Sell in May and Holiday Seasonality Effects

Direct Answer

"Sell in May and go away" (also called the Halloween indicator) is the claim that U.S. and international equities have historically produced most of their returns during the November-April half of the year, with May-October returns close to flat on average. Academic studies, most notably Bouman and Jacobsen (2002), found statistical support for this pattern across dozens of markets over long historical windows — but the effect has weakened and become inconsistent in many recent years, and it is not a reliable annual trading rule. Pre-holiday return effects (higher average returns on the trading day before a market holiday) show a similar pattern: documented historically, debated, and weaker in more recent data.

Both patterns are descriptive statistics about the past, not predictive signals with an agreed-upon economic mechanism, and both carry the same small-sample, multiple-testing fragility common to seasonality research generally.

Key Takeaways

• Sell in May has real historical support, but it is debated: Bouman and Jacobsen (2002) found the November-April period outperformed May-October in 36 of 37 countries studied, but later work has found the effect inconsistent or weaker after the original publication.
• Publication itself may have changed the pattern: A calendar effect that is widely known and easy to trade tends to attract capital that arbitrages away part of the edge, which is one candidate explanation for why the pattern has weakened since it became well documented.
• Pre-holiday effects are smaller and even more fragile: Early studies (Ariel 1990; Lakonishok & Smidt 1988) found higher average pre-holiday day returns, but the effect size is small and recent-decade replications are mixed.
• Six-month backtests routinely omit round-trip costs: A raw comparison of November-April versus May-October price returns ignores the two transactions per year, the bid-ask spread paid twice, and potential short-term tax treatment needed to actually capture the pattern.
• Neither pattern has a settled structural explanation: Proposed mechanisms include seasonal risk aversion and institutional trading calendars, but none commands the consensus that structural drivers like retail holiday demand or crop harvest cycles do in sector seasonality.
• Small sample size compounds the fragility: A single-country test of Sell in May over even 50 years is fewer than 50 independent "May-October" observations — far too few to rule out chance with confidence, especially once multiple countries, sub-periods, and holiday windows have effectively been tested.

Core Concepts

What does the Sell in May evidence actually show?

The core empirical claim traces to Bouman and Jacobsen's 2002 paper "The Halloween Indicator, Sell in May and Go Away," which tested the November-April versus May-October return split across 37 countries using data back to the 1970s (and longer in some markets) and found statistically significant outperformance for the November-April half in 36 of the 37 countries studied. This breadth across markets is the strongest part of the case for the pattern being more than a single-country coincidence.

But breadth across countries does not settle the question of whether the pattern persists going forward. Subsequent research covering more recent decades, including work revisiting the same authors' data with extended samples, has found the effect weaker and less consistent after the mid-2000s in a number of markets, including years where May-October outperformed November-April by a wide margin. A pattern that was robust across a 1970s-1990s sample and has weakened since is consistent with two very different explanations: the original pattern being partly a product of the specific sample studied (a form of the same overfitting risk covered in multiple testing and researcher degrees of freedom), or a genuine historical effect being arbitraged away once it became widely known and easy to trade with index funds. Neither explanation supports treating the pattern as a dependable rule for the years ahead.

What is the pre-holiday effect?

Separately from the Sell in May calendar split, a body of research has examined returns on the single trading day immediately preceding U.S. market holidays (Thanksgiving, Christmas, Independence Day, and similar). Ariel (1990) and Lakonishok and Smidt (1988) both found average pre-holiday day returns notably higher than average non-holiday trading day returns in their historical U.S. samples, a pattern sometimes attributed to reduced institutional trading activity and lighter volume ahead of a market closure, leaving retail order flow with more relative influence on the closing price.

As with Sell in May, later work covering more recent decades has found the pre-holiday effect has weakened substantially or become statistically insignificant in some markets and periods, which again is consistent with either sample-specific noise in the earlier studies or arbitrage-driven decay. The effect size itself was always modest — typically a fraction of a percent of additional average return on a single day — which means it is easily erased by a single wide bid-ask spread, a modest slippage cost, or a below-average pre-holiday session.

Why 6-month backtests can overstate real profitability

A common way "Sell in May" backtests are presented is a simple comparison: average price return during November-April versus average price return during May-October, computed on an index level with no transaction modeled. This comparison is not the same thing as the return an investor actually executing the strategy would receive, for three reasons.

First, transaction costs: a strategy that exits in May and re-enters in October or November requires two round-trip transactions per year — a sale and a purchase, twice — each incurring commissions (small for most modern brokers, but not zero) and, more importantly, the bid-ask spread, paid on both the sale and the purchase. Second, taxes: unless the position is held in a tax-advantaged account, the semi-annual holding period on the "in" side of the trade may not qualify for long-term capital gains treatment in every jurisdiction and personal tax situation, meaning realized gains could be taxed at a higher short-term rate than a buy-and-hold investor would pay. Third, cash drag: money held in cash during the May-October "out" period earns a money-market or cash-equivalent return, not zero, but also not the equity risk premium that buy-and-hold captures in years when May-October happens to be a strong period (which the historical average masks — some individual years are strongly positive in the "sell" window).

