Research

Multi-scale pullback — exists yes, net tradability no

On EURUSD TIMB, short vs long trend conflict exists and predicts OOS — but net EV collapses after costs.

In brief. When the short trend (DC 5p) diverges from the long one (DC 15p), the market enters a measurable “pullback” state. In 2023, ~79% of episodes resolve as Resume (R), stable disc/dev. The causal score predicts R vs F OOS (ΔLL up to +137). Net tradability fails: EV −0.22 pip on dev under 0.2 spread.

Why I ran this study

The most visual price-action motif: a background trend with a shorter counter-move. Before trading anything, I asked whether this is a process (R/F states with competing risks) or labelled noise.

The gates 0→4 framework enforces: existence → detection → predictability → tradability — without tuning upstream gates on PnL.

Definition (summary)

An episode starts when (long DC 15p vs short DC 5p), with minimal conflict of 3 bars. Outcomes:

OutcomeSymbolMeaning
ResumeRShort realigns with long
FailFLong structure breaks
CensorCSeries end / split

R and F are structure states, not PnL.

Key results

Framework gates 0→4 — pullback multi-échelle (EURUSD, 2023)
GateVerdictReading
0–1PASSCircular-shift null rejected; stable hazards
2PASSCausal score separates R/F, especially intra-episode
3PASSOOS ΔLL +32 (onset) to +143 (age 8) vs vol context
4WEAKNet EV disc +0.14 → dev −0.22 pip
Gate 4 — EV net (pips) disc vs dev, même cellule gelée (q=0.8, d=4, cap=32)

Score quintiles (dev): Q1 → 73.7% Resume, Q5 → 83.5% — the score calibrates outcome, not optimal entry timing.

Takeaways

QuestionAnswer
Does the pullback exist?Yes — beats null, stable R rate
Can we predict outcome OOS?Yes — significant ΔLL
Trade it as-is?No — unfavourable MFE/MAE asymmetry

Strong state predictability does not imply an executable edge.

Recommended use: regime feature / filter in a downstream stack (HMM, Tr8dr permission) — not a standalone signal.


Research note — Aug 2026. EURUSD · TIMB · disc/dev split 2023.