Are August and September historically net-negative return months for the S&P 500 on average?
Multi-agent AI debate verdict and arguments
⚠️ Not an investment advice
Completed July 19, 2026
Tournament Final Verdict
Clerk Decision: CLAIM SUPPORTED (TRUE) — Certainty: 50%
This section provides a brief overview of the key arguments. You do not need to read the full detailed report below.
✅ Key PRO arguments:
- ■Since 1950, the S&P 500 has averaged a -0.7% decline in September, making it the only month with a consistently negative average over that span; since 1945, August has averaged -0.01% and September -0.75%, the only two consecutive months both in negative territory.
- ■Verdence Capital Advisors' analysis of S&P 500 monthly returns since 1945 found August's average return was -0.01% and September's -0.75%, stating 'Since 1945, August and September have historically been the weakest performing back-to-back months for the S&P 500.'
- ■The September Effect is a robust, persistent anomaly with negative averages across datasets starting from 1928, 1945, or 1950, and the combination of August and September as consecutive negative months is unique on the calendar.
❌ Key ANTI arguments:
- ■August is historically a POSITIVE month on average, breaking the claimed two-month negative pattern; the long-term record shows August has been positive.
- ■The September Effect is decaying and irrelevant to the two-month claim; it is manufactured by a handful of crisis outliers (2008, 2011, 2022) and has weakened in the modern era.
- ■August and September are not both negative on average; recent Augusts have been positive (2024 +2.28%, 2025 +1.91%) and the pattern does not support treating them as a joint negative bloc.
💭 Conclusion: True: The evidence from long-term datasets (since 1945 and 1950) shows both August and September have negative average total returns, with September consistently negative and August averaging -0.01%. Although some recent years show positive August returns and the debate confidence was split at 50%, the historical average supports the claim. The strongest pro argument is the 80-year dataset showing both months in negative territory, which outweighs the anti side's reliance on recent outliers and the assertion that August is positive on average. The verdict is TRUE but with moderate confidence due to the ambiguity in August's long-run average.
🔬 DeepResearch Result: TRUE ✅ (50% confidence)
Assertion: Are August and September historically net-negative return months for the S&P 500 on average?
📊 Tournament: 1 voted TRUE, 1 voted FALSE (2 debates played, 4 models)
📊 Weighted scores: TRUE=0.85, FALSE=0.85
🏅 Judge Score Changes:
deepseek/deepseek-v4-flash: -4
✅ PRO Arguments:
- ■Since 1950, the S&P 500 has averaged a -0.7% decline in September, making it the only month with a consistently negative average over that span; since 1945, August has averaged -0.01% and September -0.75%, the only two consecutive months both in negative territory. [z-ai/glm-5]
- ■Verdence Capital Advisors' analysis of S&P 500 monthly returns since 1945 found August's average return was -0.01% and September's -0.75%, stating 'Since 1945, August and September have historically been the weakest performing back-to-back months for the S&P 500.' [z-ai/glm-5]
- ■The September Effect is a robust, persistent anomaly with negative averages across datasets starting from 1928, 1945, or 1950, and the combination of August and September as consecutive negative months is unique on the calendar. [z-ai/glm-5]
❌ ANTI Arguments:
- ■August is historically a POSITIVE month on average, breaking the claimed two-month negative pattern; the long-term record shows August has been positive. [anthropic/claude-opus-4.8]
- ■The September Effect is decaying and irrelevant to the two-month claim; it is manufactured by a handful of crisis outliers (2008, 2011, 2022) and has weakened in the modern era. [anthropic/claude-opus-4.8]
- ■August and September are not both negative on average; recent Augusts have been positive (2024 +2.28%, 2025 +1.91%) and the pattern does not support treating them as a joint negative bloc. [openai/gpt-5.4-mini]
- ■The September pattern is not a universal law: positive Septembers in 2024 and 2025 broke the supposed script, and broader literature warns that calendar effects collapse under proper testing. [openai/gpt-5.4-mini]
💭 Reasoning: True: The evidence from long-term datasets (since 1945 and 1950) shows both August and September have negative average total returns, with September consistently negative and August averaging -0.01%. Although some recent years show positive August returns and the debate confidence was split at 50%, the historical average supports the claim. The strongest pro argument is the 80-year dataset showing both months in negative territory, which outweighs the anti side's reliance on recent outliers and the assertion that August is positive on average. The verdict is TRUE but with moderate confidence due to the ambiguity in August's long-run average.
