:19 PM, <f.marchini(a)studenti.unibg.it
<mailto:f.marchini@studenti.unibg.it>> wrote:
Hi all,
I'm a student at University of Bergamo and i'm writing my
dissertation on the asymmetric eur/usd exchange rate effects on
trade relations. I need some help to figure out how to perform the
Wald test on Gretl in order to test the asymmetry hypotesis.
This is a quite basic question and so initially I wanted to tell you to
read the documentation. But I have to admit that the plain case of
testing a standard (linear) restriction in a standard (OLS) regression
doesn't really seem to be covered in the user guide. (Everybody please
correct me if I'm missing something.)
So: In the estimation output window go to Test - Linear Restriction. The
resulting window looks a little empty, but it has a help button with
some explanations behind it. Further help can be found in the command
reference for the 'restrict' command block.
You might also try other help sources such as Lee Adkins' ebook.
I'm wondering whether we could beef up the interface to basic
restrictions, but that's a different topic...
If you have further questions, please be more specific about your
equation and how exactly your tested restriction looks.
cheers
sven
9:40 a.m.
On Tue, 18 Aug 2020, f.marchini(a)studenti.unibg.it wrote:
Hi all,
I'm a student at University of Bergamo and i'm writing my dissertation
on the asymmetric eur/usd exchange rate effects on trade relations. I
need some help to figure out how to perform the Wald test on Gretl in
order to test the asymmetry hypotesis.
There are several ways to do this. In many cases, asymmetry tests are
carried out by
(a) estimating a model (usually by OLS)
(b) performing a test where the null hypothesis is that a subset of the
regressors can be omitted
In the simplest case, suppose you have a dependent variable y and an
explanatory variable x that could have an asymmetric effect on y depending
on a binary indicator d (for example, d=1 <-> \Delta x > 0).
The simplest way to go is to estimate
y = b0 + b1 * x + b2 * x*d
and then testing for H0: b2 = 0
So you'd go for something like
<hansl>
series asymm = x * d
ols y const x asymm
omit asymm
</hansl>
Variations on this theme are countless.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
4:55 a.m.
Thank you for your answer. Following the literature I adopted the following equation:
DLnTBt=a'+Σb'*DLnTBt-k+Σc'*DLnYt-k+Σd'*DLnYWt-k+Σe'*DPOSt-k+Σf'*DNEGt-k+θ0*LnTBt-1+θ1*LnYt-1+θ2*LnYWt-1+θ3*POSt-1+θ4*NEGt-1+εt
where POSt and NEGt are the two variables obtained from the partial sum of positive and
negative changes of the lnREX variable.
The restrictions look as follow:
∑ e′ ≠ ∑ f ′ for short-run asymmetry;
-θ3/θ0 ≠ - θ4/θ0 long-run asymmetric effects.
To test these two inequalities I have to apply the Wald test.
11:34 a.m.
Am 22.08.2020 um 10:55 schrieb f.marchini(a)studenti.unibg.it:
Thank you for your answer. Following the literature I adopted the
following equation:
DLnTBt=a'+Σb'*DLnTBt-k+Σc'*DLnYt-k+Σd'*DLnYWt-k+Σe'*DPOSt-k+Σf'*DNEGt-k+θ0*LnTBt-1+θ1*LnYt-1+θ2*LnYWt-1+θ3*POSt-1+θ4*NEGt-1+εt
where POSt and NEGt are the two variables obtained from the partial
sum of positive and negative changes of the lnREX variable.
OK. Please try to use at least some blanks to increase the readability
of such a formula line in pseudo-code. (Of course Latex notation would
also help, but not necessary.)
The restrictions look as follow: ∑ e′ ≠ ∑ f ′ for short-run
asymmetry; -θ3/θ0 ≠ - θ4/θ0 long-run asymmetric effects.
As always, the null hypothesis will be formulated as equalities, and
rejection would be in favor of inequality.
Then, assuming that your variables in gretl are actually named DPOS_k,
DNEG_k, POS_1, NEG_1, you type in as restrictions:
b(DPOS_k) - b(DNEG_k) = 0
b(POS_1) - b(NEG_1) = 0
Watch out that all b terms appear on the left-hand sides. (Developer
note: The error message that users get when they violate this says "b
doesn't exist" or something like this and IMO is a bit misleading.)
You could also use square brackets in this context.
(BTW, I've seen these constructed partial sums before, but I don't
remember whether it was actually OK to apply the standard regression
inference theory to them. The answer above ignores those doubts and is
just about a restriction formulation in gretl.)
cheers
sven
11:59 a.m.
Am 22.08.20 um 17:34 schrieb Sven Schreiber:
Am 22.08.2020 um 10:55 schrieb f.marchini(a)studenti.unibg.it:
> Thank you for your answer. Following the literature I adopted the
> following equation:
>
>
DLnTBt=a'+Σb'*DLnTBt-k+Σc'*DLnYt-k+Σd'*DLnYWt-k+Σe'*DPOSt-k+Σf'*DNEGt-k+θ0*LnTBt-1+θ1*LnYt-1+θ2*LnYWt-1+θ3*POSt-1+θ4*NEGt-1+εt
>
>
> where POSt and NEGt are the two variables obtained from the partial
> sum of positive and negative changes of the lnREX variable.
OK. Please try to use at least some blanks to increase the readability
of such a formula line in pseudo-code. (Of course Latex notation would
also help, but not necessary.)
(BTW, I've seen these constructed partial sums before, but I don't
remember whether it was actually OK to apply the standard regression
inference theory to them. The answer above ignores those doubts and is
just about a restriction formulation in gretl.)
Standard regression inference theory applies here too as shown by Shin
et al. They show that for both long- and short-run tests on symmetry.
The reference is:
https://link.springer.com/chapter/10.1007/978-1-4899-8008-3_9
Artur
12:56 p.m.
Am 22.08.2020 um 17:59 schrieb Artur Tarassow:
Am 22.08.20 um 17:34 schrieb Sven Schreiber:
> Am 22.08.2020 um 10:55 schrieb f.marchini(a)studenti.unibg.it:
>> Thank you for your answer. Following the literature I adopted the
>> following equation:
>>
>>
DLnTBt=a'+Σb'*DLnTBt-k+Σc'*DLnYt-k+Σd'*DLnYWt-k+Σe'*DPOSt-k+Σf'*DNEGt-k+θ0*LnTBt-1+θ1*LnYt-1+θ2*LnYWt-1+θ3*POSt-1+θ4*NEGt-1+εt
>>
>>
>> where POSt and NEGt are the two variables obtained from the partial
>> sum of positive and negative changes of the lnREX variable.
>
> OK. Please try to use at least some blanks to increase the readability
> of such a formula line in pseudo-code. (Of course Latex notation would
> also help, but not necessary.)
>
> (BTW, I've seen these constructed partial sums before, but I don't
> remember whether it was actually OK to apply the standard regression
> inference theory to them. The answer above ignores those doubts and is
> just about a restriction formulation in gretl.)
Standard regression inference theory applies here too as shown by Shin
et al. They show that for both long- and short-run tests on symmetry.
The reference is:
https://link.springer.com/chapter/10.1007/978-1-4899-8008-3_9
Strictly speaking they require a conditional ECM it seems. From
F.Marchini's formulation is not clear whether that holds. (Whether k
starts at 0.)
cheers
sven
1507
days inactive
1511
days old
7 comments
5 participants
participants (5)
-
Artur Tarassow -
Dhanasekaran Kuppusamy -
f.marchini＠studenti.unibg.it -
Riccardo (Jack) Lucchetti -
Sven Schreiber