LATE-CYCLE WARNING SIGNS
Where the market usually cracks first.
Equity indexes are late to tell you something is wrong. Credit markets, the yield curve, and the labor market are early. This page tracks the five signals that historically move before the S&P does, with recession shading so you can see the pattern.
Top Watch — how big could the next drop be?
Nobody can call the exact top, and the first leg down is usually set off by surprise news. What we can measure is how fragile the market is when that news hits. Two separate readings, each tested on the S&P 500 from 1991 to 2025.
How these readings looked in the past
S&P 500 (log scale), with the three Top Watch readings as they stood each week in the strips below. Point at the chart for that week's readings.
Each week is scored only with data that had been published by then (for example, the unemployment rate arrives about five weeks late), so this is what the page would have shown at the time. A crash-risk level has to hold for four weeks before its color changes, to keep one-week blips from cluttering the chart. Oil history starts in 2000, so the stress strip counts one fewer sign before 2001.
This is a forecast, not a sell signal. We also tested selling when crash risk rose and the index broke below its 200-day average. It cut the worst loss roughly in half, but earned less than simply holding (8.4% vs 9.2% a year since 1991), because it kept missing rebounds. Odds are for any 12-month window, measured from each day's close to the lowest close in the following year.
Signal dashboard — right now
Each card is colored by its current status. Click any signal below for the chart and interpretation.
1. Credit spreads — the canary in the coal mine
The extra yield corporate bonds pay over Treasuries. When this "risk premium" is tight, credit markets are complacent. When it widens fast, they're pricing in defaults — and equities usually follow within weeks. High-yield (junk) spreads are the more sensitive signal; investment-grade is the second wave.
2. Yield curve — the recession clock
Long-term Treasuries normally pay more than short-term ones. When the curve inverts (long yields fall below short yields), it's the market saying "the Fed will have to cut soon." Every US recession since 1970 was preceded by a 10Y-3M inversion. The lag from inversion to recession is usually 6-18 months; the un-inversion sometimes marks the trough.
3. Employment — the last shoe to drop
Jobs are the slowest signal here — unemployment usually starts rising after the recession has begun. But two patterns fire early: initial jobless claims turning up before the peak, and the Sahm rule (unemployment up 0.5% from its recent low), which has fired at the start of every US recession since 1970.
Data sources & methodology
Credit spreads (HY): the most recent ~3 years are the
real
BAMLH0A0HYM2
(ICE BofA US High Yield Index Option-Adjusted Spread) via the
authenticated FRED API. FRED restricts the full history of ICE BofA data
to paid ICE licensees, so the pre-2023 portion is a proxy
computed from Yahoo's TLT (20+yr Treasury) and HYG (high-yield ETF)
daily closes: log(OAS) = a · (TLT ÷ HYG) + b where
(a, b) are fit by OLS to the recent overlap so proxy
numbers visually match the real ones where they meet.
Proxy caveat: the SHAPE (every major stress event since 2007) is captured correctly, but pre-2011 magnitudes are compressed because Treasuries didn't rally proportionally to HY's crash in 2008. Real 2008 peak was ~22%; proxy shows ~7%. Real 2020 peak was ~11%; proxy shows ~15%. Read the pre-2023 line as "how much worse than today," not as a literal OAS level.
Credit spreads (IG): BAMLC0A0CM (ICE BofA US Corporate Index Option-Adjusted Spread), same FRED restriction — last ~3 years of real data, no proxy back-fill for now.
Yield curve: T10Y2Y (10-Year minus 2-Year) and T10Y3M (10-Year minus 3-Month). Both are daily constant-maturity Treasury series computed by the Federal Reserve. The 10Y-3M is the Fed's preferred recession signal; the 10Y-2Y is the more-cited market version.
Employment: ICSA (initial jobless claims, weekly) and UNRATE (civilian unemployment rate, monthly). Both seasonally adjusted.
Trailing 10-year CAGR: the S&P 500's compound annual growth rate (total return, dividends reinvested) over the past 10 years — what the index earned per year, on average, with each year's gains building on the last — compared with every 10-year window ending since 1936 (so the windows start in 1926). History comes from Robert Shiller's monthly price and dividend data, extended to today with the S&P 500 Total Return index. The current month uses the latest close.
Recession shading: USREC (NBER US recession indicator, monthly, 0/1). The National Bureau of Economic Research dates US business cycles — they're the official arbiter of when a recession started and ended. Shaded regions on every chart mark those recessions.
Reminder: None of these signals predict recessions with certainty. The yield curve stayed inverted for two years (2022-2024) and no recession materialized. Credit spreads can stay tight for years while the market keeps rising. Use these as context, not as buy/sell triggers. Always do your own research or talk to a financial professional.