A skewed Student-t model cut daily Nasdaq return log loss against a symmetric model with mean zero in a 2018–2025 archive test

Can an asymmetric return density improve a daily forecast after a symmetric heavy-tailed model uses recent volatility? We test a known skewed Student-t density for 2,011 Nasdaq Composite price-return sessions in 2018–2025. Monthly fits use 1,260 earlier residuals from dated ALFRED records. Both forecast means are zero, and both models use the same volatility signal and delivery cutoff. The skewed model reduced mean log loss by 0.01251 nats per date, with a paired 95% 21-session block interval of 0.00584 to 0.01875 nats of gain. The gain remained on fresh-input dates and without 2020. The model had lower annual loss in six of eight years. Lower and upper 5% quantile score gains were uncertain. A later direction control found no clear Brier gain over the past positive-return rate. This is a test on a known historical sample for one price index and one horizon. It does not establish reliable direction timing or trading profit. The package preserves code, source hashes and rights, all forecast parameters and fit attempts, aggregate scores, uncertainty draws, negative controls, and exact repeat checks. Protected source outcomes remain in a separate retained research ledger.

density forecast · skewed Student-t · Nasdaq Composite · point-in-time data · chronological test · log score · negative control