Elections and the stock market

Key findings

1. Since about 1950, year 3 of a term has been the strongest.It averaged +17.2% and fell only in 2011 (flat) and 2015 (−0.7%). The pattern has held since the Stock Trader's Almanac popularised it from 1968 (+16.9% a year). Before 1950 it did not show up: over 1873–1949 year 3 averaged +4.0%, no better than the other years (Welch t = 0.13).
2. The six months after a midterm have been unusually good.Since 1930 the S&P 500 was higher six months after midterm election day in 21 of 24 cases, averaging +10.7% against +4.0% from an ordinary day. Over twelve months the edge is smaller and within noise (+12.6% against +8.2%). This is largely the same stretch of time as the midterm year's end and year 3's start.
3. The "midterm low" statistic is mostly hindsight.From each midterm year's lowest close, the next year averaged +28.3%. From the low of every other year: +23.8%. The gap is within noise.
4. The president's party explains little.Years under Democratic presidents averaged +7.9%, under Republicans +5.4% (1873–2025). The difference is within noise (Welch t = 0.86). Dropping each party's three worst years leaves a gap of 2.7 points, about the same.
5. Presidential elections show no reliable pattern.The first close after a presidential election averaged −0.6%, within noise. Twelve months on, presidential elections did no better than an ordinary day.

Key numbers

The figures behind the findings, side by side. Shaded rows have the largest t statistics; they overlap in time (the six months after a midterm are the first half of year 3), so they are not two independent findings.

N is the number of years or elections. Mean is the average price return; Up is the share that were positive. t compares the measure with the line in "Compared with" (Welch t, or for the midterm windows the difference from any trading day over its standard error); about 2 or more is conventionally called significant, and with this many comparisons some will clear that bar by chance. The tape row gives the share of elections the rule called right and a one-sided binomial p for one window; other windows give 15 to 20 of 25.

ResultsEvery election and what followed

Who won has not reliably moved stocks. No split by result below is statistically distinguishable from zero, though the groups are small (8 to 16 elections). The calendar of the term shows more, but mainly in the six months after a midterm (Section 2).

Mean price return of the S&P 500; Up is the share of elections followed by a gain over twelve months. Daily windows start from the last close on or before election day (from 1928). The monthly measure runs from October's average price, the month before the vote, to the average a year later, so it covers every election since 1872 with one method: Shiller's monthly series through 1927 (his own average prices), averages of daily closes from 1928. No election window crosses the seam. A chamber counts as lost when the president's party held it before the vote and not after. Midterms where the president's party lost a chamber were followed by +6.3% on average over twelve months, against +17.1% when it lost none (losses minus no loss: Welch t = −1.64 daily; −0.98 monthly). This is the largest gap on the daily measure; on the monthly measure the split by unified government becoming divided is larger (−9.2 points, t = −1.49). The no-loss group includes 5 midterms where the president's party held neither chamber (1958, 1970, 1974, 1990, 1998); among those where it held at least one and kept it, the next twelve months averaged +17.0% (t against losses −1.46). Leaving out 1930, the losses averaged +11.1%. Democrat against Republican wins: t = 1.29 daily (1928–2024) and −0.11 monthly (1872–2024). Over 1928–2024 alone the monthly measure agrees with the daily one (Democrat +12.2%, Republican +5.0%); before 1928 it is the other way (4 Democratic against 10 Republican wins, t = −2.27). The year 2000 counts as divided because Democrats took the Senate in June 2001; counted as unified, "winner also gets both chambers" against "lacks" goes from +8.5% against +4.5% to +6.8% against +7.6%.

Every presidential election since 1872

Every midterm, 1874–2022

Rows in amber text were lower twelve months later (daily measure where it exists, monthly before 1928). Congress after is the party that organised each chamber; an asterisk marks a coding note (hover it, or see elections.csv): a disputed result (1876), close or tied organisation (1880 Senate, 1916 House), chambers that changed hands during a Congress (1930, 2000) and 50–50 Senates (2000, 2020). The monthly columns can disagree with the amber (1928, 1946, 1952, 1972). Through 1980 for presidential elections and 1966 for midterms the exchange was closed on election day, so "+1 day" runs from Monday's close to Wednesday's. The full table is in election_results.csv.

Section 1The four-year cycle

Since about 1950 the year before a presidential election has stood out. Before 1950, no year of the term did.

Calendar-year price return of the S&P 500. The 1968–2025 columns start after Yale Hirsch's Stock Trader's Almanac began to publish the cycle, so they are the closest thing here to an out-of-sample test, though the start year was chosen after seeing the data. The "weakest midterm year" is not stable either: over 1950–1987 the midterm year averaged +8.1%, ahead of year 1 (−1.4%). One explanation often offered for year 3 is that administrations take unpopular steps early and push for growth before the next vote; this page does not test it.

Since 1950, year by year of the term

Calendar-year price return; the small figure after Best and Worst is the year. In 6 of the 19 complete terms since 1949 (the table above starts in 1950), year 3 was the term's best year: about one in three, where chance alone gives one in four. What sets year 3 apart is that it rarely lost.

