Tuning Parameter-Free Nonparametric Density Estimation from Tabulated Summary Data

In 2017, I came up with an idea of using the maximum entropy principle to nonparametrically estimate income distributions from tabulations provided by tax authorities. I showed my notes to a few econometricians but they didn’t show interest or didn’t have time, and because I didn’t know how to formulate the problem statistically and prove asymptotic results, the project became dormant. In 2021, I was invited by my coauthors to work on a different project. While working on this project, I told my coauthors about my old idea and they showed interest, so we wrote a paper together.

Although it’s a relatively short and simple paper, I like it. The replication files are posted here. Although we have not conducted a systematic simulation study, I believe our method, which is based on rigorous asymptotic theory, is superior to other ad hoc methods such as Blanchet et al. (2022). I would like to encourage readers to test the method and conduct simulation studies.