REGHDFE: Linear and IV Regressions With Many Fixed Effects

Quick News

3.0.x three key changes: i) faster underlying algorithm (symmetric transforms and cg acceleration perform much better on "hard" cases); ii) slow parts rewritten in mata, iii) simpler syntax
2.2.x [internal] murphy-topel (unadjusted, robust, cluster), double-or-nothing IV/control function
2.1.x reghdfe does not report constants anymore. If you really want to see them, run summarize on the first fixed effect. The last version that supported constants is available with net from https://raw.githubusercontent.com/sergiocorreia/reghdfe/866f85551b77fe7fda2af0aafccbbf87f8a01987/package/

Description

This is the readme file for developing the reghdfe project, which is comprised of the reghdfe package and the underlying hdfe package. What you probably want is the tutorial/documentation, available here.

Version 3.0.0
Date: May 11, 2015

Reghdfe is a Stata package that estimates linear and instrumental variable regressions while controlling for any number of fixed effects. It is most useful when dealing with two or more highly dimensional fixed effects (HDFE).

It extends the algorithm of reg2hdfe (by Paulo Guimaraes and Pedro Portugal) in several aspects, drastically speeding it up, as well as generalizing it to any number of sets of fixed effects, slope interactions, multicore, etc.

It's objectives are similar to the lfe package in R by Simen Gaure, and to several precursor Stata packages, such as a2reg by Amine Ouazad.

HDFE is the package that powers reghdfe. It is a utility function, like avar, and intended to be used for those that want to allow HDFEs in their own regression packages.

Features

Can estimate any number of highly-dimensional fixed effects.
Extended to allow for intercept and/or slope effects by group (i.e. by absorbing variable).
Allows instrumental variable regressions by using the ivregress and ivreg2 packages.
Allows GMM2S/LIML/CUE estimators using the above packages.
Supports multi-way-clustering, as well as a battery of HAC methods for VCE estimation, thanks to the avar package (the same package that powers ivreg2).
Supports analytic, frequency, and probability weights.
Careful estimation of the degrees-of-freedom lost due to the fixed effects.
Can easily save the fixed effect estimates (whether or not they are identified is another matter..)
Very fast by default. Controlling for many fixed effects is usually extremely slow and memory consuming. Therefore, the code has been carefully profiled to avoid any bottlenecks within the Stata and Mata code.
Even faster experimental options: Although Mata does not provide custom multicore support, it can be achieved by opening multiple Stata instances and parallelizing by variable. This has been tested only in Windows, but can be extended to OSX/Linux easily. There are also other optimization options that can further speed it up.
Attention to the detail. For instance, you can just include firm#year instead of writing egen firm_year = group(firm year). This works in the varlist, in the absvars, and even with the vce(cluster ..) option.
Many other options.

Installing the beta version of reghdfe 3.0

With Stata 13:

cap ado uninstall reghdfe
net from https://raw.githubusercontent.com/sergiocorreia/reghdfe/updated_mata/package/
net install reghdfe

With Stata 12 or older:

Download the zipfile
Extract it into a folder (e.g. C:\SOMEFOLDER)
Run: (changing SOMEFOLDER with whatever you picked)

cap ado uninstall reghdfe
net from "C:\SOMEFOLDER"
net install reghdfe

Installing the development version of reghdfe (i.e. from Github instead of SSC)

cap ado uninstall reghdfe
net from https://raw.githubusercontent.com/sergiocorreia/reghdfe/master/package/
net install reghdfe

Note: this requires Stata 13, as older versions of stata cannot handle https. If you have an older version, see below.

If you have an older version of Stata, first extract this zipfile to a folder on your computer, and then replace the second line above with net from My/Path/ where My/Path is the path where you saved the files. Note that with this workaround you will not be able to use adoupdate, update to automatically update from github.

Installing HDFE

cap ado uninstall hdfe
net from https://raw.githubusercontent.com/sergiocorreia/reghdfe/master/package/
net install hdfe

(The only difference is the name of the file)

Contributing

Contributors and pull requests are more than welcome. There are a number of extension possibilities, such as adding two-way clustering, multicore support for OSX and Linux, estimating standard errors for the fixed effects using bootstrapping, and exact computation of degrees-of-freedom for more than two HDFEs.

Building the package

For clarity reasons, the source code is spread through several files and folders (in the source folder). To modify the package, do the following:

Download the entire project to your computer (through the "Clone Desktop" or "Download ZIP" buttons on the right).
Uninstall any existing versions of reghdfe (ado uninstall reghdfe in Stata).
In Stata, change the current working directory to the source folder. Do any changes that you want on the files in that folder. You can run reghdfe without problems as long as the working directory is in that folder.
To build the package, run the build.py file (in the build folder), using either Python 2.7 or Python 3.x (3.x is not tested but should work). This python script will carefully combine all the files and update the version/date.
Finally, you can upload it back to github and submit a pull request.

Author

Sergio Correia, Duke University

Acknowledgments

This package wouldn't have existed without the invaluable feedback and contributions of Paulo Guimaraes and Amine Quazad, as well as the great bug-spotting abilities of many users.

kolexiang/reghdfe