Replication Package for "FEMA Flood-Map Updates, Insurance Coverage, and Mortgage-Market Margins"

This repository contains the replication materials for the paper “FEMA Flood-Map Updates, Insurance Coverage, and Mortgage-Market Margins”. The paper studies FEMA flood-map update timing as a flood-risk-management implementation event. It links FEMA map-change timing, current flood-hazard exposure, NFIP insurance records, HMDA mortgage records, disaster controls, social vulnerability, prior flood-loss measures, National Risk Index indicators, and CRS/local capacity variables for census tracts in eight flood-prone U.S. states over 2018–2024. The preferred treatment is the first FEMA map-change timing in tracts with current Special Flood Hazard Area share of at least 1%. The estimates are interpreted as conditional associations around map-change timing, not as effects of randomly assigned map shocks. Main results show that map-change timing in flood-exposed tracts is associated with higher aggregate NFIP insurance activity. The preferred estimate for log aggregate NFIP premiums is 0.074 log points, with a standard error of 0.022. Log NFIP policy count increases by 0.032 log points, and log total coverage increases by 0.118 log points. Building and contents coverage both rise, while premium per $1,000 of coverage is close to zero. Mandatory policy count and mandatory policy share do not show robust increases. Mortgage-market outcomes are smaller and secondary to the insurance and coverage results. The package provides two replication paths: Reproduce results from frozen cleaned data: python code/reproduce_from_cleaned.py --output outputsRebuild from public raw data and then reproduce results: python code/rebuild_from_raw.py --output outputs The cleaned-data path uses the included tract-year analysis panel and simplified treated-tract geometry. The raw-data path requires the user to download public source files listed in docs/raw_data_manifest.csv; the package provides file names, required paths, source information, and file sizes. Heavy public raw data files are not bundled. The repository includes Python code, cleaned data, raw-data download instructions, data dictionaries, output crosswalks, reference tables and figures, package manifests, and checksums. Final manuscript and Supporting Information outputs are stored in outputs_reference/