S&P 500 crash check

Today the market is not inside any of the four groups. The macro backdrop (oil, rising yields, a hiking Fed) looks like the fast-correction episodes, but price has not yet done what those episodes did. The chance that the coming week marks the start of a drawdown deeper than 15% is about 1%, roughly twice the 0.5–0.7% weekly base rate. A fall of 3% or more at some close during the week is about 7% likely.

Where price stands

The index sits just under its 13 August high, above every moving average, with volatility low and credit calm. That is the state of the market in the weeks before every one of the thirteen peaks, so it does not tell the two futures apart.

What the macro backdrop says

This is where today differs from the tops of the past. At the peak of the thirteen episodes, yields and oil were flat or falling. Today both are rising faster than at any of them.

Which phase does it resemble

Distance between today and the two weeks before each episode’s peak, on five technical and four macro measures. Shorter bar means more similar.

• Closest: Gulf War 1990 and Q4 2018, both fast corrections. One was an oil shock, the other a Fed tightening with the 10-year rising. Today has both ingredients.
• Furthest: Covid, the financial crisis, dot-com and LTCM. Those started from different valuation, credit or volatility conditions.
• The rate shock of 2022 is in the middle (0.81). It is the extended-bear template, but the policy stance then was far looser at the peak.
• The single closest day in the whole 1990–2026 history is 12 September 2023, with the 10-year heading to 5%. It was followed by a fall of about 10%, a case the study left out because it stayed under the 15% threshold.

The test price has not passed

If 13 August were the top of a real episode, the index would probably already be much lower. Drawdown from the peak 25 sessions later, for each episode, against where the market is today.

Every one of the ten real episodes was already 3.9% or more below its peak at this point. The market today, at −1.9%, looks like the controls (China and oil 2015, carry unwind 2024). Once a genuine episode starts, the first week is usually mild: five sessions after the peak the twelve non-Covid episodes were between −0.6% and −3.9%. So a week of small losses would not confirm anything.

Odds for 21–25 September

Worst daily close during the week, measured from Friday’s 7,650.5. Estimates blend a volatility model, a model that adds rates and oil, the option-implied move, and the historical days most like today.

Volatility alone gives about 5% for a −3% close. With rates and oil added it is 5.4%, and the option-implied move gives 7%. The 25 most similar historical days point higher, about 20%, but that is only 5 cases. The estimate takes the top of the model range on purpose, because rates and oil are unusual. Out of sample, adding them to the volatility model did not improve it. One-week range implied by the VIX: about ±157 points (7,494–7,807); by realised volatility: about ±102 (7,548–7,753).

Probability that the week is the start of an episode

Base rate is the number of episodes in the group divided by 36.7 years of weeks. The uplift for today is a judgement, not a fitted value: thirteen episodes are too few to measure it. An extended bear cannot be recognised within a week; it can only be confirmed months later. On a three-month horizon, in days that looked like today (high 4–6 weeks old, price 1–3% below it, above the 200-day), about 7% were followed by a fall of 15% or more from that high.

Hedge Gauge on 18 September: USA100 52%, NASDAQ100 35%. A middle-of-the-range reading, consistent with this picture rather than an alarm.

What would raise the odds

None of these has triggered. Any of them would move the odds towards the fast-correction group, whose entry trigger the study shows is the most likely to arrive late.

• S&P 500 close below 7,617 (50-day), then 7,521 (100-day). The 200-day is 7,183, 6% lower, where a moving-average trigger would fire, mid-decline. A new high above 7,799 would take the market back to the pre-peak state.
• VIX above 20, or the 9-day VIX rising above the 30-day one. Both would show that the calm has broken.
• High-yield credit turning down against Treasuries. Credit is the cleanest early warning for the deeper groups, and it is quiet today.
• 10-year Treasury above 5.1%, or WTI above $110. The two macro drivers that separate today from past tops.
• Semiconductors (SMH) closing below their 200-day average, about 487, with equal-weight breadth still falling.
• USD/JPY falling sharply. That would be the carry-trade unwind pattern of August 2024, which the study classes as a control, but which starts the same way.

Method and limits

• Data: S&P 500 daily closes 1990–2026, VIX, VIX3M, VIX9D, SKEW, Treasury yields, 3-month bill, WTI, dollar, sector and credit ETFs, all from Yahoo Finance to 18 September 2026.
• Probabilities are statistics, not forecasts. They come from overlapping daily windows, so the sample is smaller than it looks, and the model gains from rates and oil disappear out of sample.
• The similarity ranking rests on thirteen episodes. It says which past setups look most like today, not that history will repeat.
• Macro context comes from news reports (listed below) and was not checked against primary releases. The economic calendar for 21–25 September was not verified; the likely catalysts are Iran headlines, US inflation data and the 5% level on the 10-year.
• The technical peak date, 13 August 2026, is the highest close in the index data. The probability of a drawdown deeper than 15% is measured from that level.

The market is in a state that has been the calm before a fast correction and also before nothing at all. What is new is the macro: yields and oil are rising faster than at any past top. That puts the odds of a break next week a little above normal, and still small. What separates the two paths is a close under the 50 and 100-day averages, not the backdrop.

Sources for the macro context: CNBC, Fed decision, 16 Sep · Detroit News, Wall Street on 18 Sep · Bloomberg, triple witching, 18 Sep.

Research material only — not personalized financial advice. Not a recommendation to buy or sell any security. Past drawdowns do not predict future ones. Contact: domderrico@gmail.com.