Hello. I'm trying to estimate an effect using panel data with three time periods (0, 1 and 2). Time 0 contains pre-treatment information.

I have very few treated observations so that propensity score estimation is not very good. I'm instead trying to use a nearest neighbor matching.

The package nnmatch, when used on longitudinal data, matches neighbors in every time period and runs regressions as if estimating a cross-section in every t.

What I which is to use time 0 data to establish the matches using the treatment status for the unit (so that if a unit is treated only on T=2, it is considered treated in 0) and then run a fixed-effects estimation using the neighbors established at t = 0.

I don't seem to find a package that is able to perform this estimation (NN) using longitudinal data... The biggest problem of doing it manually is related to inference (robust and clustering). Is anyone familiar with such a package?

Thanks in advance