Correlation Is Not a Signal

Stocks and job openings moved together for twenty years, then split the month ChatGPT launched. What that chart does and does not prove — and the same trap on your own charts. This free lesson covers: The chart everyone shares, Check the dates, Look for a third cause, Name it, Check what the line actually measures, Two axes, one illusion, The first question, The same trap on your chart, Put it together.

Measuring Your Edge · Learn · about 3 min

The chart everyone shares

From 2001 to 2019 the S&P 500 and US job openings moved together: a correlation of 0.93 on monthly data. Since late 2022 they have moved in opposite directions — stocks up, openings down.

The popular explanation: ChatGPT launched on 30 November 2022, so AI must be replacing workers while Wall Street cheers. It makes a great story. That is exactly why it is worth checking.

Academy

Going deeper

More on correlation is not a signal

Data: S&P 500 month-end closes (FRED series SP500; before October 2016, the daily history bundled with the Python arch package, which matches FRED exactly where the two overlap), US job openings (BLS JOLTS, FRED series JTSJOL) and the effective federal funds rate (FRED series FEDFUNDS), as published in October 2026.

The 0.93 correlation is between monthly levels from 2001 to 2019 (0.92 using the logarithm of the S&P 500). From December 2022 to August 2026 the same correlation is −0.83. Correlations of levels between two trending series are high almost by default, which is part of the lesson: a big number is not evidence of a link.

ChatGPT launched publicly on 30 November 2022. Job openings peaked at 12.30 million in March 2022, the month of the Federal Reserve’s first rate increase of that cycle; the effective funds rate rose from 0.08% in February 2022 to 5.33% from August 2023.

Common mistakes4

Accepting the date on the chart as the date something began, without checking where the change actually started.
Reading job openings as employment. Openings are vacancies; the number of people in work is a different series.
Treating two indicators built from the same price data as independent confirmation.
Trusting a relationship because it fits a story you already believe.

From experience2

For any chart claim, ask three things: what changed first, what else changed at the same time, and what each axis really measures.
Before trading a pattern you noticed, write down what would prove it wrong, then test it on data you did not look at while noticing it.

Questions people ask2

So is AI not affecting jobs?

This chart cannot tell you either way. The fall in openings began before ChatGPT launched and lines up with rising interest rates, so the chart does not show AI caused it. Whether AI changes hiring over time is a separate question that needs different evidence.

Is correlation useless for trading?

No. It is a starting point for a question, not an answer. A relationship becomes usable when it holds on data it was not fitted to, has a mechanism you can explain, and survives the obvious third causes.