Matrix completion: Part 1 - by scott cunningham
Susan Athey, Mohsen Bayati, Nikolay Doudchenky, Guido Imbens and Khashayar Khosravi published a paper in 2021 entitled “Matrix Completion Methods for Causal Panel Data Methods” in the Journal of American Statistical Association. It has not gotten nearly the attention of the robust diff-in-diff estimators used for staggered adoption that started popping in publications at the same time, and the original paper compares its performance with that of various alternative estimators, but not the robust diff-in-diff ones. So I was thinking what might be useful was a series that did that in an effort to fully understand what’s under the hood and how the method performs relative to other methods under different situations.
Each of these substacks is going to be about a couple of data generating processes and comparing matrix completion with nuclear norm regularization (hereafter MCNN) to a couple other estimators. I’m going to start off today using the simplest data generating process for p…
ncG1vNJzZmibkarAorjIp51nq6WXwLWtwqRlnKedZL1wucCtqaKwXZi8rrzLnquip55ivaK%2B02Zo