Past issued errors gave no clear 90% Nasdaq interval score gain over t(5) in a 2018–2025 archive test

We tested central 90% intervals for the next daily Nasdaq Composite log price return. A fixed rule used 504 errors from earlier issued forecasts. Each error entered only after both target prices became available in a dated ALFRED record, and later revisions did not replace it. All rules used fixed settings. The main test has 1,909 dates with a current preceding close in 2018–2025. Against a Student-t model with five degrees of freedom and the same risk scale, the issued-error rule changed mean interval score by +0.521 basis-point score units. The paired 95% date-block interval was -6.911 to +7.589. Lower score is better. Coverage rose from 87.32% to 89.63%, while mean width rose from 426.52 to 461.32 log-return basis points. Thus the test gives no clear score gain. Planned sample and block-length controls give the same conclusion. A shorter error window, separate signed bounds and rebuilt historical errors also give no clear score gain. Independent date, rank, scale and score checks passed. A complete repeat matched all nine numerical files. The package preserves code, all forecasts, source hashes and requests, aggregate results, uncertainty draws, figures, negative findings and repeat instructions. Provider source data and measured outcomes remain in separate retained records under their original terms. This is a retrospective risk-forecast test for one price index. It does not establish stock direction skill, conditional coverage or trading profit.

return intervals · Nasdaq Composite · dated data · forecast calibration · interval score · delayed feedback · negative result