The Evidence
The Warning Lights
History's bubble peaks were preceded by measurable signals. Here is each one in plain language: what it means, where the danger zone starts, what it reads now, and whether it is flashing.
Updated
Two findings from the historical record frame everything below. First, bubbles typically peaked within 0 to 18 months of central-bank [tightening](/glossary#tightening): in six or seven of eight major bubbles since the 1840s, the top came during or shortly after a rate-hiking cycle (NBER (opens in a new window)). Second, even correct consensus warnings ran 12 to 24 months early, and early expert warnings ran earlier still; acting on them cost real money before it saved any (Brunnermeier (opens in a new window)). A dashboard tells you about fragility, not timing. How these readings roll into the scorecard and the scenarios is on Conclusions.
Statuses below are as assessed in the underlying research (readings 2025 through mid-2026): flashing means at or beyond historical danger levels, amber means elevated or mixed, quiet means within normal range, and not scoreable means no published aggregate metric exists behind the indicator. Valuations and insider selling are the two not-scoreable indicators; each would become scoreable with a named data source and methodology, a locked vendor multiple and earnings basis for valuations, a defined issuer basket and aggregation method for insider selling.
Market concentration & breadth
FlashingBreadthHow many stocks are actually participating in a rally. Narrow breadth (a few giants carrying the index) has historically been a late-cycle warning.Full definition in the glossary asks how many stocks are actually rising; concentrationHow much of a stock index's value sits in its few biggest companies. Record-high concentration means index investors are making a bigger bet on fewer firms than they may realize.Full definition in the glossary asks how much of the index sits in its giants. Narrow rallies led by a few names preceded the 1929 and 2000 peaks by 6-18 months.
- Danger zone
Top-10 share of the S&P 500 above roughly 25-30%, with equal-weightA version of a stock index where every company counts the same, instead of big companies counting more. Comparing the two shows whether gains are broad or concentrated.Full definition in the glossary persistently lagging.
- Current reading
Top-10 weight sits in the high 30s in fund-holdings data: 37.6% by ticker, 39.1% counting Alphabet's two share classes as one company (SPY holdings, 27 Aug 2026), below the site's earlier 40.7% "record" claim, which no located methodology reproduces. Equal-weight has lagged cap-weight badly since 2023 (Apollo (opens in a new window)). Beyond prior-era norms; the research calls this the strongest current warning.
Margin debt & leverage
AmberMargin debtMoney investors borrow from their brokers to buy stocks. It amplifies gains on the way up and forces selling on the way down.Full definition in the glossary is stock-buying with borrowed money. It amplifies rises and forces selling in falls; surges of 50-60% over two years preceded the 2000 and 2007 peaks.
- Danger zone
Rapid two-year growth above ~50-60% and record levels relative to the market's size.
- Current reading
Record nominal levels: FINRA's own series set an all-time high of $1.502 trillion in June 2026, the maximum of the full 1997-2026 series, then fell $84.8 billion (-5.65%) in July to $1.417 trillion, the largest one-month dollar drop in that 29-year record, still up about 38.6% year over year (FINRA (opens in a new window)). The earlier "~1.9% of market value" denominator does not trace to FINRA's data or any located source and is dropped rather than restated. Amber, not red.
Insider selling
Not scoreableExecutives selling their own stockExecutives and directors selling shares of their own company. Regularly seen before market tops, but noisy: insiders sell for many innocent reasons too.Full definition in the glossary clustered before past tops, though insiders sell for many innocent reasons, so this signal only matters alongside others.
- Danger zone
Heavy net selling clustered in the boom sector during parabolic price moves.
- Current reading
Sustained AI-winner insider sales are real (Nvidia, Oracle, Palantir, 2024-25, much of it on pre-set 10b5-1 schedules) but no aggregate metric exists against any defined issuer universe or historical baseline. Would become scoreable with a named issuer basket and an aggregation methodology; none is published anywhere.
New issuance
AmberNew investment productsThe flow of newly created investments being sold to the public: IPOs, new bonds, new funds. Issuance waves tend to crest right around market tops.Full definition in the glossary, IPOsWhen a private company first sells shares to the public. Waves of speculative IPOs tend to cluster near market tops.Full definition in the glossary, bonds, funds, crest when promoters can sell anything. Issuance peaked within 0-18 months of the 1929, 2000, and 2021 tops.
- Danger zone
A wave of speculative issuance concentrated in the hot theme.
- Current reading
Public AI IPOs elevated but far from 1999's blow-out. Bloomberg's (opens in a new window) reporting puts private credit into AI data-center vehicles and neoclouds at an estimated $50 to 80 billion, but no regulator filing, ratings-agency methodology, or consolidated dataset was located behind that range in this pass, so it is presented as Bloomberg's own reporting, not an independently verified aggregate, and is not scored here as a figure. Hot in private credit by reputation, cooler in public equities: amber.
Credit spreads
Quiet (sector amber)Credit spreadsThe extra interest a risky borrower pays compared to the safest borrower (the US government). Widening spreads mean lenders are getting nervous.Full definition in the glossary are the extra interest risky borrowers pay. They widened months before the telecom and housing busts, one of the better-timed signals when it fires.
- Danger zone
Sustained widening of 50-100+ basis pointsHundredths of a percentage point. 50 basis points = 0.50%. Used for small moves in interest rates and bond yields.Full definition in the glossary in the boom sector's debt.
- Current reading
Broad investment-grade and high-yield spreads tightened, not widened, in the second half of August 2026 (IG 81bp to 79bp, HY 270bp to 263bp, 17-27 Aug), the opposite of the "modest widening" this page previously described (FRED (opens in a new window)). A narrower claim that AI-adjacent credits are wider than the broad market has no named issuer basket behind it and remains untestable either way. Credit markets are not yet worried, which is either reassuring or complacent.
Central-bank policy
MixedThe single most consistent bubble-ender: peaks cluster during or within ~18 months of tighteningWhen a central bank raises interest rates or drains money from the system to slow the economy. Most historical bubbles peaked during or within 18 months of tightening.Full definition in the glossary. Easing after big hikes has cut both ways, sometimes fueling one last leg up (1998-2000).
- Danger zone
A tightening cycle in progress, or its first 18 months of aftermath.
- Current reading
The 2022-23 hiking cycle (to 5.25-5.5%) gave way to cuts from September 2024; the funds rate has held near 3.5-3.75% through mid-2026 (Federal Reserve H.15 (opens in a new window), June 2026). Early-to-mid easing after aggressive hikes: historically compatible with both a melt-up and a late-cycle plateau. Genuinely mixed.
Valuations
Not scoreablePrice-to-earningsPrice divided by yearly earnings: how many dollars you pay for one dollar of annual profit. Higher means more expensive, or more optimism about growth.Full definition in the glossary and price-to-salesCompany value divided by yearly revenue. Used when profits are small or absent. Above 10 is historically expensive; some AI names trade far above that.Full definition in the glossary extremes accompanied every equity bubble, but they measure risk, not timing: expensive markets can stay expensive for years.
- Danger zone
Leaders above 60-100x forward earnings or 10-30x sales, dot-com territory.
- Current reading
No named consensus-estimate vendor, timestamp, or earnings basis (GAAP vs. adjusted) is specified for a "Magnificent-7 forward P/E" reading, so no current figure can be certified. Would become scoreable with a locked vendor (for example LSEG/I-B-E-S or FactSet), a stated horizon, and a named earnings basis. Nvidia's price-to-sales above 25 at peaks and the 1999-2000 100x-earnings comparators are historical reference points, not re-verified in this pass.
Capex vs operating cash flow
Amber / red in spotsWhen a sector's building spreeMoney a company spends on long-lived physical things: buildings, machines, chips, data centers. Spent now, paid back (hopefully) over years.Full definition in the glossary approaches or exceeds the cash its operations generate, and gets funded with debt and off-balance-sheet vehiclesA separate legal company created to hold one project and its debt, keeping both off the parent company's books.Full definition in the glossary, it matches the railway, telecom, and fiber signatures.
- Danger zone
Capex persistently above ~100% of operating cash flow, funded by rising leverage.
- Current reading
Hyperscaler capex has risen sharply against operating cash flow across 2024-25 in each company's own filings, but no consolidated cross-company aggregate exists for a single "Big-5" dollar or percentage figure; the site's earlier $356B-to-$433B debt aggregate and 50-60%-of-OCF ratio are dropped as unsupportable rather than restated. Oracle's negative free cash flow through 2029 is a market forecast, not an observed reading, and smaller AI builders leaning on debt and SPVs remains the overbuild signature, in spots.
Credit-to-GDP gap
QuietThe credit-to-GDP gapHow far total lending in an economy has risen above its long-term trend. Readings above about +10 points have preceded many banking crises.Full definition in the glossary measures whether economy-wide borrowing is running above trend, the best-tested predictor of banking crises, including 2008.
- Danger zone
Roughly +10 percentage points above long-run trend (BIS/IMF threshold).
- Current reading
The US gap has been negative continuously since 2021 Q4, at -11.5 percentage points in 2025 Q4, the latest available reading, ranging roughly -7.6 to -12.6 points through 2023-2025, not modestly positive as this page previously stated (BIS (opens in a new window)). A negative gap signals credit growing below trend, not excess credit; the sign matters here, not just the level. The clearest quiet light on the board, with one caveat: AI's leverage sits substantially in private structures this aggregate measures poorly.
Reading the board
The pattern is not "everything is flashing." Concentration is at records, leverage and issuance are hot in places, and the capex signature is visible in the weaker builders, while the economy-wide credit measures that preceded 2008 stay calm and credit markets remain relaxed. That mixed board is roughly what 1998 looked like: sector-level excess without a system-wide credit boom. It is also what a durable boom looks like early on, which is why the other reading of the same board deserves its page. What the board cannot do is tell you the date; historically, even correct alarms rang one to two years before the top.