Calendar Anomalies: Sell in May, the January Effect, and 300 Years of Evidence

An Old Adage That Refused to Die

On the floor of the London Stock Exchange in the early nineteenth century, brokers who could afford it would leave the city in late spring, retreat to their country estates, and not return until the autumn social season began. The phrase they left behind — "Sell in May and go away, come back on St. Leger's Day" — was practical advice anchored in human behavior, not academic theory. St. Leger's Day, a horse race held in September, marked the point when wealthy Londoners returned to the city and financial activity resumed.

For most of the twentieth century, this seasonal adage was treated as colorful folklore, the kind of advice that no serious investor would entertain. Then Bouman and Jacobsen (2002) examined 37 equity markets with data stretching back as far as 1694 and found something that shook the academic consensus: the Halloween effect was real, statistically significant, and present in 36 of 37 countries studied. The puzzle was not that it existed in one or two markets. The puzzle was its global ubiquity.

This article compares the Halloween indicator with the January Effect — two distinct seasonal anomalies with different origins, different magnitudes, and different fates after academic publication. Examining them side by side reveals what calendar patterns can and cannot tell us about market efficiency.

The Halloween Indicator: What Bouman and Jacobsen Actually Found

Bouman and Jacobsen defined the Halloween indicator precisely: average equity returns for the November-through-April period significantly exceeded average returns for the May-through-October period in the overwhelming majority of countries they examined. Their primary dataset covered 37 stock markets. The effect was statistically significant in 36 of them.

The strategy they tested was simple. An investor switches from equities to bills in May and back into equities in November. Across their full sample, this rotation improved average annual returns relative to a buy-and-hold strategy in the majority of markets, while also reducing risk as measured by standard deviation. The United Kingdom, with data beginning in 1694, showed the effect persisting across centuries with no evidence of decay.

Several features of their findings stand out.

The magnitude varied considerably across markets. Developed markets showed differences between summer and winter half-years averaging 6-8 percentage points annualized. Emerging markets showed even larger spreads in some cases, though with greater statistical uncertainty given shorter data histories.

The pattern was not concentrated in a small number of extreme months. May, June, and September each contributed to below-average summer returns rather than any single month driving the result. Similarly, November and December were among the stronger winter months, but the effect distributed across the full November-April window.

The effect proved robust to various specifications. Bouman and Jacobsen tested whether it was driven by January alone, whether it disappeared in the later portion of their samples, and whether transaction costs eliminated it. The Halloween rotation remained profitable even after accounting for conservative estimates of transaction costs in most markets.

Jacobsen and Zhang (2013) extended the analysis using three centuries of monthly UK stock return data. Their central finding was sobering for those hoping the anomaly would eventually self-destruct: neither the magnitude nor the statistical significance of the Halloween effect showed any trend toward disappearance over the entire 300-year sample. An anomaly that has survived three centuries of observation, including the modern era of hedge funds and algorithmic trading, is not behaving like a simple statistical artifact waiting to be arbitraged away.

The January Effect: A Different Animal

The January Effect carries a different history. Wachtel, writing in 1942, first observed that January tended to produce unusually strong returns. Keim (1983) formalized the phenomenon and tied it specifically to small-capitalization stocks. His analysis showed that roughly half of the annual small-firm premium documented by Banz (1981) accrued in January alone, concentrated in the first few trading days of the new year.

The proposed mechanism was intuitive. Investors sell losing positions in December to realize capital losses for tax purposes. This selling pressure drives down prices of small, illiquid stocks disproportionately. In early January, these stocks snap back as the selling pressure evaporates and new capital enters the market. The tax-loss-selling story predicted several testable features: stronger effects in countries with end-of-calendar-year tax years, larger effects in smaller and more illiquid stocks, and potentially stronger effects following years with many losing positions to harvest.

The evidence largely supported these predictions. The effect was strongest in the United States and other markets with December fiscal year-ends. Small-cap stocks exhibited dramatically larger January returns than large caps. Thaler (1987) documented the anomaly's features in the context of behavioral economics, noting that the predictability itself raised questions about why sophisticated investors did not trade it away.

Here the two anomalies diverge sharply.

Publication, Decay, and Persistence: A Divergent Fate

The January Effect and the Halloween indicator have followed strikingly different post-publication trajectories. This contrast matters enormously for understanding what calendar anomalies can teach us about market efficiency.

The January Effect attenuated substantially after publication. Haugen and Jorion (1996) examined January returns in U.S. markets through 1993 and found that while the effect persisted, its magnitude had declined relative to the pre-publication period. The small-cap January premium shrank as institutional investors, aware of the pattern, began positioning ahead of December, driving up year-end prices and dampening the January bounce. By the 2000s and 2010s, the raw January small-cap premium was a fraction of what it had been in the 1970s.

This decay is consistent with the efficient market hypothesis in its semi-strong form. Once information about the January Effect became public knowledge, rational arbitrageurs competed to exploit it, and in doing so, they largely eliminated it. The anomaly was genuine — but it was also arbitrageable, and it got arbitraged.