Because the historical Sell in May edge itself has been on the order of a few percentage points of average annual outperformance versus buy-and-hold — not a large multiple of it — round-trip transaction costs and tax drag compounded over decades can consume a meaningful share of a modest edge, and in some published backtests turn a positive-looking price-return comparison into a marginal or negative total-return-after-costs comparison. A realistic evaluation of the pattern requires modeling both transactions, a realistic bid-ask spread and commission schedule, and the applicable tax treatment — not just the headline price-return split.

Worked Scenario

An investor backtests Sell in May on a broad U.S. equity index over a 30-year period and finds November-April price returns averaged 5.5% versus 1.8% for May-October, a gap that looks compelling on paper.

1. Raw price-return gap: 5.5% vs 1.8% suggests switching to cash every May and back every November would meaningfully beat buy-and-hold.
2. Add transaction costs: Two round trips per year at a conservative 0.05% bid-ask spread plus commission per leg costs roughly 0.2% per year in aggregate — a modest but real drag over 30 compounding years.
3. Add tax treatment: If the May-October cash position is held in a taxable account and the November-April equity position is sold each May (a new realization event), gains may not always qualify for long-term treatment depending on exact holding periods, potentially taxing a portion of gains at short-term rates.
4. Add cash-drag reality: The 1.8% May-October figure is an average; in some individual years May-October was strongly positive, meaning the strategy sat in cash during a year it should have stayed invested — the average masks meaningful year-to-year variance, which a Monte Carlo resampling test (see Monte Carlo Resampling) would reveal as a wide range of possible outcomes, not a guaranteed edge.
5. Conclusion: After realistic costs and accounting for dispersion around the average, the net edge of a literal Sell in May switching strategy is considerably smaller than the raw 3.7 percentage point price-return gap suggests, and may not clear the bar of a strategy worth the added complexity and cost versus simply staying invested.

Measurement Framework

Common Misconceptions and Failure Modes

Treating "documented in a peer-reviewed study" as "proven to work going forward"

Bouman and Jacobsen's finding across 36 of 37 countries is a real, published, historical result — but a historical result, however broad, describes what happened in the sample studied. It does not guarantee the pattern continues, especially once a pattern is well known enough that market participants can position around it in advance, which is itself a plausible reason the effect has weakened in more recent data.

Ignoring round-trip transaction costs and taxes in the backtest

As covered above, a raw November-April versus May-October price-return comparison is not the return an investor executing the strategy actually receives. Any Sell in May backtest that does not explicitly model both transactions, the spread, and tax treatment is measuring a different (and more favorable) thing than a live strategy would deliver.

Cherry-picking the sample window

Because the effect has been inconsistent across sub-periods, it is possible to choose a start and end date for a backtest that makes the pattern look stronger or weaker than a longer, unfiltered sample would show. This is the same researcher-degrees-of-freedom risk covered in multiple testing and researcher degrees of freedom — the sample window should be fixed before looking at the result, not chosen after.

Assuming a small effect size scales to a meaningful strategy

Both Sell in May and the pre-holiday effect, even when statistically present in a given sample, are modest in magnitude — a few percentage points a year for Sell in May, a fraction of a percent for a single pre-holiday day. Building a strategy that concentrates capital or leverage around a small, debated, and shrinking historical edge carries risk out of proportion to the documented size of the effect.

Frequently Asked Questions

Sources and Further Verification

• Bouman, S. & Jacobsen, B. (2002). "The Halloween Indicator, 'Sell in May and Go Away': Another Puzzle." American Economic Review, 92(5), 1618–1635. The primary empirical study documenting the pattern across 37 countries. Available at aeaweb.org.
• Ariel, R.A. (1990). "High Stock Returns before Holidays: Existence and Evidence on Possible Causes." Journal of Finance, 45(5), 1611–1626. Foundational study on the pre-holiday effect. Available via jstor.org.
• Lakonishok, J. & Smidt, S. (1988). "Are Seasonal Anomalies Real? A Ninety-Year Perspective." Review of Financial Studies, 1(4), 403–425. Long-sample review of calendar anomalies including holiday effects. Available at academic.oup.com.
• Jacobsen, B. & Zhang, C.Y. (2018). "The Halloween Indicator: Everywhere and All the Time." Extended-sample follow-up examining persistence of the effect. Available at ssrn.com.

Educational Disclaimer

This guide is for educational purposes only and does not constitute investment, financial, or trading advice. Historical seasonal patterns discussed here are descriptive statistics from academic research, not predictions or recommendations. Past performance does not guarantee future results. Consult a qualified financial professional before making investment decisions. Trading involves significant risk of loss.