Skip to content

v1b_macro_regressor.md

Latest commit

 

History

History
38 lines (29 loc) · 2.87 KB

v1b_macro_regressor.md

File metadata and controls

38 lines (29 loc) · 2.87 KB
Raw

V1b — Macro-event regressor ablation

Question. Does adding a FRED-derived macro_event_next_week flag (FOMC + CPI + NFP) to the F1_emp_regime log-log model close the shock-tertile coverage gap documented in §9.2 of the paper (~80% realized at τ=0.95 in the highest-realized-z-score tertile)?

Calendar. 324 macro events 2014–2026 (31 FOMC decision dates from DFEDTARU, 146 CPI releases, 147 NFP releases). 48% of weekends have a macro event in the following week — the flag is frequent; if shock weekends concentrate within this 48%, we'd expect to see differential coverage improvement.

Pooled — τ = 0.95

variant scope claimed n realized sharpness_bps
M0_deployed pooled 0.950 5486 0.923 253.726
M1_with_macro pooled 0.950 5486 0.921 254.138
M2_macro_swap pooled 0.950 5486 0.922 254.023

Shock-tertile (post-hoc realised-move tertile)

variant scope claimed n realized sharpness_bps
M0_deployed realized_shock 0.680 1855 0.322 102.462
M1_with_macro realized_shock 0.680 1855 0.319 103.164
M2_macro_swap realized_shock 0.680 1855 0.318 103.098
M0_deployed realized_shock 0.850 1855 0.584 165.521
M1_with_macro realized_shock 0.850 1855 0.579 166.356
M2_macro_swap realized_shock 0.850 1855 0.578 166.230
M0_deployed realized_shock 0.950 1855 0.803 264.404
M1_with_macro realized_shock 0.950 1855 0.796 264.586
M2_macro_swap realized_shock 0.950 1855 0.796 264.903
M0_deployed realized_shock 0.990 1855 0.921 420.110
M1_with_macro realized_shock 0.990 1855 0.922 419.048
M2_macro_swap realized_shock 0.990 1855 0.923 417.747

Reading

  • M0 (deployed) is the baseline. M1 adds the macro flag, M2 swaps earnings for macro.
  • A meaningful improvement is +1pp realized in shock-tertile at τ=0.95 with comparable sharpness — a half-pp shift is within bootstrap noise at this n.
  • If M1 ≈ M0 (no detectable lift), shock weekends are not macro-event-driven. This is itself a paper-relevant finding: §9.2 stays as 'structural ceiling, mechanism unidentified.' If M1 > M0 with CI excluding zero, deploy M1 and update §9.2 to acknowledge the partial fix.

Raw: reports/tables/v1b_macro_ablation.csv, reports/tables/v1b_macro_ablation_shock.csv, data/processed/v1b_macro_calendar.parquet.