Read this before trusting anything else on the site.
This project studies the dot-com unwind (2000–02), the global financial crisis (2007–09) and the COVID crash (2020). Three is far too few to infer from. No confidence interval is computed across episodes anywhere on this site, and no probability of a future collapse is published. Where a pattern holds everywhere it is reported as “3 of 3” — a count, not a rate.
Nothing here predicts when the next collapse begins. Fitting a timing model to three observations produces a confident-looking number with no information in it.
The natural credit measure is the high-yield option-adjusted spread. FRED’s ICE BofA series are licensed and return data only from August 2023, so they cover none of the three episodes. Moody’s Baa minus the 10-year Treasury is used instead. It is an investment-grade spread: coarser, and less violent in a crisis than high-yield would be. Every credit finding on this site inherits that limitation.
FRED’s S&P 500 and Dow series are licensed and capped to a ten-year rolling window, covering none of the three episodes. The equity spine is the Ken French Data Library’s daily research factors: the value-weighted whole-US-market total return from 1926, plus the risk-free rate. Total return is what an investor actually earned.
Six series do not reach far enough back to cover it. The Fed balance sheet
(WALCL) begins December 2002. The market-implied breakevens
(T10YIE, T5YIE) begin 2003. Nasdaq volatility
(VXNCLS) begins February 2001 — after the Nasdaq had already peaked.
The current broad-dollar index (DTWEXBGS) begins 2006. And EUR/USD
(DEXUSEU) begins January 1999, ten months after the 24-month buildup window
this episode needs open in March 1998 — close, but not early enough. The coverage
table shows exactly which series reach which episode.
No daily gold price is available from any free keyless source. Gold comes from the World Bank Pink Sheet at monthly frequency, so no intra-crash gold path is computable.