S&P 500 Seasonality and the Four-Year Presidential Cycle

Review period: 1928-01-01 to 2026-09-21 Scope: Historical S&P 500 returns, volatility, drawdowns and win rates across the four-year U.S. presidential cycle, including monthly seasonality, political-control splits, macro regimes and the position of 2026. Source policy: Primary return data, official government records and peer-reviewed research were prioritized; major institutional studies were used for volatility, drawdowns, political-control comparisons and current-market context.

Executive summary

The strongest historical election-cycle result is the relative ranking of the four years, not a precise forecast for any one year. From 1928 through 2025, calendar-year S&P 500 total returns averaged 10.6% in Year 1, 7.1% in Year 2, 17.8% in Year 3 and 11.9% in Year 4. Year 3 also had the highest median and win rate, while the midterm year had the lowest. The Year 3–Year 2 average-return spread was an economically large 10.7 percentage points. (pages.stern.nyu.edu)

Risk statistics reinforce—but do not prove—the cycle. In an LPL/Bloomberg study beginning in 1950, midterm years had the highest realized volatility, 16.7%, and the deepest average maximum drawdown, approximately −18%. The corresponding volatility figures for Years 1, 3 and 4 were 14.5%, 15.8% and 16.0%. (mcb.com)

The monthly evidence is more nuanced than the folklore. September weakness is broad-based: its average return was negative in every cycle year in a 1928–2025 price-return dataset. But October was positive overall and particularly strong in midterm years, so the idea that “September and October are always bad” is not supported by average returns. The historically stronger midterm pattern begins in October or shortly before Election Day and carries into the following pre-election year. (gurufocus.com)

The post-midterm rally is a strong raw-data regularity but a weaker trading anomaly. The S&P 500 was positive in roughly 95% of the 12-month periods following midterms since 1938, and its average six-month post-election return since 1970 was 14.1%. However, peer-reviewed research finds that much of the excess return disappears after adjustment for systematic risk. (fidelity.com)

Party and government-control effects are less reliable. Democratic administrations have historically coincided with higher returns, a finding documented in peer-reviewed research, but the observation is confounded by inherited economic conditions, monetary policy, recessions, valuations and very small party-by-cycle samples. Claims that divided government is always best are contradicted by reasonable alternative samples and measurement conventions. (onlinelibrary.wiley.com)

As of the last completed market close on September 18, 2026, the S&P 500 had returned approximately 12.7% including dividends in 2026 after suffering an earlier price drawdown of about 9.8%. That is stronger and less volatile than the average completed midterm year so far, but September and the November 3 election were still incomplete. This comparison describes where 2026 stands; it is not a forecast of the fourth quarter. (slickcharts.com)

1. Data and methodology

The central annual-return series is the S&P 500, including reinvested dividends, from the NYU Stern historical-returns dataset maintained by Aswath Damodaran. Data before the modern S&P 500’s 1957 launch represent predecessor or reconstructed large-cap U.S. equity series rather than a contemporaneously investable 500-stock index. (pages.stern.nyu.edu)

Cycle years are classified mechanically:

• Year 1: post-presidential-election year;
• Year 2: midterm-election year;
• Year 3: pre-presidential-election year;
• Year 4: presidential-election year.

The annual-return sample provides only 24 or 25 observations per cycle bucket. Volatility and drawdown comparisons generally begin in 1950, leaving roughly 19 completed observations per bucket. Subdividing further by party, congressional control, inflation or recession creates very small samples.

Results from different sources should not be assumed directly comparable. Some use price returns, others total returns; some measure calendar years, others inauguration-to-inauguration periods or election-date windows; and start dates range from 1928 to 1970. Those distinctions explain many apparently conflicting published averages.

2. Returns, volatility, drawdowns and win rates

Annual returns, medians and win rates cover 1928–2025; risk statistics marked with an asterisk use a separate LPL/Bloomberg study beginning in 1950. The two panels therefore describe the same pattern over different periods and definitions rather than one homogeneous dataset. (simianx.ai)

Interpretation

• Year 3 is the clearest return effect. It led on mean return, median return and win rate. Since 1950, all 19 completed pre-election years were positive in the cited total-return calculation, averaging 20.8%. Since 1980, the 11 completed Year 3 observations were also all positive. (simianx.ai)
• Year 2 is more clearly a risk year than an inevitably negative year. Its 58% full-sample win rate is still above 50%, but its average return, realized volatility and average drawdown were the least favorable of the four buckets. (simianx.ai)
• Election years are not unusually poor. Their historical average and median were above Year 1 and Year 2, and 84% finished positive. S&P Dow Jones Indices separately found that election-year returns and volatility were broadly similar to non-election years. (simianx.ai)
• The distributions overlap heavily. Year 3 contains the worst annual observation in the dataset—1931—while Year 2 contains the best, 1954. Knowing the cycle position therefore does not tightly constrain the possible outcome. (pages.stern.nyu.edu)

3. Outlier sensitivity

A simple one-outlier-per-bucket sensitivity check preserves the ranking. The following are Revoqra-derived calculations using the published group means, sample sizes and largest positive or negative observation in each bucket:

