Will NASDAQ fall at least 65% from its historical peak at any time before the end of 2030?

TL;DR

I put the probability of YES at 9%. The historical backbone is one qualifying crash in the full 1971–2026 Nasdaq daily record, with the 1973–74 near-miss stopping at 59.9%. AI concentration and debt-linked capital spending raise the tail, but strong earnings, loose broad financial conditions, and resilient bank capital keep NO the clear favorite.

Context

The Nasdaq Composite closed at 26,180.45 on August 21, 2026, versus its record closing high of 27,093.90 on June 2, 2026. It was 3.37% below that peak. If no new high is set, the qualifying close is 9,482.865, and the index must fall another 63.8% from its August 21 close. These figures come from the complete FRED/Nasdaq daily-close series, covering February 5, 1971 through August 21, 2026, with 14,004 numeric observations and an August 21, 2026 data vintage.

The rule is path-dependent. A rally to new records in 2027–29 would raise the threshold in index points, after which a 65% decline could still qualify before December 31, 2030. There are 4.356 years remaining, enough time for a speculative boom and a multi-year decline like the 31-month dot-com collapse.

Evidence

The historical record is stark. Only the 2000–02 collapse breached 65%: the index fell from 5,048.62 on March 10, 2000 to 1,114.11 on October 9, 2002, a closing drawdown of 77.93%. The next-largest fresh decline was 59.9% in 1973–74. The 2007–09 financial crisis produced a 55.6% local decline, while the 2021–22 bear market was 36.4% using the full daily series. The frequently published 23.7% figure for the last episode uses a later May 2023 endpoint rather than the December 28, 2022 trough of 10,213.29.

A simple event-count model gives:

$$P=1-e^{-H/T}=1-e^{-4.356/55.54}=7.5\%,$$

where $H$ is the remaining horizon and $T$ is the Nasdaq's full observed history in years. A Jeffreys-prior treatment of the unknown event rate gives 10.7%. Both are unstable because they rest on one event.

There is a useful conditioning adjustment. The first qualifying close in the dot-com episode was September 5, 2001, when the index closed at 1,759.01, below 35% of the 5,048.62 peak. It did not regain that peak until April 23, 2015, when it closed at 5,056.06. Those 13.63 years were not exposure time for a new first breach: the prior event had already happened and no new all-time high had reset the process. Excluding that dead time leaves 41.92 at-risk years, producing an event-rate estimate of 9.9%; the analogous Jeffreys estimate is 13.8%. I treat this as an upward correction, not as a complete model, because market-crash arrivals are neither stationary nor truly Poisson. The underlying closes are in the official FRED table.

The quantitative anchors are:

I give the most weight to the event and bear-severity models. I give less weight to return simulations because a 65% loss normally comes from a persistent change in earnings, valuation and credit regimes, not one or two extraordinary trading days. The resulting pre-current-conditions baseline is about 8%.

Current conditions push that baseline upward. The Nasdaq Composite factsheet dated June 30, 2026 contained 3,374 securities, but technology represented 63.51% of index weight and the ten largest securities represented 53.29%. Nasdaq projected roughly $700 billion of 2026 capital expenditure by five hyperscalers, up 60%–70% from 2025, while the BIS found that more of the buildout was moving into bonds, leases and off-balance-sheet financing. A Federal Reserve survey of 20 market contacts cited AI valuations, debt-funded capital spending and private-credit spillovers, though the Fed warned that these were respondents' views rather than the Board's forecast.

The same evidence also argues against treating this as another 2000. The Nasdaq-100—an imperfect large-cap proxy for the Composite—had 48.5% trailing earnings growth and a 24.7 next-12-month P/E on July 10, down from 25.6 at the start of 2026. Nasdaq's July review found that semiconductors fell 25% from their June high and the Nasdaq-100 lost 6.6% during July, yet broader and equal-weighted stocks held up much better. The market has already shown that part of the AI complex can reprice sharply without creating a system-wide collapse.

The latest macro-financial indicators were calm overall. Each row below summarizes the full history currently available from its cited FRED series rather than selecting a recent window:

The split between a 2.75% broad high-yield spread and a 10.35% CCC spread is an amber signal. It shows pressure at the bottom of the credit market but not a general funding seizure. The Federal Reserve's May report found elevated asset valuations and near-record hedge-fund leverage, but moderate business and household debt, high bank capital and broadly normal funding risks. Its 2026 stress test projected more than $708 billion in losses while aggregate large-bank capital still remained above required minimums.

I synthesize the historical models at roughly 8%, then add about one percentage point for today's unusual concentration, high valuation pressure and AI-financing cycle. Strong earnings and the absence of broad macro or banking stress prevent a larger adjustment. This produces the final estimate of 9%.

What's non-obvious

The raw “one crash in 55 years” calculation is slightly too low for this starting point. It counts more than 13 years after the dot-com threshold had already been breached but before the index recovered its old high. Removing that period raises the maximum-likelihood estimate from 7.5% to 9.9%. I do not use the full increase because broad-market reference classes are lower and the single Nasdaq event provides weak evidence about the true rate.

A 65% decline needs more than ordinary multiple compression. Starting from the Nasdaq-100 proxy P/E of 24.7, unchanged earnings would require the multiple to fall to 8.6. Even with a 35% earnings decline, the multiple would have to compress to about 13.3. That points to two reinforcing failures—such as an AI earnings bust plus recession or credit stress—rather than a mere disappointment in AI adoption. Long-run US research also finds that large booms raise later volatility but do not reliably predict crashes, so the existence of an AI boom is not itself enough to make YES likely. (Nasdaq valuation data; NBER US market history, 1792–2024).

Uncertainties

The main limit is sample size. There is one exact Nasdaq precedent, one 59.9% near-miss, and large differences between reasonable priors and model forms. My subjective 80% interval for the correctly calibrated probability is about 4%–17%. The lower end corresponds to strong AI monetization and continued financial resilience; the upper end requires the capital-spending boom to meet recession, sticky inflation, credit stress or another large shock.

AI spending figures are projections rather than completed investment. Lease obligations, private-credit exposures and reciprocal financing among hyperscalers, chip suppliers and AI laboratories remain poorly disclosed. Better project-level data on utilization, returns on invested capital, customer concentration and financing terms would narrow the estimate. FRED also began retaining only three years of the cited ICE credit-spread histories in April 2026, limiting long-run comparison on the weakest borrowers; that absence of accessible history is a data gap, not evidence that the current divergence is harmless.