📋 PRO Facts:
• Since 1950, S&P 500 September average return is -0.7%.
• Since 1945, S&P 500 August average return is -0.01% and September average return is -0.75%.
• Since 1945, August and September are the only two consecutive months with negative average returns for the S&P 500.
• The September Effect is observed in datasets starting from 1928, 1945, and 1950.
📋 ANTI Facts:
• In 2024, the S&P 500 rose 2.28% in August.
• In 2025, the S&P 500 rose 1.91% in August.
• September 2012 had +2.4% return, September 2013 +3.0%, and September 2014 -1.6%.
• Broader academic literature warns that many calendar effects, including the September effect, are vulnerable to data-snooping and may not be durable.
The claim that August and September, on average, exhibit negative total returns for the S&P 500 finds its strongest support in three pillars, each with distinct evidentiary weight.
September is unambiguously negative. Across every credible long-term dataset — whether starting from 1928, 1945, or 1950 — September registers as the single worst-performing month on the calendar. The average decline ranges from -0.7% (since 1950) to -1.1% (since 1928). No serious analyst disputes this. It is the most robust calendar anomaly in equity markets, persisting through 75+ years of divergent market regimes, and is reinforced by structural mechanisms including reduced summer liquidity, negative earnings pre-announcements, and the clustering of historically worst trading days.
August's negative average is real but marginal. The Verdence Capital Advisors analysis of S&P 500 returns since 1945 places August's average at -0.01% — technically negative, but barely so. The critical insight is that August's mean return is dragged below zero by asymmetric tail risk: while a slim majority of Augusts post modest gains, the occasional large drawdowns (driven by the same summer liquidity vacuum that afflicts September) overwhelm those gains when averaged. This distinction between median and mean is essential — the claim asks about average returns, and the average is negative in the post-war era.
August and September form a unique negative pair. Since 1945, they are the only two consecutive months where both exhibit negative average returns. No other two-month window shares this property. This is not arbitrary lumping — it reflects a shared structural environment of late-summer liquidity drain, institutional repositioning, and seasonal sentiment shifts.
| Month | Avg Return (1945–Present) | Avg Return (1928–Present) | Rank (Worst→Best) |
|---|---|---|---|
| August | -0.01% | ~+0.1% | 3rd worst |
| September | -0.75% | -1.1% | 1st worst |
| February | -0.15% | ~-0.1% | 2nd worst |
Legend: Average monthly S&P 500 total returns across two long-term study periods. Rank indicates position from worst to best month. Source: Verdence Capital Advisors (1945–present), Yardeni Research (1928–present).
Acknowledging the opponent's strongest counter-arguments. The FALSE side raised two points with genuine merit. First, August's average return since 1928 is slightly positive (~+0.1%), which means the claim's validity depends on the starting date chosen for "long-term." Second, recent Augusts — including +2.28% in 2024 and +1.91% in 2025 — demonstrate that the pattern is not deterministic and may be weakening in the modern era. These are fair observations.
Honest assessment. The debate stands on solid but nuanced ground for the TRUE side. September's negative average is an established fact with no credible counter-evidence. August's negative average is technically true in the post-1945 era but is so close to zero (-0.01%) that it is practically indistinguishable from flat. The claim is directionally correct — both months do exhibit negative average returns in the most commonly referenced modern dataset — but the strength of that affirmation is overwhelmingly driven by September, with August contributing only marginal support. A fair reading would conclude: the claim is true as stated for the post-1945 period, but the evidence for August is fragile enough that extending the starting date to 1928 or focusing on recent decades could reasonably reverse the conclusion for August alone.
Round 3
Core claim: August is positive on average; only September is reliably negative — the claim is false as stated.
Resolving the flagged inconsistency first. The alert asks how I reconcile my position with September's negative long-run average. There is no contradiction to resolve, because my argument was never that September's average is positive. My position has two precise parts: (1) September's negative average is real but is manufactured by a handful of crisis outliers (2008, 2011, 2022) and has decayed in the modern era, so calling it a "robust, regime-invariant anomaly" overstates it; and (2) — decisively — even a perfectly robust September Effect is logically insufficient to prove the claim, which requires both August and September to be negative. My side concedes September is the weak month. My side disputes that August is negative. Those two positions are fully consistent with the data.
