US unemployment changes gave no clear gain in 127 monthly Nasdaq return forecasts with dated inputs in 2015–2025

We tested whether a three-month change in US unemployment improves next-month Nasdaq Composite price-return forecasts. Each monthly fit used only the data version available before entry, with a two-day cutoff. The fixed test had 131 planned origins in 2015–2025. One missing unemployment input and three stale price inputs left 127 main test months. A ridge model with unemployment had 9.38% more mean squared error than a past-return mean. The 95% time-block interval for error reduction was −25.37% to +1.34%. The two fixed time periods gave opposite signs. Adding unemployment to a price-trend model also gave no clear gain. A later descriptive audit found that June and July 2020 account for 99.68% of the net extra error. The June input was 33.19 training standard deviations above its mean. Revised-input and first-release target checks did not change the conclusion. In a price-index and cash simulation with a 10-basis-point trading cost, the unemployment rule gave lower point growth than the mean rule. This is a historical test of one lag and one linear model family. It does not test announcement surprises or establish a general trading result. The package preserves code, dated forecasts, model states, source requests and hashes, uncertainty draws, missing cases, figures, and independent checks.

stock returns · unemployment · point-in-time data · Nasdaq Composite · real-time data · return forecasts · negative result · paper trading