US returns with a two-day archive delay gave no clear Nikkei forecast gain with rolling OLS in 2018–2025

We tested whether delayed US stock returns improve next-session Japanese stock return forecasts. Historical ALFRED records supplied Nasdaq Composite and Nikkei 225 inputs. A fixed two-calendar-day archive delay kept inputs before Tokyo entry. Rolling ordinary least squares used up to 756 earlier available targets. In 1881 common-market origins in 2018–2025, the model with delayed US and Japanese inputs had R squared of -0.50% against a past mean. The paired 95% 21-origin-block interval was -1.10% to +0.15%. The test gave no clear gain. Direction accuracy was 51.20%, against 53.38% for an always-positive rule. A control used the US close on the Tokyo entry date. That close occurred after entry, inside the target interval. This deliberately unavailable control gave R squared of +21.67%, with interval +14.31% to +29.86%. The large score gap shows why daily date labels do not establish input availability. Calendar checks removed numerical values on 49 Japanese non-session dates and one US holiday. Fixed window, freshness and period checks gave no clear primary gain. The test concerns two price indexes, one delayed archive system and a retrospective period. It does not test a direct market feed, establish a new transmission mechanism, or show trading profit. The package preserves code, source requests, forecasts, fits, aggregate scores, uncertainty draws, figures and repeat instructions. Provider-owned prices and measured outcomes are retained separately.

cross-market forecasts · information timing · point-in-time data · Nikkei 225 · Nasdaq Composite · out-of-sample tests · negative result