ivreg2 and multi-way clustering

[Jack Simha]

Hello,

Apologies for posting on this topic again. I have gone through the prior postings on this topic but none of the suggested solutions seem to work for me.

I have a dataset with over 40,000 obs with unit1-unit2-time as the unit of analysis (i.e. it is dyad-time). I am trying to run the ivreg2 command (with two instruments) and a bunch of controls. I am specifically interested in the Kleibergen-Paap Wald rk F statistic.

It seems to run fine and generate the F-stat (as well as the SE in the first stage) when I cluster by dyad (i.e. a unique number based on the unit1-unit2 combination). That is, when I use the option cluster(unit1-unit2-uniqueid)

But it fails to generate the Kleibergen-Paap Wald rk F statistic when I use multi-way cluster of unit1 and unit2. That is, when I use the option cluster(unit1-uniqueid unit2-uniqueid)

The error message I get is

warning: -ranktest- error in calculating underidentification test statistics;

Warning: Error in calculating first-stage id statistics above;
dof of SW statistics is 3, should be L-(K-1)=..
Underidentification test
Ho: matrix of reduced form coefficients has rank=K1-1 (underidentified)
Ha: matrix has rank=K1 (identified)
Kleibergen-Paap rk LM statistic Chi-sq(.)=. P-val= .

Weak identification test
Ho: equation is weakly identified
Cragg-Donald Wald F statistic .
Kleibergen-Paap Wald rk F statistic

I have tried partialling out indicator variables as well as dropping collinear variables (though this gets dropped anyway), but these do not seem to help. I think this is something to do with the size of the variance matrix but I can't seem to understand why it works with one-way clustering but not with multi-way clustering.

I will appreciate any insight. Thanks

Best,
Jack Simha