Putting limits on estimated coefficient values

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

jons
Posts: 3
Joined: Mon Oct 22, 2012 6:11 am

Re: Putting limits on estimated coefficient values

Postby jons » Mon Oct 22, 2012 6:49 am

I’m currently trying to estimate a model on panel data that should look like
y = c(1) + c(2)*x + c(3)*(composite)
with composite being the "best" weighted average of 4 indicators (a, b, d, e). Thus, compo should be of the form
compo = c(4)*a + c(5)*b + c(6)*d +c(7)*e
with 0 < c(4), c(5), c(6), c(7) < 1 and c(4)+c(5)+c(6)+c(7) = 1
I’ve looked at the above but it does not seem to work, results are quite strange… Any idea ?

ele
Posts: 1
Joined: Thu Nov 07, 2013 9:24 am

Re: Putting limits on estimated coefficient values

Postby ele » Thu Nov 07, 2013 10:02 am

Hallo,

I have tried to use the logistic transformation, i.e. (H-L)*@logit(c(1)) + L, to estimate a two-sided restricted parameter within a State Space framework, but I always get the same error message "WARNING: Singular covariance - coefficients are not unique". More specifically

{%model}_ss.add @STATE sv1 = sv1(-1)+ (0.4*@logit({%theta1})+0.3)*d(ltu)+[VAR = {%varstate}]

where H=0.7, L=0.3 and %varstate is equal to an assigned specific value.
The same error message is produced also in case of the one-sided restriction transformation (0.7-exp({%theta1}))*d(ltu). It is not necessarily linked to the weakness of SS models to treat nonlinearities because both the logistic and the exponential transformations can be correctly solved. Hence, I was wondering what is the reason and the solution.
Thanks in advance.

trubador
Did you use forum search?
Posts: 1518
Joined: Thu Nov 20, 2008 12:04 pm

Re: Putting limits on estimated coefficient values

Postby trubador » Fri Nov 08, 2013 1:15 am

ele wrote:I was wondering what is the reason and the solution.

Hard to say without seeing the workfile. But you should consider the possibility that data strongly disagrees with your constraint.

kruupy
Posts: 11
Joined: Thu Aug 14, 2014 11:38 pm

Re: Putting limits on estimated coefficient values

Postby kruupy » Wed Aug 27, 2014 4:56 am

HI All,

I would like to put a restriction on one of the coefficients that it is equal to 1.

So Y1 = a + b Y2 + c X2 where b=1.

But what I was wondering is, if I run the regression

Y1-Y2 = a + c X2

Does this regression above impose the restriction that b=1? - Im pretty sure it does, but it would be great if I could get some confirmation.

Cheers,
Andrew.

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Putting limits on estimated coefficient values

Postby startz » Wed Aug 27, 2014 6:21 am

You got it right.

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Tue Oct 07, 2014 7:39 am

Hi All,

Is it possible to force c(1) = 1 in the following equation, and how could one do it?

Y = c(1)*Y(-1)*(X/X(-1))

Thanks!

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13306
Joined: Tue Sep 16, 2008 5:38 pm

Re: Putting limits on estimated coefficient values

Postby EViews Gareth » Tue Oct 07, 2014 7:41 am

If you do that there is nothing left to estimate...
Follow us on Twitter @IHSEViews

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Tue Oct 07, 2014 8:04 am

Right. But if I do the following,

sample large 2014q2 2024q2
smpl large
series X = (@pch(Y)+1)*X(-1)

it performs the calculation for the first time period. How do I get it to perform this calculation for all other periods?

Thank you so much!

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Putting limits on estimated coefficient values

Postby startz » Tue Oct 07, 2014 8:13 am

If you are generating the variable Y, rather than running an estimate, try

Code: Select all

series Y = c(1)*Y(-1)*(X/X(-1))


taking some care about the sample to avoid NAs from being propagated.

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Tue Oct 07, 2014 8:55 am

Oh yes, thank you!

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Thu Oct 23, 2014 1:16 pm

Hi, this still doesn't work for some reason. Please help!

I am trying to do something simple. I have quarterly series, call it X, for period 2014q1 to 2018q4. I would like to make series Y by using series X's percent change. I only have one observation on Y, 2014q1.

I set my sample to 2014q1 - 2018q4.

If I try to do this

series y = (x/x(-1))*y(-1), I get only one value for y for 2014q2, the rest is NA. The same happens when I use 'genr' instead of 'series'.

What am I doing wrong?

Thanks so much!

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Putting limits on estimated coefficient values

Postby startz » Thu Oct 23, 2014 3:05 pm

You may want to post your workfile.

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Fri Oct 24, 2014 5:54 am

Thank you so much! Here is the workfile.
Attachments
example.wf1
(9.05 KiB) Downloaded 502 times

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Putting limits on estimated coefficient values

Postby startz » Fri Oct 24, 2014 6:10 am

set your sample to

Code: Select all

smpl 2014q2 2018q1

ecardamone
Posts: 40
Joined: Wed Sep 17, 2014 12:23 pm

Re: Putting limits on estimated coefficient values

Postby ecardamone » Fri Oct 24, 2014 6:17 am

Oh, I can't thank you enough! That worked. Thanks, again.


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 38 guests