Research

Compression → expansion — structure fixed, no edge

V1 std_logp onset is lagged; V1b restores structure. Incremental vol vs slope+ER ≈ 0 — context, not a trigger.

In brief. Vol compression followed by expansion is a seductive narrative — it needs a strict protocol. V1 (std_logp) reacts after the move (pre-onset AUCΔ ≈ 39 vs post ≈ 6). V1b fixes lag and clustering. Ablation: adding vol on top of slope + ER adds ≈ 0 (Δ log-loss −0.001). Rolling z-score persistence is a mechanical filter artefact.

Why I ran this study

The “compressed vol → expansion” story is everywhere. I wanted the same rigour as the rest of the lab: causal onset, progressively tighter matching nulls (TOD → compression → pre-move), incremental ablation.

Method arc

  1. Naive V1 — escape event vs null already saturated (0.89 vs 0.86 at H=10)
  2. Lag diagnostic — detector sees displacement in progress, not pure compression
  3. V1b — early + causal reformulation
  4. Ablation — does vol add anything beyond slope + efficiency ratio?
  5. Filter null — is z persistence a “regime”?

Key results

V1 vs V1b — correction du detection lag et du clustering

Under N_C null (compression + controlled): Δ MFE50 H=10 ≈ +0.06 only — no exploitable edge even with tight matching.

Ablation incrémentale — surplus vol vs base (slope + ER) ≈ 0
QuestionAnswer
Does V1 prove compression→expansion?No
Does V1b fix structure?Yes
Vol surplus?≈ 0
z persistence = regime?No — ACF excess ≈ +0.004 vs IID-RW null

Takeaways

We thought we saw a vol process; the diagnosis shows a lagged detector, then a clean structure with no incremental edge.

Recommended use: std_logp and derivatives as maturity / activity context — not a standalone trading onset.


Research note — Aug 2026. EURUSD · TIMB · 2023.