#57 — Finite-margin re-validation of the live momentum book

Hypothesis — what it tests

Whether the live momentum strategies' backtested +30 p/d edge survives when losses must actually be realized — i.e. under a hard stop, finite margin, and OANDA's closeout mechanics — and whether the entry signal beats random entry under the identical exit rules.

Description

After the portfolio-variation campaign put a TP-only / no-stop-loss momentum book live on six OANDA accounts (001-004, 011, 012), a critical review asked the question the validation pipeline had never asked: is the backtested edge realizable with finite margin? A re-validation script replayed each deployed config through the same out-of-sample window with the account's actual balance, units and OANDA's 45%/50% margin-closeout rules, plus an entry-randomization Monte Carlo and forward-return IC. The answer was devastating: the positive expectancy was loss-deferral, not prediction — adding even a 30-pip stop flips +27.3 p/d (99.5% WR) to −206 p/d; all 6 accounts hit margin closeout during the 'validated' OOS window and 5 of 6 went NAV-negative; random entries under the same exits earn most of the headline p/d; and forward IC of the signals is slightly negative everywhere. The book was wound down and the lesson became a standing rule: gate on realizable-under-finite-margin risk, not just statistical significance.

Key result

Indicators

Algorithms

Code