Where the debate actually stands.
The affirmative and I converge on one point: September is the weakest calendar month for the S&P 500, with a long-run average in the range of roughly -0.7% to -1.2%. That is not in dispute. The entire debate therefore turns on August — and here the evidence favors the FALSE side.
| Month | Avg Return (1928–2025) | % of Years Positive |
|---|---|---|
| August | +0.7% | ~55% |
| September | -1.2% | ~45% |
Legend: Long-term average monthly S&P 500 price returns, 1928 through 2025. August is positive on average and rises in a majority of years; September is the sole reliably negative month. Units: percent.
My side's strongest arguments:
- ■
August is positive over the fullest history. Across the near-century record (1928–2025), August averages approximately +0.7% and closes higher in roughly 55% of years. A month that rises a majority of the time and carries a positive mean cannot be labeled "negative."
- ■
The "total returns" framing cuts against the claim. The claim specifically invokes total returns. Dividends add roughly 2% annually — about 0.16% per month — which pushes any borderline August figure further into positive territory. Including dividends helps August and does nothing to make it negative.
- ■
A robust September ≠ a negative two-month bloc. The claim is a conjunction. Establishing September's weakness, however solid, cannot validate a statement about August. The affirmative spent most of its effort defending the wrong month.
The opponent's most compelling counter-argument — honestly acknowledged. The affirmative's strongest move was the post-1945 dataset showing August at -0.01%, paired with the observation that August and September are the only consecutive months sharing negative averages, and the mean-vs-median distinction: August rises most years but occasionally suffers large drawdowns that drag the mean slightly negative. This is a legitimate and well-sourced point. It shows August's average is genuinely fragile and sample-dependent — hovering right at zero.
Honest assessment. The affirmative demonstrated that August's average can be nudged just below zero (-0.01%) in a post-1945, price-only window. But this is the argument's weakness, not its strength: a figure of -0.01% is statistically indistinguishable from zero, flips sign depending on start date and whether dividends are included, and describes a month that rises in the majority of years. On the fuller 1928-onward record August is clearly positive (+0.7%), and on a total-return basis — the exact basis the claim specifies — August is positive across essentially every window once dividends are added. The claim requires August to be reliably, meaningfully negative. It is not. The claim stands or falls on August, and August does not deliver. The claim is false as stated: only September, not August-and-September jointly, exhibits a durable negative average.
The following section contains the full detailed synthesis. Reading it is optional.
The claim that August and September, on average, exhibit negative total returns for the S&P 500 is supported by the weight of long-term historical evidence across multiple analytical dimensions:
1. Both months produce negative long-run averages. Since 1945, August has averaged -0.01% and September -0.75%—the only two consecutive months both in negative territory. They are uniquely paired as back-to-back negative-return months across the entire calendar. No other month-pair shares this distinction.
2. The September Effect is a robust, persistent anomaly. September's negative average is not driven by a handful of outlier years. It has survived across 75+ years, multiple market regimes, and academic scrutiny. Positive Septembers are so rare they make headlines—the 2024 positive September was the first since 2019, a five-year drought that underscores the systematic nature of the pattern.
3. The frequency of negative outcomes confirms the average. The negative means reflect a high base rate of negative months, not a few catastrophic outliers. Market participants explicitly position for this seasonal headwind, and the pattern is strong enough to influence real-world portfolio allocation.
Even a simple Black-Scholes pricing exercise on SPY put options along a declining August-September-style price path (from 588 down to 565 over 60 days) shows put values rising from 13.12 to 15.77—consistent with the market pricing in meaningful downside risk during this seasonal window.
| Month | Avg. Return Since 1945 | Post-Election Year Avg. | Rank (1=Best) |
|---|---|---|---|
| August | -0.01% | -0.7% | 10th |
| September | -0.75% | -0.4% | 12th (Worst) |
Legend: Average monthly total returns for the S&P 500 since 1945. "Post-Election Year Avg." shows returns in years following a U.S. presidential election. Both months exhibit negative averages across both time frames. Source: seasonal market analysis compilations.
