From svetosch@gmx.net Sat Mar 20 09:30:37 2010
From: Sven Schreiber <svetosch@gmx.net>
To: gretl-devel@gretlml.univpm.it
Subject: Re: [Gretl-devel] [Gretl-users] VAR: Impulse-response
Date: Sat, 20 Mar 2010 14:29:59 +0100
Message-ID: <4BA4CDD7.4040309@gmx.net>
In-Reply-To: 4BA0CA0D.40503@gmx.net
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="===============5837016271349638755=="

--===============5837016271349638755==
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit

Sven Schreiber schrieb:
> Henrique Andrade schrieb:
>> Dear Gretl Community,
>>
>> I have a question about how Gretl calculates the VAR impulse-responses.
>> Although Gretl says that an one-standard error shock is used in the
>> calculations I can't confirm this, once another value for the standard
>> error is given in the estimation results. To illustrate this I will give
>> an example:
>>
>> <script>
>> open australia.gdt
>> "Modelo 1" <- var 1 le lpus lpau --impulse-responses
>> </script>
>>
>> The standard-error of the "Equation 1" (le) is 0.048110, but the
>> impulse-response uses 0.046827 as the standard-error. I did the same
>> estimation using EViews and, according to that software, 0.046827 is the
>> standard-error with degrees of freedom correction and 0.046827 is the
>> standard-error without degrees of freedom correction.
>>
>>
> 
> Exactly, so you answered the question yourself. I tend to agree that the
> displayed standard error should be used, and then this would be a small
> bug. (there were some changes with VARs and dof-corrections recently,
> cf. the accessors $xtxinv, $sigma, and I guess that what you saw was on
> oversight -- but maybe I'm wrong and it's done on purpose)
> 
>

[moving over to devel]

Allin, so do you think this is a bug?

thanks,
sven

--===============5837016271349638755==--