Data Sources

The primary dataset is the Jordà-Schularick-Taylor Macrohistory Database (release 6), an academic panel covering 18 advanced economies with annual data on equity total returns, government bond returns, short-term bills, consumer price inflation, house prices, and exchange rates. FIRE Lab uses 16 countries with sufficiently complete return histories, starting in 1871. The market-history presets default to 1900 because that is the first year all 16 countries are covered (Switzerland, Spain, and the Netherlands begin in 1900).

JST is the standard academic source for long-run cross-country return data, underpinning the well-known "Rate of Return on Everything" research. All returns are converted to real (inflation-adjusted) local-currency terms before simulation.

The database is published under a CC BY-NC-SA 4.0 license by the original authors (Jordà, Schularick, Taylor et al.).

Official JST data ends in 2020. FIRE Lab extends each country through 2025 using public sources: equity returns from national total-return indices via Yahoo Finance (priced on annual averages to match JST's convention), CPI from the IMF World Economic Outlook, and bond returns reconstructed from OECD long-term interest rates.

The extension follows JST's methodology as closely as practical — including annual-average equity pricing and legacy-currency handling for Eurozone members — and is validated against overlapping official data. It is nonetheless unofficial: treat the most recent five years as best-effort estimates rather than curated academic data.

The US-only mode uses long-run US market data as organized by the Bogleheads community's Simba backtesting spreadsheet: S&P 500 total returns with pre-1957 predecessor indices (ultimately Shiller's 1871+ series), 10-year US Treasury returns, and for international equities the Simba 'Total International' lineage from 1970 — vintage MSCI EAFE for 1970–96 (about 0.8 points a year below the official gross index; its embedded cost and tax treatment is not fully identified) and published total-international index-fund returns from 1997, which are net of fund expenses and foreign dividend withholding.

This dataset reflects the single most successful major equity market of the past 150 years, so U.S.-only results are an optimistic reading of the future. The pooled mode is the stricter cross-check — though note it substitutes another country's inflation and domestic assets block by block, so it stress-tests a plan rather than forecasting the U.S.

The US dataset variant with international backfill replaces the pre-1970 international column (which is a US placeholder in the original spreadsheet) with a GDP-weighted ex-US equity series derived from JST, restoring genuine US/non-US diversification in early history. Per-country JST returns have been validated against corresponding MSCI indices (differences within ~0.3pp annualized). Since September 2026 this variant, from 1900, backs the recommended “U.S. history — long window” preset as the U.S. investor-experience baseline. The JST USA rows remain available as the “JST panel” preset for like-for-like comparison with the 16-country history; the two agree closely on US stocks, bonds and CPI (small series differences remain — a 100% US-stock allocation moves by under 0.1 point of SWR) and the material gap is the post-1970 international leg, where JST's GDP-weighted gross academic basket runs about 1.8 percentage points a year above the investable series — a weighting and cost difference, not a data-quality one.

All simulations run on real returns. Multi-country simulations pool the 16 countries with equal probability (1/N) per sampled block — no GDP weighting — a choice validated by sensitivity analysis showing safe-withdrawal-rate differences under 0.1pp across weighting schemes.

Block Bootstrap sampling draws contiguous blocks of 5–15 years (uniform) with same-country circular wrap-around, preserving multi-year momentum and volatility clustering. See the methodology page for details.

Survivorship and selection: the 16 countries are economies that remained investable for 150 years. Markets that closed permanently (Russia 1917, China 1949) are not represented, which biases all long-run datasets optimistic.

House-price indices are appraisal-based and smoothed, understating true volatility; they inform the buy-vs-rent tool but are not used as an investable asset in portfolio simulations.

Historical returns do not guarantee future performance. Taxes, currency risk for international investors, and trading frictions beyond fund expense ratios are not modeled.