Sell in May and go away?

Introduction

Most investors have likely heard the market adage “Sell in May and go away.” But does it actually make sense as a strategy? Numerous articles have examined its validity—primarily focusing on American stock indices. Today, we'd like to explore whether this rule can be profitably applied to one of Europe’s most important indices: the DAX.

The Origin

The original phrase “Sell in May and go away, come back on St. Leger’s Day” dates back to 18th-century London. Wealthy British investors, aristocrats and bankers left the city during the summer heat to relax in the countryside, only returning after the St. Leger Stakes horse race in September. U.S. investors later adapted this idea, suggesting one should stay away from stocks from Memorial Day in May through Labor Day in September. [1]

What is it all About?

Countless studies have tested this seasonal strategy using historical price data. It appears, for example, that the S&P underperforms in summer months compared to winter. But does that justify completely liquidating equities during that period? Likely not—the summer months may be weaker on average, but they still often yield positive returns. [2]

How Does It Look in Europe?

Since most analyses focus on U.S. indices, we’ll examine how the rule holds up for Europe’s DAX. We use daily closing prices from 1987 to 2025, not including dividends or interest. We want to see whether following “Sell in May” outperformed a simple buy and hold strategy.

The chart below compares the percentage returns of the summer months (May–August) versus the winter months (September–April), for each calendar year.

A buy and hold investment in the DAX from January 2, 1987, to December 30, 2024, yielded roughly +1309 %. Applying the seasonal “Sell in May” rule—holding only January–April and September–December each year—would have delivered only +994 %, thus underperforming buy and hold.

This alone does not fundamentally indicate whether the seasonal strategy is superior or inferior. That’s why we want to test it across various investment horizons using historical data. We begin with time frames of three, five, and ten years. For each, the buy-and-hold performance is calculated over the selected period—from January of the starting year through December of the final year. This is then compared with the performance of the seasonal strategy - that is, the returns from January to April and from September to December of each year within the respective time window.

In the case of the 3-year investment horizon, we slide the window one year at a time across the entire data set. For example: the first comparison covers January 1987 to December 1989, the next from January 1988 to December 1990, and so on. The grey bars show the return of the buy-and-hold strategy in each 3-year period, while the violet bars represent the seasonal strategy. It becomes clear that neither of the two strategies consistently outperforms the other. In 16 out of 36 cases, the seasonal strategy achieved a higher return. In the remaining 20 periods, buy-and-hold was ahead. On average, the buy-and-hold strategy returned +30.5 %, while the seasonal strategy achieved +27 %.

The 5-year investment horizon produces a similar picture. Here too, we shift the window by one year each time and compare the results. In 15 out of 34 periods, the seasonal strategy delivered a better performance. In the remaining 19 periods, the buy-and-hold strategy was ahead. On average, buy-and-hold achieved a return of +54 %, while the seasonal strategy came in at +49 %.

Using a 10-year time window, the seasonal strategy is generally superior in 20 out of 29 possible periods. However, as the following chart shows, this would have meant missing out on a particularly strong outperformance of the buy-and-hold strategy during the years 1987 to 1991. On average across all tested time frames, buy-and-hold would have achieved a return of 124.9 %, while the seasonal strategy delivered slightly less at 123.9 %. For the sake of completeness, it should be noted that the ongoing year 2025 was also included in the analysis.

We have now examined investment horizons of three, five, and ten years in detail and found that the seasonal strategy would have, on average, underperformed compared to a buy-and-hold strategy during these time periods.

Of course, many other investment periods would have been possible in the past - such as four years, 20 years, or even 38 years. The following table therefore shows the results for all possible investment periods in the DAX that could be tested using the available data. Conclusion: There are almost as many periods in which the "Sell in May" strategy would have performed better on average (20 times) as there are periods in which buy-and-hold would have come out ahead (18 times).

Based on these results, it is therefore probably unwise to choose one strategy over the other for a specific investment period just because it happened to perform better in the past. Too many factors have influenced the outcomes, and any outperformance or underperformance could just as easily have been the result of chance.

So how can we use the available data to make a forecast for the future performance of the DAX and determine which strategy might be better going forward? The following section attempts to provide an answer to this question.

Forecast

We use historical data as a basis to generate variations of possible future years and then apply these data in a so-called bootstrap simulation of potential performance.