The check is not a formal winsorized or robust-regression estimate, but it establishes an important point: the Year 3 advantage is not created by one favorable outlier, and the midterm average is actually helped materially by 1954. (pages.stern.nyu.edu)

4. Monthly seasonality inside the four-year cycle

Average monthly price returns from 1928–2025 show the following late-year pattern:

These are arithmetic monthly averages from a price-return series and will not compound exactly to the annual total-return results above. (gurufocus.com)

September–October: evidence versus folklore

September weakness is real as a descriptive historical pattern. It was the only month with a materially negative long-run average in the cited 1928–2025 data and was negative in all four cycle positions. Midterm Septembers averaged about −1.5%, while a separate Citadel Securities study estimated an average intra-month selloff of approximately 6.2% before the typical recovery. (gurufocus.com)

October weakness is mostly folklore if “weakness” means average return. October’s all-year average was positive, and midterm October was one of the strongest month-by-cycle cells at +2.30%. October’s reputation is more plausibly connected to unusually memorable crashes and elevated volatility than to a consistently negative mean return. (gurufocus.com)

The midterm pattern is therefore better summarized as late-summer/September weakness followed by a possible October–November transition, not as two uniformly bad months. Nasdaq Dorsey Wright found that only 44% of midterm first quarters were positive, versus 83% of fourth quarters. J.P. Morgan similarly found slightly negative average returns in each of the three quarters preceding the midterm quarter, followed by an average 6.6% fourth-quarter return. (dorseywright.nasdaq.com)

5. Midterm drawdowns and post-midterm rallies

The midterm year’s defining characteristic has historically been its path rather than its final return. Institutional studies place the average intra-year drawdown near 18%–20%, materially worse than in the other three cycle years. Many midterm years nevertheless recovered enough to finish positive. (mcb.com)

The rebound statistics are striking but endpoint-sensitive:

• Fidelity reports that the S&P 500 was positive in approximately 95% of the 12 months following midterm elections since 1938. (fidelity.com)
• D.A. Davidson, using a somewhat different start date, reports positive November-to-November returns after all 20 midterms since 1942. (dadavidson.com)
• BlackRock calculates an average 14.1% six-month return after midterms since 1970, with the rally historically beginning about 22 trading days before Election Day. (blackrock.com)
• Capital-market research finds that these elevated raw returns largely disappear after controlling for systematic risk, suggesting compensation for risk or changing macro conditions rather than an exploitable election-day premium. (sciencedirect.com)

Assessment: The late-Year-2/Year-3 rebound is robust as a raw historical regularity. It is less robust as evidence of abnormal risk-adjusted return, and it cannot be cleanly separated from recession recovery, monetary-policy shifts, valuation changes and ordinary mean reversion.

6. Democratic versus Republican administrations

A long-standing academic result is the “presidential puzzle.” Santa-Clara and Valkanov found value-weighted excess market returns approximately nine percentage points higher under Democratic presidents in their original sample, with lower—not higher—market risk under Democrats. Pástor and Veronesi later proposed an equilibrium explanation based on time-varying risk aversion and the tendency for more risk-averse electorates to elect Democrats. (onlinelibrary.wiley.com)

Descriptive 1928–2025 price-return data also show higher average returns under Democrats, 10.70% versus 5.26% under Republicans. But the difference varies substantially by cycle year:

The near-identical Year 2 averages suggest that midterm weakness has not been primarily a party-specific effect. However, each party-by-cycle cell contains only about a dozen observations, and large events—including the Great Depression, oil shocks, the technology bust, the global financial crisis and the pandemic—are unevenly distributed across administrations. (gurufocus.com)

Assessment: Democratic outperformance is a genuine historical observation with academic support. It is not strong evidence that presidential party alone caused the difference or can predict returns under a newly elected administration.

7. Unified versus divided government

Results depend heavily on the timing and classification method. J.P. Morgan’s inauguration-date total-return study for 1937–2025 reports:

(am.jpmorgan.com)

Other credible studies produce different rankings. First Trust’s calendar-year 1949–2025 series finds strong returns with split Congresses under presidents of either party, while RBC reports that a literally split Congress—one party controlling each chamber—occurred only six times from 1932 through 2020. Those six periods averaged 13.4% annualized, but several coincided with unusually strong secular bull-market years. (ftportfolios.com)

The samples expose the problem with the “gridlock is best” claim:

1. “Divided” may mean the opposition controls both chambers, each party controls one chamber, or simply that no party holds a trifecta.
2. Returns may be measured by calendar year, congressional term or presidential inauguration.
3. Some configurations have fewer than ten independent occurrences.
4. Government control is correlated with the business cycle, fiscal stance and election results rather than randomly assigned.

Assessment: No government configuration has a sufficiently stable advantage to serve as a stand-alone market-timing rule. The historical evidence supports market resilience under many configurations more strongly than it supports gridlock specifically.