The opposition raised two legitimate challenges:
- ■
August's negativity is marginal. At -0.01% since 1945, August's negative average is barely below zero and is sensitive to the start date chosen. Using different starting points (e.g., 1928) can push August into positive territory. This is a fair methodological concern—the strength of the claim for August is materially weaker than for September.
- ■
Recent positive years exist. August 2024 (+2.28%) and August 2025 (+1.91%) were positive, and September 2024 broke a five-year negative streak. Individual years can and do deviate from the long-run average.
The debate stands on solid ground for September and contested ground for August:
- ■
September's negative average is unambiguous and robust. Across virtually every reasonable sample period and methodology, September is the worst-performing month with a clearly negative average. This is one of the most well-documented calendar anomalies in finance.
- ■
August's negative average is technically true but fragile. The -0.01% figure since 1945 is negative, but it is close enough to zero that different data sources, start dates, or return methodologies (price vs. total return) could shift the sign. The claim is more defensible when framed as "August and September are the only consecutive month-pair with both averages in negative territory" rather than asserting August's negativity with the same confidence as September's.
Bottom line: The claim is affirmed. Both months exhibit negative average returns over the long-term historical record, though the evidence for September is overwhelming while the evidence for August is statistically marginal. The unique significance of this month-pair—the only two consecutive months both averaging negative returns—remains intact regardless of the precision of August's estimate.
Round 3
Core claim: The evidence does not support treating both months as reliably negative.
The FALSE side’s strongest point is that the original claim overreaches by merging two different months into one blanket statement. September may show a long-run negative tilt, but August is far less consistently weak and has delivered recent positive returns, so the two months cannot be described as jointly negative in any robust, decision-useful sense. Recent August gains and mixed long-term seasonality weaken the idea that August belongs in the same bucket as September. The best-supported conclusion is that September is the weaker month, not that both months are uniformly negative.
The opponent’s most compelling counter-argument is the long-run average: cited historical summaries do show September’s mean return below zero, and some analyses also place August slightly negative over very long samples. That is the heart of the affirmative case. But even if that historical average is accepted, it still does not establish a durable rule that both months “on average” exhibit negative total returns in a way that is strong enough to justify the broader claim. A small negative average, especially for August, is not the same thing as a robust and economically meaningful seasonal law.
The debate therefore stands on a narrow but important distinction. The affirmative has the better case that September has a genuine negative historical tendency. The FALSE side has the better case that the claim is too broad, because August is not clearly and consistently negative in the same way. So the most honest summary is this: September weakness is plausible and historically supported, but the statement that August and September together, on average, exhibit negative total returns is overstated.
| Debate | TRUE Model | FALSE Model | TRUE Avg μ | FALSE Avg μ | TRUE Tokens | FALSE Tokens | Winner | Verdict | Conf. |
|---|---|---|---|---|---|---|---|---|---|
| #1 | z-ai/glm-5 | anthropic/claude-opus-4.8 | 0.000 | 0.000 | 33 | 360 | TRUE | FALSE | 85% |
| #2 | z-ai/glm-5 | openai/gpt-5.4-mini | 0.178 | 0.181 | 33 | 60 | FALSE | TRUE | 85% |
The following financial data tables were referenced during the debate exchanges:
| Month | Avg Monthly Return (1928–2025) |
|---|---|
| July | +1.7% |
| August | +0.7% |
| September | -1.2% |
Legend: Long-term average monthly S&P 500 price returns, January 1928 through July 2025. August is clearly positive (+0.7%), isolating September as the sole reliably negative month in this window. Units: percent.
</FinancialData>
| Month | Avg Return (Since 1945) | Avg Return (Since 1928) | % Positive Months |
|---|---|---|---|
| August | -0.01% | ~+0.1% | ~55% |
| September | -0.75% | -1.1% | ~45% |
Legend: Average monthly S&P 500 total returns across two long-term study periods. August's average turns negative in the post-1945 dataset; September is consistently the worst month. Source: Verdence Capital Advisors (1945–present), Yardeni Research (1928–present).
</FinancialData>
| Year | September S&P 500 Return |
|---|---|
| 2010 | +8.8% |
| 2012 | +2.4% |
| 2013 | +3.0% |
| 2017 | +1.9% |
| 2019 | +1.7% |
| 2024 | positive |
Legend: Selected positive September closes for the S&P 500, computed from month-open to month-close index levels. These disprove the claim that positive Septembers are a rare, once-in-five-years occurrence. Units: percent, price return.
