Dear authors, I am a big fan of the method and I have used it in several of my papers. My comment here is not really about real issues but more about some potential add-ons to the method that could further improve it. My main points are the following:
Is it possible to add a metric (e.g., the Multivariate L1 distance) that can help the practitioner to choose among the several matching and reweighting options (also in order to reduce user discretion)?
Is it possible to add some kind of formal test able to tell the practitioner whether he/she can or cannot reject the hypothesis that all pre-treatment trends are parallel?
Is it possible to add among the matching methods, synthetic control-type algorithms (e.g. the synthetic difference-in-differences)?
One last thing concerns the importance (in my opinion) to stretch the pros of the non-parametric approach in terms of transparency and use of milder assumptions with respect to the most recent parametric methods very much used by economists (I am thinking about all recent extensions of the TWFE estimator). Is there any recent paper trying to compare these two very different approaches?
Dear authors, I am a big fan of the method and I have used it in several of my papers. My comment here is not really about real issues but more about some potential add-ons to the method that could further improve it. My main points are the following:
One last thing concerns the importance (in my opinion) to stretch the pros of the non-parametric approach in terms of transparency and use of milder assumptions with respect to the most recent parametric methods very much used by economists (I am thinking about all recent extensions of the TWFE estimator). Is there any recent paper trying to compare these two very different approaches?
Many thanks, Augusto Cerqua