Hello all,
I am working on a research, using Panel VAR model. I am using the pvar.ado package, which has been introduced by Inessa Love (Stata package is from Inessa Love's website: https://sites.google.com/a/hawaii.ed...love/home/pvar). For the early steps of my research, I hope to run some regressions without using instruments, but could not find a way to do it. Basically, my question is: how can we run a pvar regression without using instruments (i.e. without using lagged dependent variables (t-2 or t-1) as instruments for the (t-1) RHS-variables)"?
My code is:
pvar y1 y2, exog(z)
The default (without using option - lag()) is to use the first lag of each variable in depvarlist. And the result table always has one line at the bottom:
Instruments: l(1/1).(y1 y2)
I think this means that first lag of each variable is used as instruments for themselves. I also tried not to use "exog(z)", but it does not help and still reports the same line (Instruments: l(1/1).(y1 y2)) at the bottom in the result.
I assume that without using lagged dependent variables as instruments for RHS-variables, the results from running "pvar y1 y2" should be very similar to the results from running pooled OLS "reg y1 y1_lag1 y2_lag1". And result from running "pvar y1 y2, exog(z)" should be very similar to the result from running "reg y1 z y1_lag1 y2_lag1". But results from running "pvar y1 y2" are very different from running "reg y1 y1_lag1 y2_lag1", because (I think) "pvar" is using y1-lag1 and y2_lag1 as both independent variables and instruments.
My question is very similar to an earlier question posted here in 2014 as below.
https://www.statalist.org/forums/for...l-var-pvar-ado
If I understand correctly, the 3rd question from Tim Grünebaum is very similar to my question above.
3) Maybe I did not understand the PVAR approach sufficiently, but why do we have to use lagged dependant variables (t-2) as instruments for the t-1 rhs-variables? The problem of the fixed effects in dynamic regression analysis should be solved with the forward-demeaning using "helm" shouldn't it? Couldn't we just use the variables in levels as in a normal VAR after eliminating the fixed effects?
However, after reading the post and the replies, I still don't understand how to solve the problem. I've also checked Ryan Decker's website (http://www.rdecker.net/code), suggested by Marie-Pierre in #6 replying to the post but could not find a solution.
Does anyone have a solution or suggestion on this? Thank you very much in advance.
Happy New Year!
Chenli
I am working on a research, using Panel VAR model. I am using the pvar.ado package, which has been introduced by Inessa Love (Stata package is from Inessa Love's website: https://sites.google.com/a/hawaii.ed...love/home/pvar). For the early steps of my research, I hope to run some regressions without using instruments, but could not find a way to do it. Basically, my question is: how can we run a pvar regression without using instruments (i.e. without using lagged dependent variables (t-2 or t-1) as instruments for the (t-1) RHS-variables)"?
My code is:
pvar y1 y2, exog(z)
The default (without using option - lag()) is to use the first lag of each variable in depvarlist. And the result table always has one line at the bottom:
Instruments: l(1/1).(y1 y2)
I think this means that first lag of each variable is used as instruments for themselves. I also tried not to use "exog(z)", but it does not help and still reports the same line (Instruments: l(1/1).(y1 y2)) at the bottom in the result.
I assume that without using lagged dependent variables as instruments for RHS-variables, the results from running "pvar y1 y2" should be very similar to the results from running pooled OLS "reg y1 y1_lag1 y2_lag1". And result from running "pvar y1 y2, exog(z)" should be very similar to the result from running "reg y1 z y1_lag1 y2_lag1". But results from running "pvar y1 y2" are very different from running "reg y1 y1_lag1 y2_lag1", because (I think) "pvar" is using y1-lag1 and y2_lag1 as both independent variables and instruments.
My question is very similar to an earlier question posted here in 2014 as below.
https://www.statalist.org/forums/for...l-var-pvar-ado
If I understand correctly, the 3rd question from Tim Grünebaum is very similar to my question above.
3) Maybe I did not understand the PVAR approach sufficiently, but why do we have to use lagged dependant variables (t-2) as instruments for the t-1 rhs-variables? The problem of the fixed effects in dynamic regression analysis should be solved with the forward-demeaning using "helm" shouldn't it? Couldn't we just use the variables in levels as in a normal VAR after eliminating the fixed effects?
However, after reading the post and the replies, I still don't understand how to solve the problem. I've also checked Ryan Decker's website (http://www.rdecker.net/code), suggested by Marie-Pierre in #6 replying to the post but could not find a solution.
Does anyone have a solution or suggestion on this? Thank you very much in advance.
Happy New Year!
Chenli