Every term since 1949

The small letter is the party that won the election before the term. The best year of each term is in bold; shaded terms lost money over the four years. Whole term compounds the four calendar-year price returns. Before 1937 a term began in March, so Year 1 includes January to March of the previous administration. The 2025–28 row shows 2025 only.

Section 2After election day, 1928–2024

Midterms have been followed by unusually good six months. The twelve-month edge is smaller, and presidential elections show none.

Return from the last close on or before election day. The exchange was closed on every presidential election day through 1980 and every midterm election day through 1966; in those years "+1 day" runs from Monday's close to Wednesday's. A month is 21 trading days, a year 252. Midterm minus any-day mean, t: 3.04 at six months, 1.40 at twelve. The twelve-month count is sensitive to the horizon: 1946 finished −0.13%. "Any trading day" starts from every trading day since 1928, for comparison. 25 presidential and 24 midterm elections.

When the president's party loses the House

Midterms 1930–2022, measured from election day. The president's party lost control of the House in 1930, 1946, 1954, 1994, 2006, 2010, 2018, 2022. 1930 counts as a loss because Democrats organised the House in December 1931 after deaths and special elections, though Republicans led narrowly on election night; it was also the worst midterm for stocks. Counted as a hold, the 7 losses averaged +8.7% at six months and +13.3% at twelve, against +11.6% and +12.3% for the rest. Lost vs rest, Welch t at twelve months: −1.04, or 0.17 with 1930 moved. The gap rests on one year; with 8 cases this is an anecdote, not a pattern.

Twelve months after a presidential election, by result

The three worst years followed 1936 (D win, −35.2%), 2000 (R win, −20.4%), 1956 (R win, −14.6%). With 12 or 13 elections in each group, one crisis moves a mean by several points (Democrat vs Republican, Welch t = 1.29).

Section 3The midterm low

The rebound from a midterm year's low looks impressive, but the low of almost any year is followed by a rebound.

A year's lowest close is known only once the year is over, so measuring from it builds in hindsight. The fair comparison is the same measurement for every other year. Midterm lows did somewhat better, but not reliably (Welch t = 0.93).

Section 4Party and Congress

Neither the president's party nor the mix of parties in Congress separates good years from bad ones reliably.

Each calendar year belongs to the party of the president in office for most of it. Since 1872 no president who died or resigned was succeeded by the other party. Democrat vs Republican, Welch t: 0.86 (1873–2025), 1.30 (1950–2025).

By who controls Congress

Mean calendar-year return; the small figure is the number of years. Each year uses the Congress elected the November before. Cells of 6 to 10 years are anecdotes: the weakest cell over the long record, a Republican president with a split Congress, turns positive after 1950. 2001 counts as a Democratic Senate (50–50, changed hands in June); 2021–22 as Democratic on the Vice President's tie-break.

Section 5Does the market call the election?

When stocks rose from August to October, the incumbent party usually kept the White House, but the rule has missed three of the last seven elections.

Measured from the last close of July to the last close before election day. The result depends on the window: other plausible windows give 15 to 20 of 25, and 1960, 1976 and 2000 turned on moves of under 1%. The economy that moves stocks also moves votes, so a good record here is not evidence that markets know the outcome.

Section 62026 so far

2026 is a midterm year, so far a better one than most, and the president's party is projected to lose both chambers.

The midterm vote was held on 3 November 2026. On election night CBS, ABC, NBC and CNN projected that Democrats will win both the House and the Senate; some races were still uncalled, and the data record these projections. From January 2027 a Republican president will face a Congress held by the other party, the middle column of the Congress table in Section 4. The president's party has lost the House at a midterm 8 times since 1930 (Section 2, counting 1930 as a loss), so 2026 would be the 9th.

Neither table shows a reliable difference, and the 2026 windows from election day have barely begun, so the result is not translated into an expected return here.

Notes and limits

• Prices. Robert Shiller's monthly S&P composite for 1871–1927; Yahoo Finance daily closes of the S&P 500 index (^GSPC) from 1928. Price returns only. Before 1957 the index had fewer than 500 stocks (90 from 1926, and Shiller's composite before that).
• Elections. Winners and chamber control are compiled from the historical record. Where control was disputed or changed during a Congress (1880, 1916, 1930, 2000, 2020) the coding is noted in the data file.
• Timing. A year belongs to the party in the White House for most of it. The last March inauguration was in 1933 and the first January one in 1937. Election-day windows start from the last close on or before the vote.
• Statistics. t is a mean divided by its standard error; Welch t compares two groups. Overlapping windows, a few dominant crashes and many comparisons all make the true uncertainty larger than these numbers suggest.
• Data. elections.csv · election_results.csv · calendar_years.csv · election_windows.csv · midterm_lows.csv

Research, not investment advice. Every reader gets the same page; nothing here is a recommendation to buy, sell or hold anything. Past patterns around elections describe the past. All notes