topic Re: Inconsistent Results of 2SLS Estimation (proc reg and proc syslin) in Statistical Procedures

Inconsistent Results of 2SLS Estimation (proc reg and proc syslin)

angelvisa — Fri, 19 Aug 2016 19:45:54 GMT

Hello Everyone,

I found a potential inconsistent estimation or am I missing anything here? I was estimating a 2SLS model with just-identify number of instrumental variables (SAS 9.3 M2). I first use :

proc reg data=allindq outest=a;
model  maxe = x disper2 rev me btm tradvol Alpha car ar ivmaxe;
quit;

I also ran :

proc syslin data=allindq 2sls first;
endogenous maxe;
instruments x disper2 rev me btm tradvol Alpha car ar ivmaxe;
model y=  x disper2 rev me btm tradvol Alpha car ar maxe;
run;

However, the first stage results from proc syslin and the results from proc reg are very different (different from each other from 0.01).

In theory, the first stage of 2sls is an ols. Therefore the proc reg should give the same results as the first stage of proc syslin. However it didnt happen.

Could anybody please help with this? Thank you!

Re: Inconsistent Results of 2SLS Estimation (proc reg and proc syslin)

angelvisa — Fri, 19 Aug 2016 19:48:42 GMT

Btw, the endogous variable is maxe, the exogenous variables are x disper2 rev me btm tradvol Alpha car ar ; the IV is ivmaxe

Re: Inconsistent Results of 2SLS Estimation (proc reg and proc syslin)

ballardw — Fri, 19 Aug 2016 21:15:41 GMT

Different procs often use slightly different approximations or assumptions. So I would not be surprised to see different results.

For extra added fun try changing the order of independent variables in a model, especially when using one of the selection methods. See if the same variables are kept or if the parameters remain the same.