Bad, Bad September

I read a very interesting article a few days ago about why markets so often crash in autumn. It lays out an older historical mechanism and several modern explanations. I want to focus on five of the modern ones.

Before putting its question next to two SPY seasonality charts, there is an important distinction. The article is mainly trying to explain volatility. It reports that since 1990, the median VIX rose 1.5 points from late August into September and another point in October. The SPY charts ask a different question: what happened to returns?

The strange part was not that September looked bad. It did.

The strange part was how.

Across 1993–2026, September was the only month with a negative average return: −0.44%. Yet SPY rose in 56% of those Septembers. In the 2017–2026 panel, the contradiction became sharper. September rose 60% of the time and still averaged −1.18%.

September usually won.

Its losses were just bigger.

That distinction matters because a weak average, a frequent decline, and a crash are three different things. Calendar folklore tends to pour them into one pumpkin-shaped bucket. The data will not cooperate.

The article’s deepest historical answer comes before its modern theories. Before the Federal Reserve, financing the harvest and shipment of crops increased autumn demand for cash and credit. An inelastic currency could not expand quickly, so that seasonal pull strained New York banks and sometimes helped turn a shock into a panic. The Fed was created partly to furnish an elastic currency and smooth those seasonal pressures. It reduced that particular vulnerability; it did not abolish autumn volatility.

The five explanations become more useful once we stop treating them as rival solutions to one mystery. Some may start pressure. Others delay information or change liquidity. A few merely amplify a fall that already has a reason.

1. The Ghosts of Crashes Past

October has excellent branding. The Dow fell almost 13% on October 28, 1929 and almost 12% the next day. On October 19, 1987 it lost 22.6% in one session, still its largest one-day percentage fall. Lehman Brothers failed in September 2008, followed by a brutal October. The Crash Narratives paper cited by the article finds that crash stories propagate after entering the news and predict market volatility.

We remember these dates, buy protection, reduce risk, and become quicker to see an ordinary wobble as the opening scene of a disaster. That can create feedback. It cannot be the original explanation. Memories of Black Monday did not cause Black Monday, and the 1987 decline had real mechanics: portfolio insurance forced more selling as prices fell, while market structure struggled with the order imbalance.

The ghosts can shout fire.

They did not build the theatre.

2. SAD Investors

The most elegant explanation is biological. Shorter days worsen mood for some people; depressed mood can raise risk aversion; investors then demand less equity exposure as autumn darkens.

There is serious research behind it. A 2003 American Economic Review paper found return patterns linked to daylight, latitude, and opposite seasons in the two hemispheres. There is also a serious dispute. A later re-examination argued that the result was driven by model specification and the turn of the year; the original authors replied that the critique mishandled the tests.

I would not throw the idea out. I would also not ask the length of the day to explain why a particular credit structure breaks on a particular Tuesday. SAD is a plausible background nudge, not a convincing crash clock.

3. The Fiscal Year Effect

Institutional calendars can force real trades. After U.S. mutual funds moved to a common October 31 tax year, researchers found that funds accelerated sales of losing positions before the deadline. Other research found managers buying stocks they already held near quarter-end, temporarily pushing those prices up. That is two-way price pressure with paperwork behind it, not market astrology.

But managers later spread tax-motivated sales over longer periods to reduce their own impact. Different funds, companies, governments, and investors also face different reporting and tax dates. Fiscal incentives can increase volatility around particular holdings. They struggle to explain a broad market crash by themselves.

4. The Summer Vacation Effect

This is where the evidence gets more interesting. A study of 47 countries found that market returns in the month after major school holidays were 0.6% to 1% lower than in other months. The pattern did not depend on September alone. It moved with local holiday calendars and was strongest where negative news arrived during the holiday and trading had been light.

The proposed mechanism is not that traders return from the beach in a bad mood. They return to an information backlog. During the holiday, attention is scarce and bad news is harder to trade on, especially when shorting or arbitrage needs work. September becomes the month when delayed disagreement finally gets a price.

The source article adds a more specific news-flow channel. David Yermack used corporate-aircraft trips to identify CEO vacations. Companies released less news and delayed some mandatory disclosures while the CEO was away; stock volatility fell during those absences and rose after the CEO returned. Vacation may therefore affect not only investor attention, but also when corporate information reaches the market.

5. Trading Volume

Volume is probably not a separate cause so much as the gearbox in the vacation story. Evidence from 51 markets shows that summer turnover falls, both large and small investors trade less, and bid-ask spreads widen. Thin trading can delay price discovery. When full desks return, higher volume reveals orders and disagreements that were already waiting.

High volume does not predict a fall. It accompanies rallies too. Low volume does not cause a crash. The useful point is that changing liquidity can alter how quickly accumulated information hits prices and how far an imbalance travels before buyers appear.

In my view, the vacation-attention mechanism and the volume transition form the strongest modern pair. Fiscal calendars can add two-way pressure around the edges. SAD may change the background level of risk appetite. Crash memories can make the reaction faster. The common outcome can look seasonal even when no single seasonal force is in charge. Modern SPY does not trade inside the banking system of 1907, which is a useful warning against giving one permanent cause to a pattern spanning different financial regimes.

The recent chart adds another warning. February averaged −0.40% in 2017–2026 and March −0.85%, even though their 1993–2026 averages were 0.00% and +0.97%. September has kept its weak average, but it now has competition from late winter.

Ten observations per month are not much of a jury. The panel also includes an unfinished 2026, with September still in progress when the chart was captured. The pandemic crash alone can drag February and March into a different rank. A few large events can do the same to September.

This does not make seasonality useless.

It changes the job we should give it.

A weak September average can be a prompt to check leverage, liquidity, breadth, valuations, earnings revisions, credit conditions, and policy risk. It may justify greater attention when those signals already look fragile. It does not justify selling merely because the calendar changed its page. This is also where I agree with the source article: predictable volatility can make autumn worth monitoring without making a seasonal exit rule profitable.

For that broader set of current signals, I use the State of the Market. Autumn may tell us when to look more carefully. It still cannot tell us what we will find.