Systemfit package/Autocorrelation

[Axel Leroix]

 
Hello,
 
I have question with regrad to the "systemfit" package.
I'm estimating a simultaneous system of equations. I have 5 equations in my system. I use 3SLS for estimation since it permits to take into account, both, contemporaneous error terms correlation and simultaneity bias.
 
After estimating my system, in which all equations are dynamic (each equation contain one lagged endogenous variable), I perform the Breusch-Godfrey test to test for the presence of residual autocorrelation between error terms of the same equation (E_1t ,E_1(t-1). The test results show that there is autocorrelation.
 
My question is, in the case of simultaneous system of equations, how to correct for this problem of autocorrelation (I note again that I speak about autocorrelation between erros of the same equation and not autocorrelation between errors of different equations).
 
I have seen that on the case for one single equation estimation, we use the function gls to correct for residuals autocorrelation, but in the case od system of equations there is no indication about how deal with this problem.  I wonder if I should estimate separately equations in which there is evidence of autocorrelation by using simply OLS or gls functions?
 
Any idea please ?
Help will be very appreciated.
Thank you in advance
 


     
        [[alternative HTML version deleted]]


_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Peter Carl-2]

You might use autocorrelation-adjusted returns, using the 'Return.Geltner'
function in PerformanceAnalytics.  That adjusts for first-order
autocorrelation.

Okunev and White propose a method for removing n-order autocorrelation in:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=460641
... and if anyone is willing to contribute a function that implements it,
we'd be interested in including it in a future version of PA.

HTH,

pcc
--
Peter Carl
http://www.braverock.com/~peter

>  
> Hello,
>  
> I have question with regrad to the "systemfit" package.
> I'm estimating a simultaneous system of equations. I have 5 equations in
> my system. I use 3SLS for estimation since it permits to take into
> account, both, contemporaneous error terms correlation and simultaneity
> bias.
>  
> After estimating my system, in which all equations are dynamic (each
> equation contain one lagged endogenous variable), I perform the
> Breusch-Godfrey test to test for the presence of residual autocorrelation
> between error terms of the same equation (E_1t ,E_1(t-1). The test results
> show that there is autocorrelation.
>  
> My question is, in the case of simultaneous system of equations, how to
> correct for this problem of autocorrelation (I note again that I speak
> about autocorrelation between erros of the same equation and not
> autocorrelation between errors of different equations).
>  
> I have seen that on the case for one single equation estimation, we use
> the function gls to correct for residuals autocorrelation, but in the case
> od system of equations there is no indication about how deal with this
> problem.  I wonder if I should estimate separately equations in which
> there is evidence of autocorrelation by using simply OLS or gls functions?
>  
> Any idea please ?
> Help will be very appreciated.
> Thank you in advance
>  
>
>
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-SIG-Finance@... mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only.
> -- If you want to post, subscribe first.

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[ezivot]

You can treat the autocorrelation as nuisance parameters and correct for its
effect on the standard errors by estimating the model by GMM (see the nice
gmm pagckage for this) and using a heteroskedasticity and autocorrelation
consistent (HAC) covariance matrix.


-----Original Message-----
From: r-sig-finance-bounces@...
[mailto:r-sig-finance-bounces@...] On Behalf Of Axel Leroix
Sent: Friday, November 06, 2009 8:57 AM
To: r-sig-finance@...
Subject: [R-SIG-Finance] Systemfit package/Autocorrelation

 
Hello,
 
I have question with regrad to the "systemfit" package.
I'm estimating a simultaneous system of equations. I have 5 equations in my
system. I use 3SLS for estimation since it permits to take into account,
both, contemporaneous error terms correlation and simultaneity bias.
 
After estimating my system, in which all equations are dynamic (each
equation contain one lagged endogenous variable), I perform the
Breusch-Godfrey test to test for the presence of residual autocorrelation
between error terms of the same equation (E_1t ,E_1(t-1). The test results
show that there is autocorrelation.
 
My question is, in the case of simultaneous system of equations, how to
correct for this problem of autocorrelation (I note again that I speak about
autocorrelation between erros of the same equation and not autocorrelation
between errors of different equations).
 
I have seen that on the case for one single equation estimation, we use the
function gls to correct for residuals autocorrelation, but in the case od
system of equations there is no indication about how deal with this problem.
I wonder if I should estimate separately equations in which there is
evidence of autocorrelation by using simply OLS or gls functions?
 
Any idea please ?
Help will be very appreciated.
Thank you in advance
 


     
        [[alternative HTML version deleted]]

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.