2:32 a.m.
Dear All,
I wonder, if there's any way to plot final error variance decomposition
(FEVD) output, produced after VAR or VECM, in gretl?
So far, the only way I know is to export data to GNU R and plot FEVD there.
This approach, however, has its disadvantages, as one has to estimate the
whole system in R first and R's estimates often differ significantly from
those obtained with gretl.
Many thanks in advance!
Greg
12:01 p.m.
On Sun, 27 Mar 2011, Grzegorz Konat wrote:
Dear All,
I wonder, if there's any way to plot final error variance decomposition (FEVD)
output, produced after VAR or
VECM, in gretl?
At the moment, only with a little scripting. Here's an example, which I'm
sure you can adapt to your needs quite easily.
<script>
function matrix SVMA(matrix comp, matrix Sigma, scalar horizon)
scalar n = rows(Sigma)
scalar np = rows(comp)
matrix K = cholesky(Sigma)
matrix C = K
matrix tmp = I(np)
matrix ret = zeros(horizon, n*n)
loop for i=1..horizon --quiet
ret[i,] = vec(C)'
tmp = comp * tmp
C = tmp[1:n,1:n] * K
end loop
return ret
end function
function matrix FEVD(matrix SVMA)
scalar n2 = cols(SVMA)
scalar n = sqrt(n2)
scalar h = rows(SVMA)
matrix ret = zeros(h, n2)
matrix cC = zeros(n, n)
loop for i=1..h --quiet
matrix C = mshape(SVMA[i,], n, n)
matrix tmp = C.*C
cC += tmp
tmp = cC ./ sumr(cC)
ret[i,] = vec(tmp')'
end loop
return (seq(0,h-1)') ~ ret
end function
# ------- main ---------------------------------------------
open sw_ch14.gdt
series infl = 400*ldiff(PUNEW)
var 4 LHUR infl
X = SVMA($compan, $sigma, 20)
fevd = FEVD(X)
gnuplot 2 3 1 --matrix=fevd --output=display --with-lines
gnuplot 4 5 1 --matrix=fevd --output=display --with-lines
</script>
So far, the only way I know is to export data to GNU R and plot FEVD
there. This approach, however, has its disadvantages, as one has to
estimate the whole system in R first and R's estimates often differ
significantly from those obtained with gretl.
This is more worrying. The same model should produce the same estimates
both in gretl and R. Could you provide us with an example?
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
1:06 p.m.
2011/3/28 Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it>
So far, the only way I know is to export data to GNU R and plot FEVD
> there. This approach, however, has its disadvantages, as one has to estimate
> the whole system in R first and R's estimates often differ significantly
> from those obtained with gretl.
>
This is more worrying. The same model should produce the same estimates
both in gretl and R. Could you provide us with an example?
First of all, thank you so much for the script - works perfectly fine for
me!
As far as the issue with different estimates is concerned, this is probably
all my fault (I'm quite new to gretl and completely new to R), but still
there's a slight chance that something might be wrong.
In gretl I do:
vecm 2 1 X Y Z --ct
The above produces exactly the same output as STATA's:
vec x y z, trend(trend) /vecrank and lag order given in specification
window/
and slightly different form JMulTi's estimates.
However, when I use R's (urca):
xyz.vecm <- ca.jo(gretldata[, c("X", "Y", "Z")], type =
"trace", ecdet =
"trend", K = 2, spec = "transitory") /for all options in
'type' and
'spec'/
and transform obtained VECM into VAR representation:
vec2var.xyz <- vec2var(xyz.vecm, r = 1)
the system I get has quite different parameters from those estimeted in
gretl, STATA or even JMulTi (obviously I do not compare VAR to VECM - there
are diffreneces in VAR representations).
Still, this might be some kind of my mistake, so if you see anything wrong
in the code I presented above, please let me know.
Best,
Greg
5:26 p.m.
Am 28.03.2011 15:06, schrieb Grzegorz Konat:
In gretl I do:
vecm 2 1 X Y Z --ct
The above produces exactly the same output as STATA's:
vec x y z, trend(trend) /vecrank and lag order given in specification
window/
and slightly different form JMulTi's estimates.
However, when I use R's (urca):
xyz.vecm <- ca.jo <http://ca.jo>(gretldata[, c("X", "Y",
"Z")], type =
"trace", ecdet = "trend", K = 2, spec = "transitory")
/for all options
in 'type' and 'spec'/
The usual suspects are: double-check the sample which is used, the lag
order (seems to match here though), and the precise specification of
deterministic terms (both restricted and unrestricted). Don't know the
details of urca to tell you right now.
-sven
6:07 p.m.
2011/3/28 Sven Schreiber <svetosch(a)gmx.net>
The usual suspects are: double-check the sample which is used, the lag
order (seems to match here though), and the precise specification of
deterministic terms (both restricted and unrestricted). Don't know the
details of urca to tell you right now.
-sven
The sample and lag order are correct for sure. My choice for deterministic
terms is "unrestricted constant and unrestricted trend". This might be the
issue, as in gretl and STATA you can specify this particular option (aka
'case 5'), while in JMulTi and R the specification for deterministic terms
seems to be more general and I'm not quite sure whether ecdet="trend" in
urca is an exact equivalent of "--ct" in gretl... Is it?
6:18 p.m.
Am 28.03.2011 20:07, schrieb Grzegorz Konat:
The sample and lag order are correct for sure. My choice for
deterministic terms is "unrestricted constant and unrestricted trend".
This might be the issue, as in gretl and STATA you can specify this
particular option (aka 'case 5'), while in JMulTi and R the
specification for deterministic terms seems to be more general and I'm
not quite sure whether ecdet="trend" in urca is an exact equivalent of
"--ct" in gretl... Is it?
well it _sounds_ more like what in gretl would be --crt, so maybe just
do a cross-check with that gretl option.
(The case --ct [in gretl terms] is quite strange anyway, it would be
understandable if urca didn't implement it.)
-sven
7:06 p.m.
2011/3/28 Sven Schreiber <svetosch(a)gmx.net>
Am 28.03.2011 20:07, schrieb Grzegorz Konat:
well it _sounds_ more like what in gretl would be --crt, so maybe just
do a cross-check with that gretl option.
(The case --ct [in gretl terms] is quite strange anyway, it would be
understandable if urca didn't implement it.)
-sven
That's what R's Help says about it:
"‘none’ for no intercept in cointegration, ‘const’ for constant term in
cointegration and ‘trend’ for trend variable in cointegration."
4781
days inactive
4782
days old
6 comments
3 participants
participants (3)
-
Grzegorz Konat -
Riccardo (Jack) Lucchetti -
Sven Schreiber