Do three earlier stock-split findings hold against a yardstick whose false-alarm rate stays on target?
Study 40 · No reading (stopped by its own checks) · recorded 6 October 2026 · answer: Not settled · full paper with figures
The question
Studies fifteen and nineteen (reverse splits) and twenty-two (pooled forward splits) read share-price moves against a yardstick of what chance gives. On real reverse-split days shifted so that nothing is there to find, that yardstick flagged an effect about ten to twenty-six times in a hundred against a nominal two and a half; we re-judge them against one on target.
Rules fixed before the result
Fixed in advance: a yardstick could be used only if, on made-up data built with nothing to find, it flagged an effect between one and four times in a hundred against a nominal two and a half; otherwise the study stopped before reading real data. The plan allowed an alternative that a calculation showed would pass. The four tried were named, in order, before that calculation ran, and the rate itself, not its range of sampling error, decided.
Data and method
Each earlier finding was to be re-judged with its own split days, measure and rules, changing only the yardstick: a redraw of whole companies' split days, shifted so that nothing is there to find. The yardstick, and then two alternatives for the median and two for the mean, were each run on many made-up data sets shaped like the real split days, skewed and symmetric, and their false-alarm rates counted.
The answer
Not measured: the re-check could not be made. Before any real data was read, the yardstick had to pass a size check on made-up data built with nothing to find, flagging an effect between one and four times in a hundred against a nominal two and a half. The registered yardstick flagged an effect a little too often, so the study stopped there; four alternatives, named before a calculation on made-up data only, each failed the same check on at least one test set, for the median move and for the mean alike. No real data was read, so the three earlier findings are neither upheld nor corrected by this study and keep the status they had. The registration expected the findings to be re-read, and gave this ending a low chance.
What it changes
Nothing in the study's registered plan changed during the run. The re-checks of the three earlier findings end as not measured, because no yardstick passed the false-alarm check; as registered, the public answers that would have rested on this re-check stay unpublished, and the three findings keep the status they had. Reopening the re-check with another yardstick is not proposed.
What remains uncertain
The three studies had already read these split days; only made-up data was used here. One alternative for the mean came close: on two test sets its rate sat just above the ceiling, within its sampling error, and the rule judged the rate itself. Two alternatives were tested with a scale for their spread, fixed in the instructions before the run, larger than the usual form's, so their failure does not show the usual form would fail.