#49 — No-VPS migration + SMA16 momentum

2026-05-24 (Session 072)

Verdict: positive · ✅ positive — validated and deployed live (later drained with the momentum book)

A 5-phase sweep finds SMA16 momentum with lags (8,10,15) on H1+M30 agreement, TP=20p: +29.8 p/d OOS, 10/12 pairs pass IS and Monte Carlo, deployed live on account 012.

Hypothesis — what it tests

Is a multi-lag SMA-momentum agreement signal across H1+M30 (all 6 momentum values same sign) a tradeable cross-pair entry with a fixed take-profit?

Description

A two-part session: the Hetzner CX43 VPS was decommissioned with all fx-core containers moved to the local 24/7 'trader' box, and an SMA momentum signal was researched from scratch and deployed. The signal computes SMA(N) on H1 and M30 closes and takes momentum at three lags on each; a trade fires only when all 6 values agree in sign (LONG all >0, SHORT all <0), exiting at a fixed broker-side take-profit. Five phases ran: a single-pair exit sweep on GBP_JPY (simple TP=20p beat every trailing stop), a 12-pair sweep with Monte Carlo, a TP feasibility frontier (5-100p), the key lag+period sweep (all 56 lag triplets from {1,2,3,5,8,10,15,20} x SMA periods {16,22,34,55} = 224 configs x 12 pairs), and MC validation of the winner with 2,000 sign-shuffles per pair. The winner — SMA16, lags (8,10,15), TP=20p — delivered +29.8 p/d OOS portfolio (627 OOS trades, 99.5% WR, mc_p=0.0000, 10/12 pairs passing both gates), +40% over the initial SMA22 guess. It went live as fx-sma-live on account 012 (25u/trade, 10 pairs). Two caveats recorded: the 7 JPY crosses fire nearly simultaneously (one JPY bet x7, not diversification), and the TP-only/no-SL design later proved to hide a catastrophic finite-margin tail (entry 57: +27.3 -> -206 p/d once any stop is added); the momentum book was drained 2026-06-10.

Key result

Winner SMA16 lags=(8,10,15) TP=20p: +29.8 p/d OOS, 1.47 t/d, IS3 10/12, portfolio 627 OOS trades / 99.5% WR / mc_p=0.0000; +40% p/d vs SMA22 (1,5,10) baseline (+21.3); best pairs USD_JPY +6.8, EUR_JPY +4.6, GBP_JPY +4.2 p/d; deployed acct 012 @ 25u/trade

Indicators

SMA(16)multi-lag SMA momentum (lags 8,10,15)H1+M30 timeframe agreementfixed take-profit

Algorithms

grid search (lag/period/TP sweeps)walk-forward validation (IS 3/3 chunks)Monte Carlo sign-shuffle test (2000/pair)

Code

research/experiments/sma_momentum/backtest_sma22_momentum.py research/experiments/sma_momentum/backtest_sma22_sweep.py research/experiments/sma_momentum/feasibility_study.py research/experiments/sma_momentum/lag_sweep.py research/experiments/sma_momentum/mc_validate_winner.py research/experiments/sma_momentum/results/sma_lag_sweep.csv research/experiments/sma_momentum/results/sma_mc_validation.csv services/strategy_sma_live/main.py

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

← #48 Multiscale shock Phase 10 + joint timing-direction model #50 Price momentum M15+M5 live →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.