Non-Parametric Earnings-Dynamics Toolkit
MIT-Licensed • Maintainer: Tong Zhou ( zhoutong.econ [at] gmail )
The Panel Study of Income Dynamics (PSID) underpins U.S. fiscal, monetary, and social-insurance models at the Federal Reserve, CBO, SSA, and other agencies.
semqreg supplies the first fully non-parametric estimator that:
- Separates persistent vs transitory risk in earnings without functional-form assumptions;
- Captures heavy-tailed, ARCH-type volatility documented in PSID data;
- Reproduces Table 2 of Zhou (2024) with < 2 min runtime on a laptop.
Accurate persistence and volatility inputs flow directly into FRB/US stress tests, CBO dynamic tax scoring, and Social-Security actuarial balance calculations—central arguments in the accompanying National-Interest Waiver filing.
| Module | Purpose | Key Output |
|---|---|---|
np_id.py |
Spectral operator identification of a hidden-Markov earnings process | Persistent shock rhô and transitory variance σ̂²ₜ |
psid_loader.py |
One-click PSID 1968-2017 extract (792 balanced households) | Clean long panel earnings.csv |
forecast_tools.py |
5-year Gini-path & mobility-elasticity projection | gini_proj.csv, mobility_proj.csv |
demo_notebook.ipynb |
Replicates Figures 5–9 in Zhou (2024) | Plots + LaTeX tables |
# clone & enter
git clone https://github.com/Tongzhoutz/semqreg.git
cd semqreg
# create environment
conda env create -f environment.yml
conda activate semqreg
# run demo replication
python np_id.py # estimate HMM
python forecast_tools.py # 5-year projections
jupyter notebook demo_notebook.ipynb