## 2SLS regression - multiple endogenous variable

I have to perform a 2 SLS regression with multiple endogenous variable.

1st endogenous variable: a

Instrument variable for 1st endogenous variable: b c d

2nd endogenous variable: p

Instrument variable for 2nd endogenous variable: q r s

My final model: y = a p b c d q r s.

is there any simple way to do it?

for example using proc syslin or something like that ?

1 ACCEPTED SOLUTION

Accepted Solutions

## Re: 2SLS regression - multiple endogenous variable

Typically, when there is only one dependent variable in your model you use all the instrumental variables to instrument both endogenous variables; however, if you want to limit which instrumental variables are used to correct for the endogeneity of each endogenous explanatory variable you could use the following:

proc tmodel data=yourdata;
eq.a = y - ka*a;
eq.p = y - kp*p;
fit a p / ols 2sls;
instruments (a, b c d) (p, q r s);
quit;

3 REPLIES 3

## Re: 2SLS regression - multiple endogenous variable

In both PROC MODEL and its multithreaded replacement, PROC TMODEL, you can estimate your model using two stage least squares with the following statements:

proc tmodel data=yourdata;
y = ka*a + kp*p;
fit y / 2sls;
instruments b c d q r s;
quit;

Let me know if you need any further clarifications on how to use PROC (T)MODEL for your problem.

## Re: 2SLS regression - multiple endogenous variable

but how do i assign different instrumental variables to different endogenous variables? like how to separate them to their respective endogenous variable ?

## Re: 2SLS regression - multiple endogenous variable

Typically, when there is only one dependent variable in your model you use all the instrumental variables to instrument both endogenous variables; however, if you want to limit which instrumental variables are used to correct for the endogeneity of each endogenous explanatory variable you could use the following:

proc tmodel data=yourdata;
eq.a = y - ka*a;
eq.p = y - kp*p;
fit a p / ols 2sls;
instruments (a, b c d) (p, q r s);
quit;

Discussion stats
• 3 replies
• 2035 views
• 0 likes
• 2 in conversation