Stock market curiosities: the omens, tested

The scoreboard: omens against a coin

Every omen here makes a call — up or down, or “this year rises” — and we count how often the call came true. Two yardsticks keep it honest. A fair coin lands in a range around 50% (the grey band, its width set by the number of calls); a rule inside the band did no better than heads or tails. And a stubborn optimist who says “up” every year is the real rival (the black tick; for a rule about some years or months only, the tick is how often all years rose, or all months fell), because the index rose in 101 of 154 years since 1872.

Where we know when a rule was first written down, the hollow dot is its record before that and the solid dot its record afterwards — the only part that is a real test. None of the omens beat the optimist by more than luck allows.

The Super Bowl indicator

The rule: if the Super Bowl is won by a club from the old National Football League, the stock market rises that year; if a club from the old American Football League wins, it falls. Since the two leagues merged in 1970 that means NFC up, AFC down — except for three old NFL clubs that moved to the AFC, the Colts, the Browns and the Steelers, which still count as “up”. Sportswriter Leonard Koppett pointed it out in February 1978, just after that year’s game, when it had never missed.

It really had not: from the first game in 1967 to 1977 it went 11 for 11. Then it was published, and the market stopped reading the sports pages. Since 1978 the indicator has been right in 29 of 48 years (60%). That sits inside the 35%–65% range a coin flip lands in over 48 calls. The optimist who never watches football did better: 77% (37 of 48).

Most of its record rests on two plain facts: old NFL clubs win more often than not, and the market rises more often than not. When the winner said “up”, the index rose 31 times in 38; when it said “down”, the index rose anyway 12 times in 21. Count the conferences strictly, with no exception for the Colts and Steelers, and the rule drops to 54% — a coin. Its worst miss came in 2008: the New York Giants won, and the index fell 40.7%.

The January barometer

“As January goes, so goes the year” — a rule Yale Hirsch wrote about in 1972: if the index rises in January, the year will rise too. It is a different animal from the January effect, the claim that January itself is a strong month, which we tested in Irrational money. Here January is not the prize but the weather vane.

Scored on the whole year, the barometer looks good: right in 70 of 101 years up to 1972 and 37 of 53 since (70%). Still, since 1973 the “always up” optimist did better: 74%. But the whole year includes January itself, so part of every call forecasts the past.

The honest test is the other eleven months. Over 1872–2025, when January rose, February to December rose too in 69 of 102 years (68%); when January fell, the rest of the year fell in 28 of 52 (54%). So a down January was about as good a warning as a coin, which lands in a 37%–63% range over as many years. Since 1973 its calls for the other eleven months were right 64% of the time — below the 72% of “always up”.

The averages differ more than the hit rates: after an up January the other eleven months averaged +6.9%, after a down one +0.5% — though a few big years can move an average a long way. This year January moved +1.1%, a call for “up”; to October 2026 the index is +12.7% for the year.

Years ending in 5, and other decimal magic

The decennial cycle says the last digit of the year matters, and that years ending in 5 are the market's lucky number. In this data the folklore has a point: years ending in 5 rose in 14 of 16 and averaged +19.3%, against +6.4% for all years. Those that did not rise: 1875 (−3.7%), 2015 (−0.01%). At the other end sit years ending in 7: −2.0% on average, down in 9 of 15 years.

Law or fluke? A digit is only a label, so we dealt the 154 years' changes at random into ten groups of the same sizes, 10,000 times. A gap between the best and the worst group as wide as the real one came up in only 3% of the deals — rare, but with dozens of calendar patterns tried over the decades, a few such rarities are to be expected. And the 5s' record of rises is less special than it looks: in 22% of the deals some digit rose at least as often, because among ten digits one usually looks blessed.

Split the record in two and years ending in 5 come out on top both times: +19.6% in 1872–1948 and +18.9% in 1949–2025. It is the most stubborn curiosity on this page — and still one with no reason behind it, spotted by looking at these very years, and resting on 16 data points.

