Tackling Autocorrelation (Panel Least Squares)
Posted: Mon Jul 08, 2013 4:42 am
Hello,
I am using Eviews 8.
I wanted some advice on the procedures available in Eviews to tackle the problem of Autocorrelation (serial correlation) under Panel data (Least squares regression approach).
I am running a Panel (Ordinary) Least Squares regression with Period Fixed Effects.
The dependent variable is dividend payout and independent variables are proxies for agency/transaction costs. Basically, I am investigating agency and transaction cost influence on dividend policy of firms. My sample is an unbalanced panel of 84 firms across 8 years (2004-2011).
My regression equation has the following functional form:
Dividend Payout = F(Firm Size, free cash flow, Government Ownership, Leverage, Growth and Investment opportunities).
I believe I ran this regression correctly in Eviews as a Panel Least Squares regression (with Period Fixed Effects). However, the model seems to suffer from the problem of Autocorrelation in error terms, a key violation of OLS assumptions. The Durbin Watson value was rather low at 1.003.
I am now trying to tackle this problem of Autocorrelation.
I am planning to follow the approach of Generalized transformation (as shown in Gujarati, 2012). I plan to transform my original regression equation whereby I would take the following difference: (P*previous year variable values) from all my original variable values. I would first estimate the value of p (e.g. 1-DW/2). This transformation would mean some loss of sample size (1 (year) per observation).
I plan to do the transformation in Excel and then import the file in Eviews.
I would then re-run this transformed equation as a Panel (Ordinary) Least Squares regression with Period Fixed Effects.
Now I have the following query:
My original regression was run as Panel Least Squares with Period Fixed Effects.
Can the process of Generalized Transformation be applied to Panel data?
If yes, can I re-run my transformed regression equation as Panel Least Square with Period Fixed Effects?
My knowledge of Econometrics in general is low. I am not sure whether the above approach can be applied in my case.
I would be very grateful for any advice offered in this regard.
If the above approach is not valid could someone possibly suggest other ways/options available in Eviews 8 to tackle the autocorrelation problem in panel data?
Are there any choices available in Eviews 8 under GLS weights/coefficient covariance methods that are suitable for tackling Autocorrelation (serial correlation) problems in panel least squares regressions?
Thank You…
I am using Eviews 8.
I wanted some advice on the procedures available in Eviews to tackle the problem of Autocorrelation (serial correlation) under Panel data (Least squares regression approach).
I am running a Panel (Ordinary) Least Squares regression with Period Fixed Effects.
The dependent variable is dividend payout and independent variables are proxies for agency/transaction costs. Basically, I am investigating agency and transaction cost influence on dividend policy of firms. My sample is an unbalanced panel of 84 firms across 8 years (2004-2011).
My regression equation has the following functional form:
Dividend Payout = F(Firm Size, free cash flow, Government Ownership, Leverage, Growth and Investment opportunities).
I believe I ran this regression correctly in Eviews as a Panel Least Squares regression (with Period Fixed Effects). However, the model seems to suffer from the problem of Autocorrelation in error terms, a key violation of OLS assumptions. The Durbin Watson value was rather low at 1.003.
I am now trying to tackle this problem of Autocorrelation.
I am planning to follow the approach of Generalized transformation (as shown in Gujarati, 2012). I plan to transform my original regression equation whereby I would take the following difference: (P*previous year variable values) from all my original variable values. I would first estimate the value of p (e.g. 1-DW/2). This transformation would mean some loss of sample size (1 (year) per observation).
I plan to do the transformation in Excel and then import the file in Eviews.
I would then re-run this transformed equation as a Panel (Ordinary) Least Squares regression with Period Fixed Effects.
Now I have the following query:
My original regression was run as Panel Least Squares with Period Fixed Effects.
Can the process of Generalized Transformation be applied to Panel data?
If yes, can I re-run my transformed regression equation as Panel Least Square with Period Fixed Effects?
My knowledge of Econometrics in general is low. I am not sure whether the above approach can be applied in my case.
I would be very grateful for any advice offered in this regard.
If the above approach is not valid could someone possibly suggest other ways/options available in Eviews 8 to tackle the autocorrelation problem in panel data?
Are there any choices available in Eviews 8 under GLS weights/coefficient covariance methods that are suitable for tackling Autocorrelation (serial correlation) problems in panel least squares regressions?
Thank You…