ManyIV

Note: Currently in beta. Various instrumental variables regressions (OLS, TSLS, LIML, MBTSLS, JIVE, UJIVE, RTSLS) with absorbed instruments and controls.

version 0.7.2 09Jun2024 | Installation | Usage | Examples | Compiling

Installation

From Stata:

local github "https://raw.githubusercontent.com"
net install manyiv, from(`github'/gphk-metrics/stata-manyiv/main/)
manyiv _plugin_check

You can also clone or download the code manually, e.g. to stata-manyiv-main, and install from a local folder:

git clone git@github.com:gphk-metrics/stata-manyiv stata-manyiv-main

(or download the code manually and unzip). Then from Stata:

cap noi net uninstall manyiv
net install manyiv, from(`c(pwd)'/stata-manyiv)

NOTE: If you get an error when checking the plugin you can still use manyiv but its capabilities will be limited. In particular, without the plugin the number of absorb variables is capped at 2, and when the absorb variables have many levels the function's performance will be very slow. If the plugin fails to load, please see the compiling section below.

Usage

manyiv depvar (endogenous = instrument) [exogenous], options
help manyiv

Examples

clear
set seed 1729
set obs 1000
gen u  = rnormal()
gen z1 = rnormal()
gen z2 = rnormal()
gen e  = rnormal() + u
gen c  = int(runiform() * 10)
gen fe = int(runiform() * 15)
gen iv = int(runiform() * 8)
gen w  = rnormal()
gen x  = 1 + 0.1 * z1 - 0.2 * z2 - 1/(1 + iv) + u
gen y  = 1 + x + w + 1/(1 + fe) + e

manyiv y (x = z1 z2) w
manyiv y (x = z1 z2) w, cluster(c)
manyiv y (x = z1 z2) w, absorb(fe) cluster(c)
manyiv y (x = z1 z2) w, absorbiv(iv) cluster(c)
manyiv y (x = z1 z2) w, absorb(fe) absorbiv(iv) cluster(c)
manyiv y (x = .)     w, absorb(fe) absorbiv(iv) cluster(c)

You can also save any of the estimators to a Stata table:

manyiv, est(TSLS)  vce(cluster)
manyiv, est(UJIVE) vce(cluster)

There are instances where jive/ujive will fail because they are leave-one-out estimators and a covariate or fixed effect group will be collinear with a given observation. If dealing with the issue manually is otherwise prohibitive, the forcejive option will prompt manyiv to drop observations and/or covariates until jive/ujive can run.

clear
input z1 z2 fe iv
-1 -1 0 0
1  1 0 0
3  3 1 0
-3 -3 1 1
0  0 2 1
0  0 2 1
1 2 3 2
2 3 3 2
end
gen z3 = _n^2
gen x  = z1 + z2 + z3 + rnormal()
gen y  = x + rnormal()
manyiv y (x = z1 z2), absorb(fe) absorbiv(iv)
manyiv y (x = z1 z2), absorb(fe) absorbiv(iv) forcejive

Compiling

To compute the jive and ujive estimators with multiple absorb levels, manyiv uses a plugin compiled with the Eigen3 library. Pre-compiled binaries are provided in this reposirtory. To re-compile the plugin from source, download Eigen3 using your system's package manager (e.g. brew, apt, etc.) or from their website and run:

git clone git@github.com:gphk-metrics/stata-manyiv
cd stata-manyiv
make all EIGEN=/path/to/eigen3

(NB: On Windows, this needs to be compiled using MinGW.) If your system's libraries are not compatible with the latest version of Eigen3, you can still compile manyiv using an older version. For example,

git clone https://gitlab.com/libeigen/eigen.git
cd eigen
git checkout 3.3.9

Replace 3.3.9 with the version you want to try. (Run git tag to see all available versions.) Then you can run make in the stata-manyiv folder, setting EIGEN to the folder you just downloaded from gitlab. Once the plugin is compiled, open Stata from the stata-manyiv directory and run:

cap ado uninstall manyiv
cap net uninstall manyiv
net install manyiv, from(`c(pwd)')