LeaveOutTwoWay

This Matlab package implements the leave out correction of Kline, Saggio and Soelvsten (2020) for estimating variance components in two-way fixed effects models. It can also report asymptotically valid standard errors on coefficients obtained from a second stage regression of estimated fixed effects on observables.

See this vignette for a description of the package.

Julia Version

The Julia version of the package (developed by Paul Courcera) can be found here. On the same page, one can find an executable that permits estimation of the leave-out correction even if the user does not have MATLAB or JULIA.

Summary of Changes

These are the most significant changes introduced with this new version

• New documentation that describes in detail the functioning of leave_out_KSS.
• By default, the code runs a leave-out correction by leaving a match out as opposed to leaving an observation out. See vignette for details.
• New, optimized, random projection algorithm for calculation of the statistical leverages which scale extremely well to large datasets.
• Code no longer requires MATLAB BGL but runs automatically on MATLAB built-in network functions.
• Requires MATLAB R2015b or higher.

Replication Package

The replication package of the Econometrica article can be found here.