Přehled trhů
Why Not to Rely on Seasonal Patterns — The Myth of the September Effect
Key takeaways
- The September effect (historically a weak month for equities) exists in the historical data, but it is not reliable for predicting the future.
- Seasonal patterns are a product of data mining — the more data you search through, the more false patterns you find.
- Every well-documented pattern loses strength once it becomes publicly known — arbitrageurs exploit it and erase it.
- Investment decisions built around a seasonal calendar lead to excessive trading and tax costs.
The September effect is one of the most cited stock market anomalies — but it predicts the future about as reliably as astrology. Here is why you should not trust it.
What do historical data show?
Historical data do indeed show that September is on average the weakest month for US equities over the past 70 years — the average return of the S&P 500 in September is slightly negative. But "average" hides enormous dispersion. In individual years September ended up +9%, −11%, +5% — in short, anything is possible. An average negative result does not mean that next September will be negative.
Why are seasonal patterns a statistical illusion?
Data mining (searching through a large volume of historical data) almost always finds some pattern. If you test 12 months, 5 days of the week, and decades of data, you will statistically inevitably find combinations that "work" — but only in the past. This phenomenon is called overfitting. The September effect was identified in exactly this way.
- After it was published and popularised, the pattern weakened significantly.
- Outside the US, the September effect is even less consistent — different markets produce different "weakest" months.
- Arbitrageurs who exploit the pattern systematically eliminate it.
How do seasonal strategies actually hurt?
An investor who sells in September and buys in October pays transaction fees, the bid-ask spread, and in many jurisdictions potentially triggers a taxable disposal — unless an applicable holding period exemption applies. Meanwhile the forecast is unreliable. The result is negative when you add up fees, taxes, and foregone returns from days out of the market. Read more about taxes in the article taxes on ETFs.
What to do instead of seasonal patterns?
Do not buy or sell based on the month of the year. Invest regularly — a DCA strategy automatically buys regardless of the calendar and averages the purchase price over time. Markets are unpredictable in the short run — but the long-term trend has historically been upward.
FAQ
Is the September effect real?
In historical averages, yes, but with enormous dispersion. In individual years September behaves completely differently from the average. It cannot be reliably used for investment decisions — especially after accounting for trading costs and taxes.
Are there seasonal patterns that actually work?
Some anomalies have more consistent data (for example the Turn-of-the-Month effect), but after accounting for transaction costs and once widely known, most disappear. No seasonal pattern provides a robust foundation for an investment strategy.
What is data mining bias?
Data mining bias occurs when you search through a large dataset looking for patterns — you will always find some purely by chance. The pattern is then only valid for the historical data it was "trained" on, but has no predictive power going forward.