8. Macroeconomic regimes

Recession versus expansion

Post-1950 recessions have been disproportionately concentrated in Years 1 and 2. The limited number of Year 3 recession episodes—principally 1975, 1991 and the beginning of the 2007 recession—helps explain why Year 3 returns look unusually strong. In other words, part of the presidential cycle may be an uneven business-cycle distribution rather than a direct electoral-policy effect. (hussmanfunds.com)

This creates a causal-identification problem: if markets fall during a Year 2 recession and rebound when the recession ends in Year 3, the calendar will record a presidential-cycle effect even when the dominant mechanism is ordinary recession and recovery.

Inflationary regimes

RBC identified seven earlier midterm periods accompanied by significant inflation challenges and found that the market reached a new high in the following year less frequently than after noninflationary midterms. The sample is too small for a firm statistical conclusion, but it weakens any claim that the post-midterm rally is unconditional. (rbcwealthmanagement.com)

Monetary policy, valuations and earnings

Institutional comparisons consistently emphasize that interest rates, monetary policy, economic growth, corporate earnings and valuation have more explanatory and forecasting relevance than party control. These variables also help explain why superficially similar election years can produce radically different returns. (am.jpmorgan.com)

9. Statistical and economic significance

A limited significance check

The Year 3 win rate is 22 positive years out of 24. Using the full-sample 73% positive-year rate as a fixed benchmark, a Revoqra-derived one-sided exact binomial calculation gives an unadjusted p-value of approximately 0.025. That sounds significant, but it is not a clean hypothesis test: the benchmark is estimated from overlapping data, returns are not independent Bernoulli trials, and the pattern was selected from many possible calendar partitions. (simianx.ai)

There are 48 basic month-by-cycle cells before adding party, congressional control, term number or macro regimes. At a 5% threshold, roughly 2.4 false positives would be expected under an illustrative independent-test assumption; a simple Bonferroni threshold would be near 0.001. The tests are correlated, so this is not a formal correction, but it shows why isolated monthly “edges” deserve skepticism. Academic work on calendar effects demonstrates that repeated searches of heavily studied market data can materially inflate conventional significance levels and make single crash observations disproportionately influential. (sciencedirect.com)

10. Where 2026 stands

The 2026 federal midterm election is scheduled for November 3, 2026. As of September 2026, Donald Trump is serving the second year of his second administration, and Republicans retain unified control: the Senate has 53 Republicans, 45 Democrats and two independents who caucus with Democrats; the House’s latest official breakdown showed 218 Republicans, 214 Democrats and two vacancies. (fec.gov)

At the September 18 close, the S&P 500’s 2026 return was approximately 11.8% on price and 12.7% including dividends. Its principal completed drawdown was approximately 9.8%, from the January 28 high to the March 30 low. (slickcharts.com)

Relative to history, this means:

• 2026’s return through September 18 was already above the 1928–2025 completed-midterm average of 7.1%.
• Its roughly 9.8% drawdown was substantially shallower than the approximately 18% historical midterm average.
• The year had not followed the classic template of a persistently weak midterm market through early autumn.
• September remained incomplete, and neither the election nor the historically stronger fourth-quarter window had occurred.

Those observations do not imply that 2026 must revert toward the midterm average, nor that the historical post-midterm rally must appear. Full-year historical averages are not conditional forecasts once most of the current year’s information—including earnings, interest rates, valuation and prior returns—is already known.

11. Bottom line: evidence versus folklore

Historically robust enough to matter as context

1. Year 3 has delivered the strongest combination of average return, median return and win rate.
2. Year 2 has produced the weakest average return and the most difficult drawdown/volatility profile.
3. September weakness is broader and more persistent than October weakness.
4. The period from late in the midterm year into the pre-election year has historically been unusually strong in raw returns.

Not robust enough to use as a stand-alone forecast

1. A negative or bear-market midterm year is not inevitable.
2. October is not, on average, a consistently losing month.
3. The election result or president’s party does not reliably determine broad-index returns.
4. Divided government is not uniformly better than unified government.
5. A historically perfect or near-perfect post-midterm record does not establish a risk-free anomaly.

The election cycle is best treated as a weak prior about the distribution and timing of risk, not as a directional trading system. It can justify heightened awareness of late-summer and midterm volatility, but earnings, monetary policy, recession risk and valuation should dominate portfolio decisions.

Sources

• NYU Stern/Damodaran annual U.S. asset-return dataset, including S&P 500 dividends. (pages.stern.nyu.edu)
• S&P Dow Jones Indices analysis of election versus non-election years. (spglobal.com)
• LPL Research/Bloomberg presidential-cycle returns, volatility and drawdowns. (mcb.com)
• Santa-Clara and Valkanov, The Presidential Puzzle, Journal of Finance. (onlinelibrary.wiley.com)
• Pástor and Veronesi, Political Cycles and Stock Returns, Journal of Political Economy. (journals.uchicago.edu)
• J.P. Morgan Asset Management 2026 Midterm Election Hub. (am.jpmorgan.com)
• BlackRock midterm-election analysis. (blackrock.com)
• Fidelity post-midterm return analysis. (fidelity.com)
• RBC Wealth Management election-cycle and government-control analysis. (rbcwealthmanagement.com)
• Official FEC, House and Senate records for the 2026 election date and current congressional composition. (fec.gov)