## Imposition of zero lower boundary in endogenous variable

For technical support, tips and tricks, suggestions, or any other information regarding the EViews model object.

Moderators: EViews Gareth, EViews Moderator

dino55
Posts: 1
Joined: Fri Jan 23, 2009 9:24 am

### Imposition of zero lower boundary in endogenous variable

Hi
I'm wondering whether there is a easy way to establish a zero lower boundary in an endogenous model
variable, e.g. interest rates cannot go below zero?
I tried to work in the equation with the function @recode, but I always got an error message.
Thanks for any hints.

startz
Non-normality and collinearity are NOT problems!
Posts: 3517
Joined: Wed Sep 17, 2008 2:25 pm

### Re: Imposition of zero lower boundary in endogenous variable

dino55 wrote:Hi
I'm wondering whether there is a easy way to establish a zero lower boundary in an endogenous model
variable, e.g. interest rates cannot go below zero?
I tried to work in the equation with the function @recode, but I always got an error message.
Thanks for any hints.

For some purposes you might try modeling log(y).

jlbrillet
Posts: 39
Joined: Mon Feb 16, 2009 6:03 am

### Re: Imposition of zero lower boundary in endogenous variable

Hello,

You could add an equation creating a new variable x as :

x = y *(y>0)

and use x as the endogenous.

Take care,

Jean Louis Brillet.

Actually, the problem is when your model uses the logarithm of the variable. In this case it needs to be strictly positive, and probably different enough from zero (to avoid large derivatives). The problem lies with the "enough".

startz
Non-normality and collinearity are NOT problems!
Posts: 3517
Joined: Wed Sep 17, 2008 2:25 pm

### Re: Imposition of zero lower boundary in endogenous variable

jlbrillet wrote:Hello,

You could add an equation creating a new variable x as :

x = y *(y>0)

and use x as the endogenous.

Take care,

Jean Louis Brillet.

Actually, the problem is when your model uses the logarithm of the variable. In this case it needs to be strictly positive, and probably different enough from zero (to avoid large derivatives). The problem lies with the "enough".

If you do this, you may need to think about estimation with a Tobit rather than a standard regression model.

jlbrillet
Posts: 39
Joined: Mon Feb 16, 2009 6:03 am

### Re: Imposition of zero lower boundary in endogenous variable

I thought you meant in a model identity. Of course if the equation was estimated the problem is different.

But if you want the variable to be positive it is probably that it should be so by nature (like the nominal interst rate, or GDP), which means it is positive on the sample period.