How Much Should We Trust the 30–40 Percent? Financial Knowledge and Retirement Wealth Inequality

Replication package for "How Much Should We Trust the 30–40 Percent? Financial Knowledge and Retirement Wealth Inequality". This package contains the code, outputs, and documentation needed to reproduce every number, table, and figure in the manuscript. The paper audits the flagship claim that financial knowledge explains 30–40 percent of retirement wealth inequality (Lusardi, Michaud, and Mitchell 2017, Journal of Political Economy). Contents: (1) a Python port of the LMM structural model plus LMM's official Fortran code for provenance; (2) NFCS 2009–2021 financial literacy and behavior pipelines; (3) HRS 2004 wealth regressions with corrected education coding; (4) PSID moment construction and formal likelihood-ratio inference for the knowledge return gradient; (5) the manuscript's 13 tables extracted verbatim and 3 figures. To replicate: unzip, place the external public-use survey files in data/raw/ following data/codebook/source-register.md, install the pinned Python 3.12 requirements (code/requirements.txt), and run "bash code/run_all.sh". Core pipeline runtime is about 2.5 hours; the PSID section requires the PSID_DATA_DIR environment variable and adds about 3 hours. Raw survey microdata are not bundled; all sources are public or registered public-use.