A fixed loss weight gave no clear daily Nikkei 225 risk forecast gain in a 2018–2025 archive test

We tested whether extra weight on past losses improved daily Nikkei 225 risk forecasts. The final test has 1938 sessions in 2018–2025. Each input uses an ALFRED archive from two calendar days before the target, with an independent exchange calendar check. A fixed EWMA rule gave negative returns three times the weight of positive returns of the same size. The primary mean QLIKE gain over symmetric EWMA was 0.01160. Its paired 95% block interval was -0.01685 to +0.04411. The evidence gives no clear gain. An annual scale fit with only past data also gave no clear sign gain. The loss rule beat a reverse-sign control and a slow 252-session reference. A secondary 5% tail score gain lost clear support with longer bootstrap blocks. The package has all 5347 forecasts from both input delays, 22 scale fits, all fixed comparisons, source requests and hashes, code, figures, negative findings and repeat instructions. Separate code checked the labels, timing, risk paths and scale fits. A new process matched 27 numerical files byte for byte. This is a retrospective test of one fixed risk rule. It does not test return direction or a trading strategy.

Nikkei 225 · volatility forecast · asymmetric risk · point-in-time data · ALFRED · EWMA · chronological evaluation · negative result