ts2sls_python

This is a Python program to calculate two-sample two-stage least squares (ts2sls) estimates based on Inoue and Solon (2010). This is a follow up to Dr. Jeffrey Shrader's Stata code for the same available here.

As a benchmark, I have also included a Python program to calculate 2sls estimates which are equivalent to Stata's ivregress 2sls.

Syntax for Python

from ivregress import ts2sls
ts2sls(S1, S2, y_var, regs, ev, inst)

where:

• S1 is the dataframe with sample 1 - this may not have the endogenous variable
• S2 is the dataframe with sample 2 - this may not have the dependent variable
• y_var is the name of dependent variable
• regs is the list of regressors or indepdent variables
• ev is endogenous variable provided as a list with 1 element
• inst is the list of exogenous instruments

For some examples, see unit_tests.py

Installation

There are multiple possible ways. I have provided 2 simple ones here

1. For a quick and dirty solution, you can download the ivregress.py file to the folder where you have your Python script (example - unit_tests.py) to call the ts2sls function. This would enable Python to find the functions easily.
2. A more permanant solution requires you to find where your Python modules are stored. Run this code on your terminal python -c "import sys; print(sys.path)" . This would print a list of locations that Python searches for a module to be imported. Look for the one which ends in site-packges (example - /opt/anaconda3/lib/python3.7/site-packages). Download the file ivregress.py to this location.

For R users

You can install reticulate library in R which allows you to call Python modules from R interface. For instructions on how to use reticulate, see this. As invregress code calls pandas library, you may also need to install that. Use the following code to complete both these tasks:

install.packages("reticulate")
library(reticulate)
py_install("pandas")

Once the installation is complete, follow the example code - unit_tests.R to run these functions in R.

References:

Inoue, Atsushi, and Gary Solon. "Two-sample instrumental variables estimators." The Review of Economics and Statistics 92, no. 3 (2010): 557-561.