We have recently been exploring MiDaS in EViews, it is a really great feature. Also, the MiDaS webinar was superb.
However, I have a minor problem with the Almon/PDL weighting in the MiDaS estimation. The issue is described in the attached PDF.
Also, when determining the degree of the Almon polynomial, I miss an estimated parameter. Suppose that I choose polynomial degree 3. Then I would expect to get 4 estimated parameters (the a_i:s) for the polynomial: a_0+a_1*X+a_2*X^2+a_3*X^3. But only three estimated parameters are presented in the output. I suspect that a second degree polynomial is estimated when setting polynomial degree to 3: a_0+a_1*X+a_2*X^2. Is this so or is it a restricted (a_0=0) third degree polynomial (a_1*X+a_2*X^2+a_3*X^3) that is estimated?
Best,
Kristian
Almon/PDL in MiDaS
Moderators: EViews Gareth, EViews Moderator
Almon/PDL in MiDaS
 Attachments

 AlmonLagsEViews.pdf
 (84.61 KiB) Downloaded 105 times

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11983
 Joined: Tue Sep 16, 2008 5:38 pm
Re: Almon/PDL in MiDaS
Here's an EViews program that manually computes an example MIDAS model, and shows the weighting calculations:
Code: Select all
!xlags = 4
!ylags=1
!horizon=2
!polyorder = 3 ' matlab order +1
wfopen(wf=midas, page=quarterly) .\mydata.xlsx range="Sheet1"
pageload(page=monthly) .\mydata.xlsx range="Sheet2"
series x1 = dlog(value(!horizon))*100
for !i=2 to !xlags
series x!i = x1(!i+1)
next
for !i=1 to !xlags
series z!i=x!i(2)
next
pageselect quarterly
series y = dlog(value)*100
smpl 1985 2009/1/1
for !i=1 to !xlags
copy(c=l) monthly\z!i x!i
next
for !i=1 to !polyorder
series z!i=0
next
for !i=1 to !xlags
for !j=1 to !polyorder
z!j = z!j + !i^(!j1)*x!i
next
next
group xlags z*
group ylags
for !i=1 to !ylags
ylags.add y(!i)
next
equation eq2.ls y c xlags ylags
show eq2
equation eq1.midas(freq=first) y c y(1) @ monthly\x1
show eq1
 Attachments

 mydata.xlsx
 (47.65 KiB) Downloaded 78 times
Follow us on Twitter @IHSEViews
Re: Almon/PDL in MiDaS
It is exectly this I am after.
Estimations in EViews and the MiDaS toolbox for Matlab use the weighting
for !i=1 to !xlags
for !j=1 to !polyorder
z!j = z!j + !i^(!j1)*x!i
next
next
In the documentation for the MiDaS Matlab toolbox (see equation 2.10 in the documentation for version 2.1 as of August 3, 2016), in the documentation for EViews 9.5 (Equation 28.6) and in Intriligator (1978) Econometric models, techniques, and applications, page 183, equation 6.7.28, the weighting described for the Almon/PDL case is
for !i=1 to !xlags
for !j=1 to !polyorder
z!j = z!j + (!i1)^(!j1)*x!i
next
next
This might of course be a minor difference (although the curvature of the estimated parameters seems to be affected), but if I understand you correctly it is the documentation that is the candidate to be changed and not the estimation routine?
Estimations in EViews and the MiDaS toolbox for Matlab use the weighting
for !i=1 to !xlags
for !j=1 to !polyorder
z!j = z!j + !i^(!j1)*x!i
next
next
In the documentation for the MiDaS Matlab toolbox (see equation 2.10 in the documentation for version 2.1 as of August 3, 2016), in the documentation for EViews 9.5 (Equation 28.6) and in Intriligator (1978) Econometric models, techniques, and applications, page 183, equation 6.7.28, the weighting described for the Almon/PDL case is
for !i=1 to !xlags
for !j=1 to !polyorder
z!j = z!j + (!i1)^(!j1)*x!i
next
next
This might of course be a minor difference (although the curvature of the estimated parameters seems to be affected), but if I understand you correctly it is the documentation that is the candidate to be changed and not the estimation routine?

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11983
 Joined: Tue Sep 16, 2008 5:38 pm
Re: Almon/PDL in MiDaS
Hi.
I post my question here as it seems the most appropiate thread.
I am using MiDaS in Eviews 10.
With PDL/ALmon weighting, if I use equation 28.5 in the Eviews documention (which is consistent with all Ghysels papers and his Matlab toolbox documentation), I cannot replicate the lag weights in the output. Please see Excel file. In order to obtain the reported lag weights I have to divide the estimates according to equation 28.5 (column (1) ) by the lag number. I cannot find a reference to this variation of Almon weighting. Is this correct and should the documentation be modified?
Thanks.
I post my question here as it seems the most appropiate thread.
I am using MiDaS in Eviews 10.
With PDL/ALmon weighting, if I use equation 28.5 in the Eviews documention (which is consistent with all Ghysels papers and his Matlab toolbox documentation), I cannot replicate the lag weights in the output. Please see Excel file. In order to obtain the reported lag weights I have to divide the estimates according to equation 28.5 (column (1) ) by the lag number. I cannot find a reference to this variation of Almon weighting. Is this correct and should the documentation be modified?
Thanks.
 Attachments

 Example_Midas.xlsx
 (23.84 KiB) Downloaded 42 times

 Example_Midas.WF1
 (50.23 KiB) Downloaded 34 times
Who is online
Users browsing this forum: Baidu [Spider] and 4 guests