S&P's September to Remember

Summary

This research examines the historical relationship between positive September performance in the S&P 500 and subsequent January returns, using 1975–2024 data. We focus specifically on the 22 years in which September finished with a positive return and analyze what happened the following January.

Key Findings (Corrected):

• When September closes positive, January has been positive in 15 of 22 cases (68.2%).
• The average January return in those 22 years is +1.96% with a standard deviation of 4.94%.
• Statistically, the 95% confidence interval for the mean January return is roughly -0.2% to +4.2% – a modest edge, not a guaranteed “January effect reboot.”
• The win rate (68.2%) is higher than a coin flip, but the sample size is small and the effect is noisy across regimes.

Bottom line: a positive September has historically tilted the odds in favor of a positive January, but the magnitude and reliability of the edge are much weaker than a naïve read of the raw averages might suggest.

---

Methodology

Data Source and Scope

• Index: S&P 500 (price-level monthly returns)
• Time Period: January 1975 – December 2024
• Sample Size: 48 complete calendar years
• Adjustment: Monthly percentage changes from month-end close to month-end close

Classification Rules

• Positive September: September monthly return > 0%
• January Performance: Return from December 31 close to January 31 close of the following year
• Sample of Interest: All Januarys immediately following a positive September (22 observations)

All of the core statistics below are computed directly from the 22 year-by-year observations shown in Appendix A.

---

Historical Context: September’s Reputation

September has a well-earned reputation as the weakest month for U.S. equities. Across long-term datasets, it is typically the only month with negative average returns and the lowest share of positive months. Within the 1975–2024 sample used here:

• September was positive in 22 of 48 years (45.8%).
• September was negative in 26 of 48 years (54.2%).

That makes a positive September a relatively uncommon – and therefore interesting – setup to examine heading into the following January.

---

Core Analysis: January After a Positive September

Primary Statistics (Corrected)

Metric	Value	Notes
Sample Size	22 years	1976–2019 Januarys following positive Septembers
Average January Return	+1.96%	Simple average of the 22 January observations
Median January Return	+2.90%	Half the observations are above, half below
Win Rate (Jan > 0%)	68.2% (15 of 22)	Approximate 95% CI: ~49% to ~88% (binomial)
Standard Deviation	4.94%	Volatility of January returns in this subset
95% CI for Mean	-0.2% to +4.2%	t-distribution with 21 degrees of freedom

The corrected numbers tell a more conservative story than the earlier draft (which had +2.84% and wider confidence bounds). The sample still leans bullish, but the uncertainty band is wide enough that “edge” should be treated as modest, not overwhelming.

Return Distribution

Grouping the 22 January returns into simple buckets makes the skew clearer:

Return Range	Probability	Historical Count
> +6%	18.2%	4 of 22
+3% to +6%	31.8%	7 of 22
0% to +3%	18.2%	4 of 22
Negative	31.8%	7 of 22

In other words, after a positive September:

• About half of the time January returns are +3% or better.
• Roughly one-third of the time January is negative.
• A non-trivial fraction of outcomes cluster in the +4–6% range.

Percentiles (Corrected)

• 25th percentile: -1.98%
• 50th percentile (median): +2.90%
• 75th percentile: +5.04%

These percentiles reinforce the theme: outcomes are skewed slightly positive, but the downside tail (–3% to –8%) is very real and shows up often enough that you can’t hand-wave it away.

---

Extreme Outcomes

Best Januarys After Positive Septembers (From the 22-Observation Sample)

• 1976: +11.96%
• 2019: +7.87%
• 1989: +7.13%
• 1997: +6.12%
• 1980: +5.85%

Worst Januarys After Positive Septembers

• 2009: -8.43%
• 2007: -6.12%
• 2010: -3.70%
• 2018: -3.49%
• 2005: -2.53%

The earlier draft referenced some years (e.g., 1978) that do not actually appear in the filtered positive-September sample. The list above is corrected to align strictly with the 22 rows in Appendix A.

---

Risk Assessment

Volatility and Drawdowns

• Sample volatility (January after positive Sept): 4.94%.
• Largest observed single-month loss: -8.43% (January 2009).
• Frequency of losses worse than -3%: 4 of 22 cases (18.2%).

Risk is absolutely not trivial. A positive September does not immunize January from macro shocks (2009 being the prime example), and anyone sizing positions off this pattern needs to plan for a –5% to –8% month as a realistic tail event.

How Strong Is the Edge, Really?

The numbers support a modest upward tilt in both probability and average outcome:

• Win rate is meaningfully above 50%, but with wide confidence bands due to small sample size.
• Mean return is positive, but the 95% confidence interval includes slightly negative values.

From a purist statistical perspective, this is a weak seasonal tendency, not a robust anomaly you’d bet the farm on. It can inform odds, but it should not dominate risk management.

---

Practical Use Cases

How a Trader Might Use This (Without LARPing as a Statistician)

• Treat a positive September as a contextual tailwind for January – a small nudge toward being more open to upside trades, not a green light to go all-in.
• Use the distribution to set expectations: roughly one-third chance of a down January, and a non-trivial chance of a sharp drawdown even in “favorable” years.
• If you size based on this pattern, keep allocations small enough that a –8% month on the index does not blow up your risk budget.

A simple, sane approach for systematic traders would be:

• Only lean on the effect when it aligns with broader trend/macro context.
• Scale risk modestly higher after a positive September (e.g., +10–20% relative to baseline), not multiples higher.
• Always pair it with hard stops or hedge structures that cap worst-case outcomes.

---

Limitations and Caveats

Key Limitations

• Sample size: Only 22 qualifying years; a couple of outliers move the averages significantly.
• Structural change: Market microstructure, index composition, and monetary regimes have evolved substantially since the late 1970s.
• Data mining risk: Any conditional seasonality can be a coincidence, especially if discovered after the fact and optimized.
• No sector/macro layers here: Earlier drafts tried to layer sector and macro correlations on top of this pattern; those claims are removed in this revision because they cannot be cleanly supported by the small conditional sample.

In short: this pattern is interesting and can be a useful small input to a broader process, but it is not a standalone signal you should worship.

---

Conclusion

The corrected analysis confirms that when September is positive, January has historically been more likely to be positive and has, on average, delivered a modestly higher return than a random month. However, once you properly recompute the statistics and account for the small sample size, the edge is:

• Real, but weak.
• Useful as a secondary filter, not a primary engine of a strategy.
• Fully compatible with responsible position sizing and risk management only if treated as a small tilt, not a guarantee.

The main value of this study isn’t that it “proves” January will be great after a green September – it’s that it shows how to translate a seductive seasonal story into actual probabilities and risk ranges before putting real money behind it.

---

Appendix A: Year-by-Year Results (Positive September Years)

Year	September Return	January Return
2019	+1.87%	+7.87%
2018	+0.43%	-3.49%
2017	+1.93%	+1.79%
2013	+2.97%	+5.04%
2012	+2.42%	+5.04%
2010	+8.76%	-3.70%
2009	+3.57%	-8.43%
2007	+3.58%	-6.12%
2006	+2.46%	+1.31%
2005	+0.68%	-2.53%
2004	+0.94%	+1.73%
1998	+6.41%	+4.18%
1997	+5.40%	+6.12%
1996	+5.61%	+3.34%
1995	+4.22%	+2.50%
1992	+1.03%	-2.03%
1989	+0.49%	+7.13%
1988	+4.00%	+4.18%
1983	+1.11%	+3.29%
1982	+11.54%	-1.81%
1980	+2.55%	+5.85%
1976	+2.35%	+11.96%

All statistics in this revised note are computed directly from the January column above.