Dynamic panel with xtregar AR(1) disturbance

[Giorgio Di Stefano]

Dear all,

I am trying to estimate dynamic panel data on GDP growth rates with a large T, far greater than N. The estimation method is xtregar. The model follows an AR(1) disturbance.
I have a quadratic equation, and my coefficients are all linear. In the model, I have a bunch of dummies, and categorical variables, among others. Macro variables are in logs or percentages, and data are stationary.


My model is specified as

 
 (yi,t − yi,t−1 )= (α − 1)yi,t−1 + β1yi,t−1 + β2T yi,t−1 + β3T 2yi,t−1 + yi, t+Σ1Nβ4+nyi,t + +β5Ζi,t+Di,t+μt+ηi,+εi,t

Where Y is the dependent variable of interest, GDP at level, Z is a set of control variables, and D is a set of dummy and categorical variables. T is an operator for some kind of indicators, continuous indicator variables, not binary, allowed to interact with them.


I am studding the impact on the calculated GDP growth, so they are entering it as data. (y_i,t − y_i,t−1 ) they have been calculated, already. GDP % growth is the dependent variable, and the lag of GDP at level enters the equation as a predictor variable. T is an indicator, and an overall constant a ((1-a)yt-1) in my model is included. Z is a vector of control variables that affect the level of GDP.


I have some trivial doubts about xtregar syntax:


1. Because in xtregar there is an AR1 disturbance, do I need to put in the syntax the lag operator for Y in my model ,gdp at level, or is the variable by itself enough?


The code I run is

xtregar, gdpgrowth,  gdplevel and other macro variables dummies. c.indicator#c.indicator, fe

should it look like that or

xtregar, gdpgrowth,  L. gdplevel and other macro variables dummies. c.indicator#c.indicator, fe

Risking in that case to case two lags instead of one?


2. Also, since there is a term (α − 1)y_i,t−1 to be entered in the equation, how is that added with Lincom in order to get the results in a nice table? How do I get the lincom to calculate (a-1), which will appear in a table with esttab? .

3. Also, since I am interested in margins will I have to change the margins code to take into account those calculations for

(α − 1)yi,t−1.  

4. How are the Baltagi-Wu LBI test reported with estab and when to perform two-step estimate of correlation?

[Mario Ferri]

I think Carlo Lazzaro and prof Jeff Wooldridge could be of your help. I am tagging them in order to have them notified on your inquiry.

[Carlo Lazzaro]

Provided that I'm, at best, 10,000 miles behind Jeff Wooldridge , I wonder if -xtregar. is suitable for T>N dynamic panel datasets.

[Giorgio Di Stefano]

Provided that I'm, at best, 10,000 miles behind Jeff Wooldridge , I wonder if -xtregar. is suitable for T>N dynamic panel datasets.

Thanks for reaching out to my question. I have tried several ways, like ivreghdfe and xtscc, with Driscoll-Kray errors with an AR (1) approach . I am interested in margins. Given primary that this is growth model, doing background engeniring, xtregar is the only way where I get the concavity condition with normal actual pairs of margins numbers for the x and y variables . As it should be in a growth model.
In the other cases of ivreghdfe and xtscc I get sharp convex and quasi convex pair of margins for x as and y, with unexpected values for y variable. Hard to explain from theory and data. Additionally, only with xtregar I get statistically significant results, while in the other cases are minimum or none.

Kindly, would you have a way to go on points 2 to 4 of #1?
Finally, I´m not sure if xtregar correct heteroskedascity and autocorrelation(the late does it with an AR1 disturbance)

[Carlo Lazzaro]

Giorgio:
sorry, but I'm not familiar with the models you're dealing with.

[Giorgio Di Stefano]

Dear all,

I am trying to estimate dynamic panel data on GDP growth rates with a large T, far greater than N. The estimation method is xtregar. The model follows an AR(1) disturbance.
I have a quadratic equation, and my coefficients are all linear. In the model, I have a bunch of dummies, and categorical variables, among others. Macro variables are in logs or percentages, and data are stationary.


My model is specified as

(yi,t − yi,t−1 )= (α − 1)yi,t−1 + β1yi,t−1 + β2T yi,t−1 + β3T 2yi,t−1 + yi, t+Σ1Nβ4+nyi,t + +β5Ζi,t+Di,t+μt+ηi,+εi,t

Where Y is the dependent variable of interest, GDP at level, Z is a set of control variables, and D is a set of dummy and categorical variables. T is an operator for some kind of indicators, continuous indicator variables, not binary, allowed to interact with them.


I am studding the impact on the calculated GDP growth, so they are entering it as data. (y_i,t − y_i,t−1 ) they have been calculated, already. GDP % growth is the dependent variable, and the lag of GDP at level enters the equation as a predictor variable. T is an indicator, and an overall constant a ((1-a)yt-1) in my model is included. Z is a vector of control variables that affect the level of GDP.


I have some trivial doubts about xtregar syntax:


1. Because in xtregar there is an AR1 disturbance, do I need to put in the syntax the lag operator for Y in my model ,gdp at level, or is the variable by itself enough?


The code I run is

xtregar, gdpgrowth, gdplevel and other macro variables dummies. c.indicator#c.indicator, fe

should it look like that or

xtregar, gdpgrowth,  L. gdplevel and other macro variables dummies. c.indicator#c.indicator, fe

Risking in that case to case two lags instead of one?


2. Also, since there is a term (α − 1)y_i,t−1 to be entered in the equation, how is that added with Lincom in order to get the results in a nice table? How do I get the lincom to calculate (a-1), which will appear in a table with esttab? .

3. Also, since I am interested in margins will I have to change the margins code to take into account those calculations for

(α − 1)yi,t−1. 

4. How are the Baltagi-Wu LBI test reported with estab and when to perform two-step estimate of correlation?

I have a mistake in the equation variables in #1. Where I state gpd level it is meant to be gdp growth and should be L. gdpgrowth instead of L. gdplevel there and everywhere . Just realized the mistake. My apologies.
I am asking here how to calculate the (a-1)term I'm having here with lincom that will allow to report in a table. Thank you

[Giorgio Di Stefano]

This is an update on the thread. I found out that

xtssc

performs way better with respect to

xtregar.

Always check your summary statistics and your scatters. Data speaks to you. Confirmed XTSSC later in the action!