#42 — Grid-trail strategy

Hypothesis — what it tests

Does a fixed-interval price grid with a latching trailing stop capture mean-reversion bounces (or crossing momentum) profitably net of spread — i.e., is FX at M5 close enough to a random walk for grid-stop win-probability math to hold?

Description

The idea: divide price space into fixed-width boxes (box width a multiple of the IS 90th-percentile spread), enter long on an upward close-to-close grid crossing and short on a downward one, and exit via a latching trail that locks in the high-water profit minus a target. v1 was a reversal system (opposite crossing closes and flips); it went 0/20 walk-forward on all 12 pairs with 21-34% win rates — every reversal exit is a loss, occurring 1.5-3× as often as trail exits, giving structurally negative EV of −3 to −7 pips/trade at any parameters. v2 added enter-only-when-flat plus an absolute stop (2-3× box): 0/480 configs, best IS −1.43 p/d. The diagnosis was instructive: the theoretical ~91% win probability assumes a random walk, but FX M5 has negative short-term autocorrelation — after an upward crossing price statistically reverts, so the stop is hit far more often than random-walk math predicts. Adverse selection plus spread cost killed it. Closed.

Key result

Indicators

Algorithms

Code