Skip to content

johnshun001/economic-simulation

Repository files navigation

Tennessee Economic Data Lake & Simulator

A county-first data lake and agent-based simulation engine for the Tennessee economy — ingest, normalize, and curate public economic data, then model how shocks propagate across the state's commuting and industry networks.

County commuting-network hub map

Commuting-network hub map: each node is a Tennessee county, sized and colored by network centrality. The top-5 structural hubs — Davidson, Williamson, Rutherford, Wilson, and Shelby — are starred. Faint edges show inter-county commuting flows.

What it does

The project has two layers:

  1. Data lake (Phase 1) — ingests, normalizes, curates, and exports county-level datasets that later stages build on. Sources include Census PEP and ACS demographics, PUMS household microdata, QCEW wages, CBP establishment counts, and Tennessee tax collections.
  2. Simulation engine (V1) — builds a calibrated "economic world" from the curated data, computes network centrality across the county commuting graph, and runs shock-propagation and spillover scenarios with interactive visual outputs.

Interactive outputs

The V1/ directory generates standalone HTML visualizations (Plotly). Included here:

  • V1/hub_map.html — the commuting-network hub map shown above, with a selectable centrality metric.
  • V1/run_comparison.html — side-by-side comparison of two simulation runs.

The full set (county drilldowns, sector corridors, scenario diffs, animated economy) is generated by the scripts in V1/ from the local data lake.

Structure

Path What's inside
pipelines/ Ingestion, transform, validation, and export code
schemas/ Table contracts for curated and feature tables
V1/ Simulation engine: calibration, centrality, scenario scripts + demo HTML
infra/ Terraform for the S3 lake bucket, Glue databases, Athena workgroup
docs/ Source registry, runbooks, reproducibility notes
tests/ Fixture-backed pipeline tests that run locally without AWS

Notes

This repository is a curated code + demo slice. The raw data lake, curated parquet outputs, trained models, and validation artifacts (tens of GB) are intentionally excluded — see .gitignore. The pipeline code regenerates them from public sources.

About

County-first data lake + agent-based simulation of the Tennessee economy, with commuting-network centrality maps.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages