In short
- The two-month window around the vote has usually been positive. From one month before the election to one month after, the S&P 500 was higher in 15 of 19 cases (79%), averaging +5.1% (median +5.4%).
- Most of the move sits before the vote, not after it. The month before averages +3.8% and was positive in 13 of 19 cases (68%); the month after averages a smaller +1.3%, positive in 14 of 19 (74%). The win rate rises through the window even as the average return per leg falls.
- The worst windows line up with shocks unrelated to the vote. 2018 (−8.7% total) coincided with the trade war and a hawkish Fed; 1978 (−7.2%) with the inflation and dollar crisis of the late 1970s.
- The sample is small and the spread is wide. Nineteen elections, and the standard deviation of each leg (4.5 to 7.1 points) is close to or larger than its average — single elections vary a lot around these numbers.
- Four elections used the next trading session. NYSE was closed on election day itself in 1954, 1958, 1962 and 1966; those four use the first session afterward as the “vote” price.
Change around the vote, by window
Average and win rate across all 19 elections, for the month before the vote, the month after, and the full two-month window.
All 19 elections, bar by bar
Percentage change in the month before the vote and the month after, for each election. Dashed lines mark the two averages.
The full record
Closing price of the S&P 500 (^GSPC) roughly 21 trading sessions before the vote, on the vote date, and roughly 21 sessions after.
| Year | Election date | Price −1m | Price, vote | Price +1m | Before | After | Total |
|---|---|---|---|---|---|---|---|
| 1950 | 1950-11-07 | 20.12 | 19.56 | 19.40 | −2.78% | −0.82% | −3.58% |
| 1954 | 1954-11-02 (next session) | 32.47 | 32.44 | 34.49 | −0.09% | +6.32% | +6.22% |
| 1958 | 1958-11-04 (next session) | 51.07 | 52.03 | 52.46 | +1.88% | +0.83% | +2.72% |
| 1962 | 1962-11-06 (next session) | 57.07 | 58.71 | 63.06 | +2.87% | +7.41% | +10.50% |
| 1966 | 1966-11-08 (next session) | 74.53 | 81.38 | 82.14 | +9.19% | +0.93% | +10.21% |
| 1970 | 1970-11-03 | 86.47 | 84.22 | 88.90 | −2.60% | +5.56% | +2.81% |
| 1974 | 1974-11-05 | 64.95 | 75.11 | 66.13 | +15.64% | −11.96% | +1.82% |
| 1978 | 1978-11-07 | 104.59 | 93.85 | 97.08 | −10.27% | +3.44% | −7.18% |
| 1982 | 1982-11-02 | 121.51 | 137.49 | 138.82 | +13.15% | +0.97% | +14.25% |
| 1986 | 1986-11-04 | 234.78 | 246.20 | 253.04 | +4.86% | +2.78% | +7.78% |
| 1990 | 1990-11-06 | 313.48 | 311.62 | 329.07 | −0.59% | +5.60% | +4.97% |
| 1994 | 1994-11-08 | 459.04 | 465.65 | 445.45 | +1.44% | −4.34% | −2.96% |
| 1998 | 1998-11-03 | 988.56 | 1,110.84 | 1,150.14 | +12.37% | +3.54% | +16.34% |
| 2002 | 2002-11-05 | 785.28 | 915.39 | 906.55 | +16.57% | −0.97% | +15.44% |
| 2006 | 2006-11-07 | 1,350.66 | 1,382.84 | 1,407.29 | +2.38% | +1.77% | +4.19% |
| 2010 | 2010-11-02 | 1,137.03 | 1,193.57 | 1,221.53 | +4.97% | +2.34% | +7.43% |
| 2014 | 2014-11-04 | 1,964.82 | 2,012.10 | 2,071.92 | +2.41% | +2.97% | +5.45% |
| 2018 | 2018-11-06 | 2,884.43 | 2,755.45 | 2,633.08 | −4.47% | −4.44% | −8.71% |
| 2022 | 2022-11-08 | 3,612.39 | 3,828.11 | 3,963.51 | +5.97% | +3.54% | +9.72% |
“Next session” marks the four elections (1954, 1958, 1962, 1966) where the NYSE was closed on election day; the first trading session afterward is used as the vote price, and the before/after windows are counted from that session.
What this says
Put together, these numbers read as a market that prices in the resolution of election uncertainty ahead of the vote more than it reacts to the outcome afterward: the pre-election average (+3.8%) runs about three times the post-election average (+1.3%), while the win rate keeps climbing across the three windows — 68% for the month before, 74% for the month after, 79% for the full two months. That is more consistent with markets settling down once the vote is behind them than with an average return that swings on which party wins which chamber.
It does not explain the outliers. 1974, 1978 and 2018 each combine a weak or negative window with visible non-electoral stress — the 1973–74 oil shock and Watergate, the 1978 inflation and dollar crisis, the 2018 trade war and a Fed still hiking into December. The calendar alone does not carry those declines.
None of this is a forecast for the midterms on 3 November 2026. It is a historical base rate built from 19 observations, with confidence bands wide enough that any single election — this one included — can and has landed far from the average in either direction.
How this analysis was done with AI
The analysis was carried out by Claude, an AI model from Anthropic (Sonnet 5), used inside Claude Code on the author's computer. The AI fetched the data, ran the calculations and wrote the page. The question, the direction and the decision to publish were the author's.
Step by step
- 1. Election datesThe 19 US midterm election dates from 1950 to 2022 (the first Tuesday after the first Monday in November, in every even non-presidential year) are public record; Claude listed them directly.
- 2. Price dataFor each election, Claude fetched daily closing prices of the S&P 500 index (^GSPC) through a market-data tool connected to the session, requesting a window of ±50 calendar days around the vote — wide enough to contain about 21 trading sessions on each side.
- 3. DelegationThe 19 data pulls and the per-election calculations were run by a subagent, to keep the raw daily prices out of the main conversation; it returned only the computed prices and percentages for each election.
- 4. AggregationMean, median, sample standard deviation and win rate were computed across the 19 elections for each of the three windows (before, after, total).
- 5. Chart and pageThe bar chart and this page were generated by a Python script from the computed numbers, then reviewed and written up.
What the code computed
Every price, percentage change and aggregate statistic on this page comes from the fetched daily closes, computed in code rather than typed by hand. The same calculation can be repeated with the same inputs.
What is the AI's judgment
The choice to define “one month” as ~21 trading sessions, how to handle the four elections with a closed market, the framing of pre- vs post-election moves as anticipation vs reaction, and the commentary in “What this says.”
What was not verified, and known limits
- Small sample. Nineteen elections; a single one (2018, 1998) moves the averages noticeably.
- No significance testing. This is a descriptive summary of the historical record, not a hypothesis test; no p-values or confidence intervals were computed.
- Price index, not total return. Dividends are not included, which understates the long-run return slightly, more so in the earlier decades.
- Approximate trading-day counts. “21 sessions” is a fixed approximation of one calendar month; the exact number of sessions in a month varies with holidays.
- Not advice. Research material, not a recommendation or a forecast for the 2026 midterms or any other event.
Sources
- Prices: S&P 500 index (^GSPC) daily closes, fetched via a Financial Modeling Prep market-data tool connected to the research session.
- Midterm election dates: US general elections held on the first Tuesday after the first Monday in November of each even, non-presidential year — public record.
Research material only — not personalized financial advice. Past performance does not predict future results. Contact: domderrico@gmail.com.