The S&P 500 closes 50% below its high

Where things stand (13 Oct 2026)

The S&P 500 set a record close of 7,818.93 on 6 Oct 2026. The latest close is 7,807.24 (12 Oct), about 0.15% below that high (FRED SP500 and Yahoo ^GSPC agree). For YES, the index would have to close at or below 3,909.47, and that line moves up whenever a new high is set. Short-term options volatility is low (CBOE 30-day implied vol of 12). No close since the 16 Sep opening date comes anywhere near the line.

Base rate (computed from Yahoo ^GSPC daily closes, 1928 to 2026)

• A close 50% or more below the prior record has happened in only two episodes: 1929–32 and 2007–09. In both, the first such close came about 13–14 months after the peak (10 Nov 1930 and 20 Nov 2008).
• Two declines came close but fell short: 2000–02 reached 49.15% and 1973–74 reached 48.2%.
• Looking at days at or near a record and asking whether a 50% close followed within 5.2 years: about 1.5% of such days since 1950 and 6–8% since 1928. Both figures depend heavily on those near-misses.
• Monthly data back to 1871 (Shiller) add probably one more pre-1928 episode (1872–77). Overall that is roughly 2–3 events in 100–155 years, a rate of about 0.015–0.025 a year. Over this window that works out to roughly 5–10%.

Models

• Volatility model. I fitted a standard volatility-clustering model (GJR-GARCH(1,1) with skewed-t errors) to daily returns since 1950. Over simulated 76-year runs it reproduces the observed number of declines fairly well: 10.0 vs 11 actual for 20%+, 4.6 vs 6 for 30%+, 2.3 vs 3 for 40%+, and 1.2 vs 1 for 50%+. It slightly under-counts the large ones.
• Simulating from today's state to 31 Dec 2031 gives 7.8–11.3% for the 1950 fit (annual drift from 8% down to 2%). Fitting from 1928, which includes the Depression, gives 11.1–15.2%.
• Options. From the CBOE SPX option chain (12 Oct 2026), the market-implied chance of the index ending below about 3,904 is 0.1–0.2% by Dec 2026, 2.1% by Dec 2027, 3.9% by Dec 2028, 5.6% by Dec 2029 and 7.6% by Dec 2030.
• When I scale the volatility model to match those option prices, the implied chance of a 50% fall from the record by end-2031 is about 30–35%. Option prices build in a premium for crash insurance, so real-world odds are usually a quarter to a half of that. That points to roughly 8–16%.

Conditions now

These point to more risk than usual:

• The cyclically adjusted price-to-earnings ratio (CAPE) is about 41, the second-highest since 1871.
• The 10-year Treasury yield touched 5.36%, its highest since 2002.
• The Fed raised rates on 16 Sep and signalled another hike. Its room to cut in a crash would be limited.
• Brent oil is above $100 because of the Iran war, and there is stress in government bond markets worldwide.
• Spreads on the lowest-rated (CCC) corporate bonds are at 4-year highs, and $18bn of Oracle data-center loans trade at 89–91 cents on the dollar.
• AI spending is increasingly paid for with debt.

On the other side, the most bearish strategist I found (Klement at Panmure Liberum) targets 5,000 by end-2027, a 36% fall, while the consensus expects about +14%. Big-tech earnings are real, and policymakers have a record of stepping in.

Paths

• YES: the AI spending boom collapses during a period of high inflation and high rates, possibly with a credit event, and the decline overshoots the 2000–02 and 1973–74 near-misses.
• NO: no bust within the window, or a bust that stops at 30–49%, as the closest historical parallels did.

How the forecast behaves over time

• A 50% fall within a single quarter starting from a record has never happened, so the chance by Dec 2026 is about 0.05%.
• The quarterly chance builds through 2027, since a crash needs about a year or more to reach 50%, and is about 0.8% per quarter in 2028–29.
• It then stays at roughly 0.75% per quarter, giving about 12.7% by end-2031. This tracks the 1928-fitted volatility model with a 4% drift and sits between that model and the options-based estimate.

Strongest objection

The two closest parallels, the 2000 tech bubble (CAPE 44) and the 1973 oil-shock bear market, both stopped just short of 50%, and since 1950 only the 2008 banking crisis got past the line. On that view the answer is about 6%. The opposite view, that record valuations combined with limited room for policy support justify 20% or more, is about as strong. My estimate sits between the two.

Sources

• FRED SP500 (fred.stlouisfed.org/series/SP500); Yahoo ^GSPC daily closes 1927–2026; CBOE SPX delayed quotes (12 Oct 2026); Shiller ie_data.xls.
• Reuters/CNN, 6 Oct 2026 (record close); NBC, 7 Oct 2026 (yields 5.36%); Guardian, 20 Sep 2026 (CAPE about 41, Fed hike); Financial Post/Bloomberg, 8 Oct 2026 (Klement target); AP/CNBC, 16 Sep 2026 (Fed hike).

Question. By the last day of each quarter, has the S&P 500 closed 50% or more below its prior all-time closing high?

Definitions.

• S&P 500: S&P Dow Jones Indices’ standard capitalization-weighted S&P 500, USD price-return index, not a total-return, equal-weight, inflation-adjusted, futures, or ETF series.
• Close: the official daily end-of-session index level for a U.S. trading day, including shortened sessions. Intraday levels and indicative or preliminary closing estimates do not count.
• Prior all-time closing high: the highest official daily closing level in the index’s entire published history before the trading day being tested. The high may precede the question’s opening date and is updated whenever a higher close occurs; it is not fixed at the opening-date high. Any purely technical rebasing must be applied consistently to both values.
• 50% or more below: for tested close C and prior high H, C ≤ 0.50 × H, using published index levels without rounding the percentage drawdown.

Resolves YES if a qualifying official close for a trading day after the opening date meets that inequality and is published by the quarter-end. A single qualifying close is sufficient; no persistence or AI-related cause is required. Apply the Observation period and Resolution window conventions. Historical highs from before the opening date may be used solely as the comparison baseline.

Resolves NO if no such close has been published by the quarter-end. Intraday crashes that recover above the threshold by the official close, drawdowns from intraday highs, declines confined to other indexes or investment products, and qualifying closes only from before the observation window do not count. Trading suspensions without an official closing value do not themselves satisfy the threshold.

Resolving source. S&P Dow Jones Indices’ official S&P 500 index page, official daily index-level files, or official historical data. The Federal Reserve Bank of St. Louis’ FRED series SP500, sourced to S&P Dow Jones Indices and specified as “Daily, Close,” is an acceptable republication; S&P Dow Jones Indices controls any discrepancy. FRED’s rolling ten-year history must not be treated as the definition of all-time history. This named statistical source governs rather than Public confirmation (A2). Apply Resolution date (A3): use the publication date of the qualifying official value, not a retrospective article’s date. An official correction that first establishes a qualifying close resolves on its publication date, provided the observation itself is eligible; corrections solely to pre-opening observations cannot trigger YES. Apply the Resolution window convention, including its once-YES rule.