An equal mean gave no clear gain over daily model choice for Nasdaq variance forecasts in a 2018–2025 archive test

An equal mean of GARCH and GJR risk forecasts gave no clear gain over a daily model choice in 2,010 Nasdaq Composite test origins from 2018 to 2025. The choice used the last 252 forecast scores with outcome labels available in the ALFRED record at each cutoff. The target was the squared next-session log price return. The mean loss gain of the equal mean was 0.00183, with a 95% 20-session block interval of -0.01852 to +0.02893. The equal mean beat an exponential variance baseline, but its gain over GJR alone was uncertain. Removing 2020 changed the primary gain to -0.00987. Inputs and selection feedback followed dated release records, including a long source outage. We removed a known false Good Friday close before fitting. This is an extension on a previously studied market history, not an independent market sample. The package preserves corrected fits, 2,765 combination forecasts, source provenance, all planned checks, 200,000 resampled contrasts, figures and repeat code. It makes no profit or return-direction claim.

variance forecasts · forecast combination · model selection · dated data · Nasdaq Composite · negative result