logit standard error

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

m9111
Posts: 2
Joined: Fri Jul 27, 2018 3:56 am

logit standard error

Postby m9111 » Tue Jul 31, 2018 1:53 am

Hello,

Could you please assist with the following query.

I have estimated a state space and used a probit function to constrain the coefficient c(1) to below 1. I understand that the estimate of c(1) must be transformed in order to back out beta, using the following transformation: beta=(1/(1+exp(c(1))).

What I am not sure is whether the same transformation can be applied to transform the standard errors. I found that when doing this, the transformed standard errors are very small, even though the coefficients are insignificant, so I suspect this approach may not be correct. Could you please advise on how to back out the standard errors?

Many thanks in advance!

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

Re: logit standard error

Postby startz » Tue Jul 31, 2018 10:16 am

It’s a touch messy. Basically var(f(b)) = f’(b)var(b)f’(b), where f’ means first derivative.

Probably the easiest way to get the standard error is to look under coefficient,views,wald test and test

Code: Select all

1/(1+exp(c(1)))=0

m9111
Posts: 2
Joined: Fri Jul 27, 2018 3:56 am

Re: logit standard error

Postby m9111 » Mon Aug 13, 2018 8:29 am

Thanks for the advice. I have tried this approach but have run into an issue - it produces a standard error which seems very low, considering the coefficient is insignificant.

The results show that: beta=0.99; SE=0.087, p-value = 0.74

Since beta has been constrained to below 0.95, the p-value result seems fine, but the low standard error doesn’t appear credible. Is there any other way the standard error can be transformed within Eviews to double check the result?

I attach the workfile. Please see tab ‘2024’ > state space object ‘ss_0’ > results for c(1) and saved ‘wald_test’ as per your suggestion.
model_a.WF1
(92.35 KiB) Downloaded 221 times

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

Re: logit standard error

Postby startz » Mon Aug 13, 2018 8:42 am

I think you've done everything correctly. But you may be misinterpreting significance.

Suppose c(1) is almost exactly zero, so not "significant." Then

1/(1+exp(c(1))) = 1/(1+e^0)=1/(1+1) = 0.5

which is significantly different from zero.


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 35 guests