The Edge That Wasn't
All 81 experiments › #31

#31 — Currency-strength lower-TF viability

2026-05-18 (Session 061)
Verdict: negative · ⛔ closed negative — no intraday currency-strength signal
The H4 contrarian StrengthSpread alpha does not exist intraday: every hold period up to ~43 hours is negative on average, and only the ~10-day hold regime (already known) breaks even.

Hypothesis — what it tests

Whether the proven H4/multi-day contrarian StrengthSpread signal holds at M5/M15/H1 signal granularity with intraday hold periods.

Description

The csi_factor_study project had established one real contrarian signal: StrengthSpread at H4 with a 64-bar hold (OOS Sharpe 0.59). This sweep asked whether that currency-strength effect works faster — at M5/M15/H1 signal granularity with intraday holds. StrengthSpread was computed as the difference of cross-sectional z-scored rolling-return strength between a pair's two currencies, swept over 4 signal timeframes x 11 hold periods x trend/counter x 12 pairs (1,056 configs), plus a top-3-vs-bottom-3 currency portfolio. Results were unambiguous: mean OOS p/d was negative at every hold up to 512 bars (5-minute hold: −365.9 p/d, 0% positive), the loss shrinking monotonically as the hold lengthens — a pure spread-amortization curve. Only at the ~10-day hold (the regime csi_factor_study already found) did 46% of configs go positive. Conclusion: currency strength is a multi-day mean-reversion effect; there is no intraday version.

Key result

Mean OOS p/d by hold: 5min −365.9 (0% positive) → 42h40 −1.5 (22%) → ~10d −0.2 (46%); portfolio variant uniformly negative in all 32 configs; signal timeframe irrelevant (aggregate means −81.7 to −86.3 for all four)

Indicators

Currency Strength Index (cross-sectional z-score of rolling log returns)StrengthSpread

Algorithms

grid search parameter sweepIS/OOS split validationcross-sectional z-scoring

Code

research/experiments/csi_strength/backtest_csi_strength.pyresearch/experiments/csi_strength/results.csv

Interactive version (search, filters, figures): the experiment explorer.

← #30 RACS reversal-accumulator IC study#32 P&F momentum vs counter-trend →
This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.