The Halloween effect has been far more stubborn. Maberly and Pierce (2004) argued that the anomaly was driven by outlier observations — specifically, the August-September 1998 collapse associated with the Russian debt crisis and the LTCM near-failure. Remove those months, they argued, and the effect was weaker. Bouman and Jacobsen responded that such retroactive exclusion of unusual months was itself a form of data mining; the full sample, unusual months and all, is what investors actually experience.

Subsequent research has confirmed that the Halloween indicator persisted in the post-publication period. For a pattern documented in centuries of data across dozens of countries to be merely a statistical artifact stretching back to 1694 requires extraordinary coincidence. The more defensible conclusion is that something structural is generating it.

Competing Explanations: Risk, Behavior, and Institutional Patterns

Neither anomaly has a fully satisfying explanation, and the explanations differ in ways that illuminate the mechanisms.

For the January Effect, the tax-loss-selling story is reasonably compelling but incomplete. It explains why small illiquid stocks should rebound, but it does not fully explain why institutional investors — who can observe this predictable pattern — do not front-run it to the point of elimination. The behavioral component involves window-dressing: portfolio managers sell underperforming stocks in December to avoid year-end disclosure of embarrassing holdings, then buy them back in January. This institutional behavior amplifies the price pattern independent of retail investor tax considerations.

The Halloween indicator resists any single clean explanation. Candidate mechanisms include:

Vacation effects and reduced trading volume. During summer months in developed markets, institutional activity declines as traders take holidays. Thinner markets can generate lower average returns through reduced price discovery efficiency and higher transaction costs. But reduced volume should increase volatility, not mechanically reduce returns.

Risk exposure shifts. One interpretation is that sophisticated investors systematically reduce equity exposure during summer months, perhaps in response to elevated uncertainty during earnings-light periods. Lower average equity exposure during summer would translate directly to lower average returns, but this merely relabels the anomaly rather than explaining why the risk shift itself occurs.

Behavioral explanations emphasize mood seasonality and attention cycles. Hirshleifer and Shumway (2003) documented a link between weather and stock returns, with sunshine correlating positively with market returns. Kamstra, Kramer, and Levi (2003) proposed a seasonal affective disorder mechanism: shorter days in autumn and winter affect investor mood, leading to risk aversion in autumn followed by optimism as days lengthen from January onward. These mechanisms are speculative and their economic magnitudes are debated.

Three Centuries of Evidence: What Persistence Means

The longevity of the Halloween indicator creates a genuine puzzle. McLean and Pontiff (2016) showed that anomalies decay after publication on average, as the academic literature effectively broadcasts trading signals to the arbitrage community. Calendar anomalies should be among the easiest to exploit: they require no proprietary data, no complex modeling, and the timing is known years in advance.

Yet the Halloween effect has not behaved like the typical anomaly. Three potential explanations deserve consideration.

First, the trading costs of implementation are non-trivial for certain investor classes. Tax-exempt institutional investors can theoretically rotate between equities and bills at minimal friction, but retail investors face transaction costs, taxes on realized gains, and the behavioral difficulty of moving against the market during summer months when equities are still generating some positive return.

Second, the effect may be compensation for a genuine seasonal risk premium. If macroeconomic risks are higher during summer months — perhaps because corporate guidance is less frequent, or because political event risk concentrates around summer recess periods — then the equity risk premium during summer may simply be lower, reflecting a genuine reduction in the compensation required for risk-bearing.

Third, the behavioral mechanism may be self-reinforcing. If enough institutional participants believe in the effect and act on it, their summer risk reduction becomes the mechanism that perpetuates the pattern. This is a coordination equilibrium rather than a mispricing, and it need not disappear even when widely known.

What Has Survived and What Has Not

Three decades of post-publication scrutiny have produced a fairly clear assessment.

The January Effect, in its original small-cap form, has largely decayed in the United States and other developed markets. A modest January premium may persist, but its magnitude falls far short of the pre-publication estimates. Investors hoping to systematically harvest January small-cap premiums will find that their competitors got there first.

The Halloween indicator, by contrast, retains its statistical signature. The rotation strategy — equities November through April, short-term fixed income May through October — continues to show positive risk-adjusted performance in most developed markets. Whether this represents an exploitable alpha opportunity or a persistent risk-premium differential is a matter of interpretation, not a matter of whether the pattern exists.

The practical question for investors is whether a simple seasonal rotation makes sense within their portfolio. The historical evidence suggests that rotating to lower-volatility assets during the May-October period can reduce drawdown exposure without proportionally reducing long-run returns. This is not magic — it reflects the empirical observation that the worst equity drawdowns (the summer of 2008, the summer of 1998, the summer of 2002) have disproportionately concentrated in the weak half-year.

What the evidence cannot tell us is whether the next cycle will look like the historical average or like the exceptions. In 2020, the worst equity decline arrived in February and March — squarely in the Halloween period's supposedly favorable window. No calendar rule eliminates risk entirely. The most honest assessment is that the Halloween effect represents a robust historical pattern with a disputed mechanism, and that exploiting it requires accepting the distinct possibility that any given year will violate the seasonal expectation.