Code
- Library
JaxEcon
A JAX library for solving high-dimensional dynamic economic models with deep learning. Its main application is a nonlinear model of the US economy with 37 sectors linked by input-output and investment networks, calibrated to BEA data, with 74 state variables and correlated sectoral productivity shocks, solved with neural networks trained on the model's equilibrium conditions (Deep Equilibrium Nets). It also adapts reinforcement learning to economics: analytic policy gradients train a policy by differentiating discounted welfare through the model, including models with irreversible investment. Grid time iteration serves as a benchmark.
- Dataset
AI Law AtlasNew
A registry of 1,574 laws, regulations and bills that govern artificial intelligence and its inputs (data, compute, chips, data centers and energy) in 57 countries and blocs. For 770 of them, the legal facts in the text are recorded as structured data: dated events, thresholds, obligations and sanctions, each with its provision and a verbatim quote. Annotated with language models and checked with a blinded validation protocol.
- Examples
Agentic Coding for Economics
Worked examples of coding agents in economic research: recovering and checking derivations in LaTeX, turning a model into TikZ figures and Python code, building slides, pulling industry data from the BEA, and turning agent transcripts into readable chats.