Those who want to learn more about the simulation approach can refer to the appendix at the end.

The result of this simulation is shown in the following chart: the simulation indicates an outperformance of the buy-and-hold strategy, which increases further with longer investment horizons. For comparison, the chart also displays the historical returns of both strategies.

The histograms in the following chart show the performance difference of the seasonal “Sell in May” strategy compared to the buy-and-hold strategy. The bars represent the distribution of performance outcomes—grouped along the horizontal axis—that would have occurred across all simulation runs.

The results of the simulation therefore suggest a clear underperformance of the “Sell in May” strategy. Critical readers might wonder why the simulation using synthetic data deviates quite significantly from the historical results. Possible reasons for this are also discussed in the appendix at the end of the article.

Conclusion

What can we conclude from the simulations?

The historical performance of both strategies was heavily influenced by individual, exceptional events (e.g., the 2008 financial crisis and its aftermath) and is therefore not a reliable indicator for the future.

The block bootstrap simulation (using independent annual blocks) suggests that, on average, a buy-and-hold strategy offers higher expected returns over medium and long investment horizons than the “Sell in May” rule.

Since both historical and simulated results vary widely and depend on model assumptions (such as the independence of years), one should not rely solely on the “Sell in May” rule. Instead, a broader analysis that includes risk management, diversification, and fundamental data is recommended.

If you prefer relying on solid data rather than seasonal rules of thumb, feel free to visit www.gravitrade.at. There, you can easily create and test a variety of strategies—based on technical indicators, news sentiment, or fundamental data. And the best part: once you’ve found a working strategy, you can activate it with just a few clicks—for automatically generated, data-driven trading signals.

Appendix: Simulation

The simulation of potential future returns for both strategies was carried out as follows:

Based on the DAX closing prices, monthly returns were calculated (e.g., Jan 2020: +2%, Feb 2020: −1%, …).

For each calendar year, exactly those 12 monthly returns were grouped and stored as a single one-year block. Using historical data, we thus obtain a list of 38 such blocks (since 38 complete years are available).

Next, a desired investment horizon H was specified—for example, 3, 5, or 10 years. For each simulation run, H blocks were drawn randomly with replacement from the list. (With replacement means that a block could be drawn multiple times—just like rolling a die several times.) These blocks were then joined into a synthetic sequence of years, from which the buy-and-hold performance and the seasonal performance were calculated.

This process was repeated several thousand times per investment horizon to achieve statistically meaningful results. The more simulation runs performed, the better short-term fluctuations can be smoothed out, allowing for reliable average and dispersion values. In our example, investment horizons of 3, 5, 10, 15, 25, 35, and 38 years were chosen, with 5,000 simulation runs each.

Reasons for Deviations Between Simulation and Historical Data

Overlap vs. random selection: In the historical analysis, investment horizons overlap significantly—a given year appears in many of them. In contrast, the simulation can randomly select an extremely good or bad year multiple times in succession, which can shift the average up or down.

Different means: The historical mean is the actual arithmetic mean of the 29 available values. The bootstrap expected value, however, is the mean of a theoretical distribution under the assumption that each year block is drawn independently. These two metrics do not have to match.

Disruption of long market phases: Because each year is shuffled independently, long crash or boom phases are broken up into individual blocks. In reality, however, markets often follow multi-year trend segments, which are not preserved in this simulation method.

Nevertheless, the approach and the division into one-year blocks are justified: one-year blocks are small enough to avoid capturing historical dependencies, yet large enough to retain seasonal patterns (monthly structures). Using the bootstrap method, we repeatedly and randomly draw these blocks with replacement to create a realistic picture of potential future scenarios.

References

• [1] A. Loo, „Sell in May and Go Away,“ [Online]. Available: https://corporatefinanceinstitute.com/resources/career-map/sell-side/capital-markets/sell-in-may-and-go-away/#:~:text=the%20summer%20months.-,History%20of%20Sell%20in%20May%20and%20Go%20Away,Established%20in%201776%2C%20the%20St.. [Zugriff am 10 05 2025].
• [2] Blind Luck Project, „Blind Luck Project,“ [Online]. Available: https://blindluckproject.com/blog/sell-in-may-and-go-away?utm_source=chatgpt.com. [Zugriff am 10 05 2025].