Datasets you can cite
Derived statistics from the series behind Market Memory, recomputed at every build and published as a table, a CSV and a press package. Each carries the date of the last close it was computed from and a freshness state: CURRENT within two days of the build, DELAYED within six, STALE after that. Free to reuse with attribution. Historical context, not a forecast.
How often markets move 1%, 2% and 3% in a day
For every index, commodity and currency series the site stores: the share of sessions with a move at or beyond 1%, 2% and 3% in each direction, how many a year that is, the sessions since the last one and the longest gap on record.
CURRENT · as of 2026-10-26 · 90 rows →The twenty worst and best days for every series, and what followed
The largest one-day falls and rises on record for each stored series, with the drawdown from the high at the time, the realised volatility going in, and the change over the following day, week, month and three months.
CURRENT · as of 2026-10-26 · 600 rows →This calendar date in every earlier year
For the build date, what each stored series did on the same calendar date in every earlier year with a session: the share of years it rose, the median, the best and the worst. Rebuilt daily.
CURRENT · as of 2026-10-26 · 15 rows →What is and is not here
Frequencies, percentage changes, counts of sessions and dates. Not index levels: those belong to the index providers, and a table of them would be redistribution. The arithmetic is in the open — the methodology page sets out every definition — and the build fails if a published figure disagrees with the source rows.
If you use a table, the attribution line on each page is enough. If you want a cut the CSV does not give, the how we work page says how to reach the desk; nothing is sent to anybody by the build.