Extra weight on past falls cut lower 5% Nikkei 225 return forecast loss by 1.7% in a 2015–2025 archive test

We tested whether a fixed extra response to past price falls improves lower 5% daily return forecasts for the Nikkei 225 index. The rule was fixed before scores were calculated. Each forecast used ALFRED inputs dated at least two calendar days before its target. The primary test had 2,674 exchange sessions in 2015–2025 with input age at most seven days. Extra weight on falls reduced mean quantile loss by 1.69% against the same symmetric filter. The paired 95% block interval was 0.51% to 2.95%. The absolute gain was 0.257 basis-point score units, with interval 0.073 to 0.469. All-age and block-length checks kept a positive gain. Gains were negative in three years, and 2020 and 2025 supplied 66.2% of the net gain. The current calendar removed 50 priced source dates marked as closed. First-common-release and latest outcomes gave the same scores. The study extends a prior Nasdaq test; it introduces no new estimator. It tests one index and a known historical period with a reconstructed calendar. It does not establish profitable trading or future prediction skill. The package preserves code, source requests and hashes, all forecasts, aggregate scores, uncertainty draws, figures and checks. A full repeat matched 15 numerical and check files byte for byte. Restricted outcome ledgers are retained separately under provider terms.

Nikkei 225 · quantile forecasts · point-in-time data · external market test · asymmetric volatility · filtered historical simulation