Reference

Historical data and its limits

Which historical return datasets Glideworth uses, where each comes from, how they are licensed and what they cannot tell you, including the limits of the UK data.

Historical Monte Carlo methods need market history: stock and bond returns, dividends, cash and inflation, by year. Glideworth bundles datasets built from public sources. Every value is an arithmetic transformation of a source cell; nothing is typed in by hand. The repository records the provenance and the licence terms of every source in DATA-LICENSES.md.

The datasets

Dataset Coverage Source Terms
US since 1928 (the default for every country, including the UK) 1928 to 2025, calendar years Aswath Damodaran, “Historical Returns on Stocks, Bonds and Bills” (stocks, 10-year Treasury bonds, 3-month bills). Inflation from US Bureau of Labor Statistics CPI-U, December to December. Damodaran: free to use, acknowledgement requested. BLS: public domain.
UK since 1983 1983 to 2025, calendar years Office for National Statistics (inflation), Bank of England (Bank Rate and 10-year gilt yields), Federal Reserve (the dollar-sterling rate), and Damodaran for equities: US shares converted to sterling. Open Government Licence v3.0 for the UK public-sector sources; Federal Reserve data is public domain.

Datasets are not updated automatically. Canada, Australia and the euro area have no history of their own, so households there test against US history unless the plan says otherwise.

Two older sets are no longer included. The UK set from 1871 to 2020 came from an academic database with a non-commercial licence. The US set from 1871 (Robert Shiller’s data) has no explicit licence. Plans that named either now run on UK since 1983 or US since 1928, and the app says so with a notice on the plan. If you run Glideworth yourself, you can download the 1871 US data for your own use with a script in the repository; it is never part of the hosted service.

How the sets are built

For each year, the engine needs stock price return, dividend yield, bond return, bond income (the coupon), cash and inflation. Bond total return is split into income and price change. Both bundled sets use calendar years, and inflation is December to December to align with calendar-year returns.

What the UK set is, and is not

The UK set uses openly licensed sources so that it can be used in a commercial service and in an AGPL repository. UK inflation, Bank Rate and 10-year gilts are UK data. No openly licensed UK share history covering more than a few years exists, so the share series is the US market (Damodaran’s S&P 500 data) converted to sterling at the Federal Reserve’s year-end exchange rate: a sterling investor holding US shares, not UK shares.

It covers 43 years, all after 1982. It contains the 2008 crash and the 2022 inflation spike and gilt sell-off, but no 1970s inflation and no wartime years, and US shares returned more than UK shares over the period. Historical runs on it are therefore markedly more optimistic than long UK history. For that reason new UK plans default to US history since 1928, which is longer and includes the Depression and the 1970s; it is still more favourable than long-run UK history, so leave a margin. You can choose the UK set in the chance-of-success settings. A plan longer than 43 years cannot use the method that replays every start year once.

Limits you should know

  • A short history is a short history. 43 years of UK data, or 98 years of US data, is a thin sample of possible futures, and overlapping start years are not independent.
  • You are probably testing against American history. The default everywhere is US stocks, bonds and inflation, in dollars; results are in real terms, so what matters is US history after inflation, not the dollar.
  • Crypto has no history and is sampled from a distribution.
  • Past returns do not predict the future.

Last checked against the code on .