rsaggio87
commented
1 year ago Hi Jorge,
Apologies for the late response, for some reason I did not get any notification of this being posted.
Can I ask you the number of years you have available? Also, do you know the number of movers? It is possible that limited mobility concerns are really minimal in your concern. As a sanity check, I would also try estimating everything with just two time periods. I would expect there to see quite a bit of divergence b/w plug-in and KSS.
Hi, thank you for making this code available. We were hoping you could provide some guidance on an issue we are experiencing.
We are estimating an AKM model with administrative data from Mexico. The connected set has between 523,000 and 695,000 firms and between 5,576,000 and 8,941,000 workers depending on the period of time of our samples. We used your package to correct the variance components' using the leave-out KSS correction. The code runs without any issues. However, our uncorrected estimates on the leave-one-out connected set, and our corrected estimates are quite similar.
We first thought there was an issue with our setup. We also thought that our estimates were not biased by limited mobility. But when we use the Bonhomme et al. correction, e.g. cluster the firms, the corrected estimates are different from the baseline estimates.
This issue arises regardless of the number of simulations or the leave-out level used. The baseline estimates and the KSS corrected estimates are not exactly the same, but differences only appear in the third or fourth decimal of the variance shares.
We are trying to understand if this may be possible or if it is the case where we need a subtle coding change since we are using an extensive database. Any suggestions?
We are using Matlab 9.6.0 (R2019a) on Windows, and version 3 of the leave-one-out code.
FYI, @jorpppp