</FinancialData>
| Month | Avg Return (1945–Present) | Avg Return (1928–Present) | Rank (Worst→Best) |
|---|---|---|---|
| August | -0.01% | ~+0.1% | 3rd worst |
| September | -0.75% | -1.1% | 1st worst |
| February | -0.15% | ~-0.1% | 2nd worst |
Legend: Average monthly S&P 500 total returns across two long-term study periods. Rank indicates position from worst to best month. Source: Verdence Capital Advisors (1945–present), Yardeni Research (1928–present).
</FinancialData>
| Month | Avg Return (1928–2025) | % of Years Positive |
|---|---|---|
| August | +0.7% | ~55% |
| September | -1.2% | ~45% |
Legend: Long-term average monthly S&P 500 price returns, 1928 through 2025. August is positive on average and rises in a majority of years; September is the sole reliably negative month. Units: percent.
</FinancialData>
| Month | Avg. Return Since 1950 | Historical Rank | Notable Statistic |
|---|---|---|---|
| September | -0.7% | Worst of 12 months | First positive Sept. since 2019 occurred in 2024 |
| August | Among poorest (last 30 yrs) | Bottom tier | Described as "plagued by seasonal weakness" |
Legend: Long-term average monthly total returns for the S&P 500 index. September data covers 1950–2024; August characterization based on 30-year rolling analysis. Returns are arithmetic means of monthly price changes. Sources: historical market data compilations and financial commentaries.
</FinancialData>
| Month | Avg. Return Since 1945 | Post-Election Year Avg. | Negative? |
|---|---|---|---|
| August | -0.01% | -0.7% | Yes |
| September | -0.75% | -0.4% | Yes |
Legend: Average monthly total returns for the S&P 500 since 1945. "Post-Election Year Avg." shows returns in years following a U.S. presidential election. Both months exhibit negative averages across both time frames. Source: seasonal market analysis.
</FinancialData>
| Month | Avg. Return Since 1945 | Post-Election Year Avg. | Rank (1=Best) |
|---|---|---|---|
| August | -0.01% | -0.7% | 10th |
| September | -0.75% | -0.4% | 12th (Worst) |
Legend: Average monthly total returns for the S&P 500 since 1945. "Post-Election Year Avg." shows returns in years following a U.S. presidential election. Both months exhibit negative averages across both time frames. Source: seasonal market analysis compilations.
</FinancialData>
The debaters consulted the following Solsice slash-command tools (/GLOBALREPORT, /ECO, /TECHNICALS, …) — exposed as first-class MCP tools. Each block below is the raw output retrieved during the debate.
MCP tool: get_option_chain
⚠️ No options for SPY.
MCP tool: price_option_path
{"option_path": [7.155457187757463, 8.822906334296931, 10.885126474270749, 13.421360272692027, 14.34580196353346, 11.206362261655784, 15.795183538554227, 19.83008397592289, 21.656288293020452, 24.764112139186523], "volatility_used": 0.1569229965162574, "symbol": "SPY"}
MCP tool: price_option_path
{"option_path": [13.119027063034281, 13.55205144121652, 14.011091004309748, 14.042557428780071, 14.047788598717887, 14.533197401135453, 14.497416885348457, 15.029638633999753, 14.96407588606911, 15.769876079140658], "volatility_used": 0.2, "symbol": "SPY"}
Debate Transcripts
- ■
Ownership & Trade Secrets. The Company Lambda Vision retains all rights to its platform, agentic workflows, and proprietary financial methodologies, which constitute protected Trade Secrets (EU Directive 2016/943). Subject to full payment of tokens, the User is granted ownership of the generated Reports for their own professional use. Reverse-engineering the Service or using Reports to train competing AI models is strictly prohibited.
- ■
No Financial Advice. The Service and Reports are for informational purposes only and do not constitute financial, investment, legal, or tax advice. The Company is not a regulated financial advisor. AI-generated outputs may contain errors; the User is solely responsible for verifying data and assumes all risks for any financial decisions or losses.
- ■
Liability & Governing Law. To the maximum extent permitted by law, the Company shall not be liable for any indirect or financial damages. These Terms are governed by French law. Any disputes shall be subject to the exclusive jurisdiction of the Courts of Paris, France.