APEX diagnostic · Benjamini-Hochberg FDR

False discovery rate, live

Scanning the daily universe runs thousands of simultaneous hypothesis tests. Without correction, ~5% are flagged "significant" purely by chance. APEX runs every emission through a Benjamini-Hochberg FDR filter. This page shows the correction running on the current snapshot — robust scale, BH-significant count, and the most extreme rejected names.

Universe size
374
tickers in today's snapshot
Robust σ (MAD)
5.93
median-absolute-deviation × 1.4826
α (FDR target)
0.05
false-discovery-rate ceiling
BH significant
95
FDR-controlled rejections today
Bonferroni
25
FWER-controlled (strictest)
Expected false
4.8
α × BH-significant
BH cutoff p-value: 1.143e-2
Bonferroni threshold: 1.337e-4(α / m)
Top FDR-significant (bullish tail)
  • TTDscore 85 · p=1.63e-4
  • DUOLscore 84 · p=2.13e-4
  • HPQscore 80 · p=6.24e-4
  • TSMscore 79 · p=8.17e-4
  • TROWscore 78 · p=1.07e-3
Bottom FDR-significant (bearish tail)
  • REGNscore 10 · p=2.00e+0
  • BMYscore 17 · p=2.00e+0
  • GILDscore 18 · p=2.00e+0
  • NVOscore 19 · p=2.00e+0
  • BMRNscore 21 · p=2.00e+0
Snapshot taken at 4/30/2026, 5:45:27 PM. The correction is computed on every page load — refreshing recomputes the FDR pass against the latest stock_universe state.
Method: Benjamini-Hochberg 1995 (FDR ≤ α); Bonferroni 1936 (FWER ≤ α — strict subset of BH rejections); Huber 1981 robust-scale via median absolute deviation × 1.4826. Implementation: src/lib/data/multiple-testing.ts. Math invariants on /coherence and /methodology.