How the Crash Indicator works

The Crash Indicator shows 26 market and macro indicators, each with its own warning rule and a status of clear, elevated or warning, grouped into valuation, credit, market structure and macro. It has no single crash score on purpose. A track-record table replays 15 of the rules against the 8 US recessions since 1969, so you can see how often each warned in time and how often it raised a false alarm.

What are the 26 indicators?

Each group gets a summary status from how many of its indicators are flagged. The summary describes conditions today. It is not a forecast and not a combined score.

What are the warning rules?

These are the 15 indicators with a fixed threshold and a long enough free history to replay. The rules are the ones the page already uses, not ones tuned afterwards.

A leading indicator is meant to move before a downturn, a coincident one confirms it as it happens, and a lagging one follows it. The Sahm rule is coincident by design.

How is the track record calculated?

• Recessions flagged: of the recessions whose 24-month run-up the indicator's data covers, how many saw the warning on at least once in the 24 months before the recession began.
• Typical lead: the median number of months between the first warning in that window and the recession's start.
• False alarms: warning episodes (runs of warning months, with gaps of up to 2 months merged) that began outside a recession and were not followed by one within 24 months. Episodes too recent to judge count as pending.
• Crash follow-through: the share of warning episodes after which the S&P 500 fell 20% or more within 12 months, next to the same share for all months. A signal only means something if it beats that base rate.

Recession dates are the NBER's. Months still inside an earlier recession are skipped, so the 1980 signal cannot count as a warning for 1981.

Which indicators are not tested, and why?

• High-yield and investment-grade credit spreads: FRED now keeps only three years of ICE BofA history (licensing), too short to judge.
• Buffett Indicator and forward P/E: Built from today's tracked stocks and analyst estimates; no long free history exists.
• Market breadth, defensive rotation, copper/gold, VIX term structure: The data starts in the 2000s or later, or only covers today's tracked stocks.
• Equity risk premium: Below zero for most of 1980-2003 (the whole 1990s bull market) and in 2008-09, so zero is not a warning line; the card is shown without a status.
• Insider sell/buy ratio: Only the trailing 90 days is available.
• Consumer credit and corporate profits growth: Shown as trends with no warning threshold, so there is no rule to test.
• Yield-curve recession odds (our own fit): The model was fitted on the same history, so replaying it would flatter it.

Where does the data come from, and what is left out?

Most series are from FRED (St. Louis Fed): yields and spreads, the financial stress and conditions indexes, bank lending standards, jobless claims, the Sahm rule, recession probabilities, the Chicago Fed activity index, the Philadelphia Fed survey and consumer sentiment. The Shiller CAPE comes from Robert Shiller's data via multpl.com, margin debt from FINRA, sector and commodity ratios from Twelve Data ETF prices, and insider activity from SEC Form 4 filings. Breadth, the Buffett Indicator and forward P/E are computed from this site's own tracked stocks. Everything is refreshed weekly, and each card shows the date of its latest reading.

What are the limits?

• The replay uses today's revised data, not what was visible at the time. Jobs, financial-conditions and earnings data get revised.
• There are few events: 8 recessions since 1969, and many indicators cover only some of them.
• Recessions are not crashes. 1987 had a crash with no recession, and 2022 had a bear market with none.
• S&P prices in the replay are Shiller's monthly averages, which understate the depth of fast crashes.
• Valuation warns years early, so a 24-month test is harsh on it.

Frequently asked questions