Identification, Estimation, and Inference in Panel Vector Autoregressions identified using external instruments

Replication package for “Identification, estimation and inference in panel vector autoregressions using external instruments”. The package reproduces every figure and table of the paper and of its supplementary material. It contains: pvariv, a Python library that estimates panel vector autoregressions identified with an external instrument (PVAR-IV). It implements the within estimator with the half-panel jackknife correction and FOD-GMM, impulse responses and cumulated responses, Anderson-Rubin and conditional likelihood ratio sets robust to weak instruments, and covariance estimators clustered by unit, by period, or in both dimensions. An extension handles spillovers across units through a known exposure mapping; the code and data of the empirical application, which estimates the effect of stock market wealth on nontradable payroll in US counties, using the county panel of Chodorow-Reich, Nenov and Simsek (2021, American Economic Review); the code of the Monte Carlo simulations, with the stored results reported in the paper. All code is in Python. The README explains how to reproduce each result. The data extract is redistributed from the replication package of Chodorow-Reich, Nenov and Simsek (2021) under its Creative Commons Attribution license; please cite that article when using the data.