Solving Models of Economic Dynamics with Ridgeless Kernel Regressions

Replication code for Solving Models of Economic Dynamics with Ridgeless Kernel Regressions by Mahdi Ebrahimi Kahou, Jesse Perla, and Geoff Pleiss.

Setup

Install uv, then from the repo root:

uv sync

Replicating the paper

./generate_all_results.sh

runs the paper experiments, renders the figures, and writes .results/all_results.json — the authoritative source for every number in the paper. Single-model runs and driver details: replication/README.md.

Running one model

Every solver in models/ is importable and has a jsonargparse CLI; each replication/ driver regenerates one model's paper data and figures:

uv run python -m models.neoclassical_growth_kernel --kernel_set.ell 5.0
uv run python -m replication.neoclassical_growth_kernel --output_dir .results
uv run python -m replication.neoclassical_growth_kernel_figures --output_dir .results

Layout

• models/ — one solver per model/method.
• replication/ — ensemble and figure drivers and the paper-number collector.
• notebooks/ — self-contained annotated notebooks for the canonical examples, with prose-free *_minimal.py companions.
• extra/ — nonessential variants; not part of the main replication.

License

MIT