The “False market connections” passage warns about misleading time-series comparisons. It uses the deliberately absurd association between storks and births to expose the weakness of causal interpretations based on similar-looking lines.
Challenge the common driver
Population, inflation or time can cause unrelated measurements to move together. As a teaching extension, compare changes as well as levels, consider common drivers and examine whether the relationship adds information beyond an appropriate baseline.
Worked example
Two fictional series both rise each year: cumulative spending and the number of archived photographs. Their rising levels may create an impressive chart without a useful causal link.
Case connection
Co-movement is not a full causal account. The staff report compared buying activity with competing explanations of the rise.
GameStop: investigate the popular story
Source-grounded facts
SEC staff found short covering contributed during some intervals, but positive sentiment sustained GameStop’s weeks-long rise.
Context
GameStop’s January 2021 rise drew attention to retail participation, short selling and possible feedback loops. The SEC staff report examined transactions to test several competing explanations.
Outcome
The report attributed the sustained weeks-long rise to positive sentiment rather than short covering alone. This is a staff interpretation of a specific episode, not a complete account of every participant’s private motives.
- Staff observed some intervals when heavily shorted accounts bought shares while the price rose.
- Those purchases were a small part of overall buying, and prices stayed elevated after their direct effect would have faded.
- Staff did not find evidence that a gamma squeeze explained GME’s January episode.
Case analysis
Two observations moving together can result from a shared cause, a direct link or coincidence. Define which account you are investigating and what evidence distinguishes it from the others. The staff report’s comparison of different buying mechanisms models that approach. A correlation coefficient alone cannot perform the same causal discrimination.
Try it
Find a claimed correlation. Ask what common trend or third variable could generate it.