Qm
Foundational

The Super Bowl Indicator as a Data-Mining Warning

A famous spurious correlation, whether an original NFL or AFL team wins the Super Bowl supposedly predicting whether stocks rise that year, used as a cautionary example of pattern-matching on coincidence.

The Super Bowl indicator claims that if a team from the original National Football League wins the Super Bowl, the stock market rises that year, and if a team from the old American Football League wins, it falls. For several decades after the indicator was popularized, it had an eerily high hit rate, well above what chance would predict, and it is still cited today, half-jokingly, as a market "signal."

There is no economic mechanism connecting a football game outcome to equity returns; the indicator's early track record is a textbook case of data mining, sometimes called the look-elsewhere effect. If you test enough candidate "predictors" against a market's historical return series, football winners, sunspot cycles, hemline lengths, which president is in office, some will show an impressive-looking correlation purely by chance, simply because so many candidates were tried. The Super Bowl indicator survived scrutiny for a while because it was found and publicized after the fact, from among the many things one could have checked, rather than proposed in advance from a reasoned hypothesis; its hit rate has since degraded toward what pure chance would predict.

The indicator is taught less as a market signal and more as a standing warning: any backtested pattern discovered by scanning many candidate variables against one price history needs a plausible causal story and out-of-sample validation before it deserves any trust, because the search process itself manufactures apparent predictors that have no reason to keep working.

The Super Bowl indicator's historical hit rate is a case of data mining, not a genuine signal, testing enough unrelated candidate predictors against market history guarantees some will look impressively correlated by chance alone, which is why any backtested pattern needs a causal rationale and out-of-sample evidence, not just a good historical fit.

Discussion

Sign in to join the discussion · reading is open to everyone

💡 Discussion rules

  1. Ask and answer about this concept. Off-topic gets removed.
  2. No homework dumps. Show what you tried first.
  3. Corrections are welcome. Cite a source when you claim an error.

Loading discussion…

Related concepts

Further reading

  • Krueger & Kennedy, An Examination of the Super Bowl Stock Market Predictor (1990)
ShareTwitterLinkedIn