The four-year term cycle

US presidential elections fall in years divisible by four, so every year has a place in the four-year term: year 1 follows an election, year 2 holds the midterms, year 3 comes before the next election, and year 4 is the election year. Folklore says year 3 is the strong one. Only the year's number counts here — no names, no parties.

Year 3 did have the best average since 1872: +10.1%, against +2.8% for year 2, the weakest. But it did not rise most often: year 3 rose in 27 of 38 years (71%), year 4 in 29 of 39 (74%). A gap between the best and the worst group as wide as the real one came up in 28% of the deals — easily luck. Nor is the lead an old one: in 1872–1948 the best was year 4 (+5.5%), with year 3 at +4.1%; only in 1949–2025 did year 3 lead, at +16.1%.

Is October the crash month?

Mark Twain called October “one of the peculiarly dangerous months to speculate in stocks” — and then listed the other eleven. The month's name rests on 1929 and 1987, and on this: of the 30 months since 1871 in which the index's monthly average fell 10% or more, 8 were Octobers.

That is the most of any month. Scatter 30 big drops over twelve months at random and some month collects 8 or more in only 3% of tries: October's pile is unusual. October also holds 6 of the 20 worst months. Its average change, −0.32%, is the lowest of the twelve months in this price data (with dividends, see the calendar in Irrational money).

But crashes do not usually start in October. Of the 13 falls of 20% or more from a record since 1871, 3 began in October and 4 in January. And most Octobers are dull: the month fell in 75 of 156 years (48%), against 41% for all months — so October still rose more often than it fell.

One caution: these are monthly averages of daily prices, which blur single days. The crash of October 1987 came on the 19th, so October's monthly average shows −12.1%, and part of the fall lands in November's.

Make your own omen: a data-mining machine

Every curiosity above was found by people staring at the same few decades of prices — and the longer you stare, the more you find. To show how, we built an omen to order. We took all 84 data series on this site that cover 1980–2024 — prices of commodities and groceries, wages, rates, money, jobs — turned each into three yearly numbers known by the end of a year, and asked which best “predicted” the next year's market over 1983–2006. That makes 252 candidates. Butter, the classic of the genre since David Leinweber's joke that butter production in Bangladesh “explained” the S&P 500, was disqualified: its price series here begins only in April 2018.

For the crown we kept to things that have no business predicting stocks: the 27 commodity and grocery prices among the candidates. The crown goes to Cocoa, change over two years: over 1983–2006 its correlation with the next year's change of the index was +0.46, strong enough to pass a textbook significance test (a fluke this strong turns up only 2.4% of the time — if you test one idea alone). It called above- and below-average years right 16 times in 24. The best fit of all, Median usual weekly earnings, full-time workers 25+, less than a high school diploma (change over the year), reached −0.55 in training — and then flipped to +0.43: a forecaster that is confidently wrong.

Then the years after 2006 arrived. Over 2007–2025 the crowned omen's correlation fell to +0.02 — next to nothing — and it called 8 of 19 years right (42%), a coin's score. The machine was never broken; it was measuring luck. Of all 252 candidates, 12 passed the significance test in training — about the 13 that pure chance hands out at the 5% level — and the training scores said nothing about the later ones (the two correlate at +0.03).

The great collapse

The same 252 candidates, each a line from the strength of its fit in training to how it did afterwards, pointed the same way (a line that ends below zero pointed the wrong way). The top tenth by training fit averaged 0.41 in training and +0.01 afterwards; 11 of those 25 turned the wrong way.

Why curiosities fade

Three things make market folklore look better than it is. Small samples: 16 years ending in 5, 11 Super Bowls before the indicator had a name. Many tries: people have scanned these same years for patterns for as long as there have been almanacs, and every hit gets a name while the misses are forgotten. And no mechanism: a football score or a calendar digit has no way to reach a company's profits.

Real anomalies do exist: researchers keep finding patterns that hold for a while and then shrink once they are published and traded on. A rule that cannot survive the years after its discovery was never a rule. That is the test this page applies, and the one worth applying to any “this always happens” you hear.