Software

Most of my OSS efforts center around py-econometrics, where we build software for applied econometrics in Python.

• pyfixest implements fast routines for fixed effects regression (OLS, IV, Poisson) and a wide range of inference procedures (iid, HC1-3, CRV1, CRV3, randomization inference, multiple testing corrections via Bonferroni & Romano-Wolf) following the syntax of the formidable fixest R package. When we started in 2022, our goal was to finish the package before the LLMs fully take over writing code. Now that LLMs write most of my code, and I am still working on PyFixest, I am not sure if we succeeded - but we have gotten quite far =)
• maketables produces publication-ready regression tables for models from pyfixest, statsmodels, and linearmodels in HTML (via great-tables), LaTeX and Word.
• within provides high-performance solvers for projecting out high-dimensional fixed effects from regression problems. It includes a Rust core and Python bindings, and exploits the graph structure of fixed-effects systems. It is very fast =)
• duckreg provides a simple interface to run lossless regressions on very large datasets that do not fit in memory by first reducing the data to a set of summary statistics out-of-memory using DuckDB and then running weighted least squares with frequency weights.

I got hooked on working on open source during my PhD, when I was writing my code in R (a language that I still love) and needed a fast routine for wild cluster bootstrapping.

• fwildclusterboot and wildboottest implement the fast wild cluster bootstrap algorithm as suggested in Roodman, MacKinnon, Nielsen & Webb and MacKinnon, Nielsen & Webb (2022) in R and Python.
• wildrwolf and wildwyoung implement the multiple-hypothesis corrections by Romano & Wolf (2005) and Westfall and Young (1993) for regression objects from the fixest package (R). Because both packages are built around fwildclusterboot and the wild (cluster) bootstrap, both are usually quite fast.