Download as pdf or txt
Download as pdf or txt
You are on page 1of 35

If the absence of IQ: cannot find a good proxy; cannot observe, etc.

Instrumental variable: Introduction 1


Instrumental variable: Introduction 2
Example of instruments
Example of instruments
Test the validity
Test the validity
To estimate
Option 1: Run OLS twice:
*A long way, and standard errors are not efficient*

First-stage:

෢ 𝑖 = 𝜋ෟ
𝑋1𝑖 = 𝜋1 + 𝜋2 𝑍1𝑖 + 𝜋3 𝑍2𝑖 +𝜋4 𝑊1𝑖 +𝑣𝑖 ➔ 𝑋1 1+𝜋
ෞ2 𝑍11 + 𝜋
ෞ3 𝑍21 + 𝜋
ෞ2 𝑊1𝑖

Second-stage:
෢ 𝑖 + 𝛽3 𝑊1𝑖 + 𝜀𝑖
𝑌𝑖 = 𝛽1 + 𝛽2 𝑋1

Gretl codes:
ols x const z1 z1 w1 -- robust
genr x_hat = $yhat (from first stage equation)
To estimate

Option 2: Run embedded two-stage least square model


*This is a rough notation

෢ 𝑖 = 𝜋1 𝑍1𝑖 + 𝜋2 𝑍2𝑖 + 𝜋3 𝑊1𝑖 ) + 𝛽3 𝑊1𝑖 + 𝜀𝑖


𝑌𝑖 = 𝛽1 + 𝛽2 (𝑋1

Gretl codes:

tsls y1 const x1 w1 ; const z1 z2 w1 --robust


Gretl codes:

tsls y1 const x1 w1 ; const z1 z2 w1 --robust


General Model: Many endogenous variables
If x is exogeneous, always use OLS!
Hausman Test for Endogeneity
𝑢ො is the unexplained part of Y after we control for Z!

𝑅 2 will be very low if Z are exogeneous


Z = exogenous when not reject the null.
First-stage: Checking the Relevant Assumption
tsls l_packpc const l_avgprs ; const tax --robust
Gretl:
open "cig_ch10_new.gdt”
logs avgprs #log value
logs packpc
smpl year=1985 --restrict –replace #the data contains 2 years: we will use only 1985.
ols l_avgprs const tax
tsls l_packpc const l_avgprs ; const tax --robust
smpl --full
Gretl: (I use both years in this example)

tsls l_packpc const l_avgprs income ; const tax income --robust

You might also like