The Edge That Wasn't
All 81 experiments › #45

#45 — GBP_USD small-box P&F sweep

2026-05-20 (Session 064)
Verdict: positive · ✅ positive backtest — monotonic gain as box shrinks (later bled in paper trading)
Shrinking the GBP_USD P&F box from 5 to 2 pips more than triples OOS pips/day (15.4 -> 54.52), with a 16-pip prior-column filter mandatory at b=2.

Hypothesis — what it tests

Do sub-5-pip P&F boxes give GBP_USD (a low-spread, low-pip-value pair) a stronger FIFO-Trends edge than the standard 5-pip box?

Description

Follow-up to the ATR sweep's hint that GBP_USD prefers smaller boxes: a systematic FIFO-Trends sweep of fixed box sizes 2, 3, 4 and 5 pips on GBP_USD (48 configs per box size, same 3-stage IS-WF/MC/OOS pipeline). Results improved monotonically as the box shrank, with the best config b2_r3_n8_E2_X3c_1_5 reaching 54.52 p/d OOS versus 15.4 at the standard 5-pip box; every winning b=2 config required n_min=8, i.e. a 16-pip prior column as a strong directional filter before entry. The recorded caution was that a 1-box trail at b=2 is only 2 pips against a 1.8-pip median spread (+0.2p effective floor), so live profitability would have to come from column-SMA exits rather than trail exits. The config was deployed to the fx-fifo-paper service as GBP_USD_b2; in the Session 087 epilogue the whole FIFO paper book was stopped at -627.3p over 849 trades, with GBP_USD_b2 itself -99p — the spread consumed the trail in every variant.

Key result

OOS p/d by box: b=2 54.52, b=3 42.9, b=4 40.3, b=5 26.3 (vs 15.4 original); best b2_r3_n8_E2_X3c_1_5; all b=2 winners need n_min=8; trail floor +0.2p (2p trail vs 1.8p P50 spread); paper epilogue GBP_USD_b2 -99p

Indicators

P&Fcolumn-SMA exit (X3c)trailing stop (1-box)n_min prior-column filter

Algorithms

grid searchwalk-forward validationMonte Carlo shuffle testNumba JIT backtest

Code

research/experiments/fifo_trends/backtest_gbpusd_small_box.pyresearch/experiments/fifo_trends/results/gbpusd_small_box_final.csv

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

← #44 ATR-adaptive P&F box sweep#46 FIFO 2-box trail sweep (X3c_2_5) →
This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.