Daily Trend Follower

The story in one line

The classic 50/200-day SMA crossover on eight FX majors plus gold, run honestly on real 2018–2026 data, loses money: the account ends at $4,251 (−15% over 8.5 years) with a −42.3% max drawdown, net profit factor 0.94, Sharpe −0.02, win rate 36.7% across 934 trades. That is below the PF > 1.2 deployment bar by a wide margin.

The cause is structural, not cost-related: of the eight instruments, only gold (XAUUSD) made money (+$2,157). All seven FX majors lost. The majors ranged rather than trended during 2018–2026, so a trend-following system has nothing to capture and bleeds whipsaw plus spread. The historical trend edge in this basket now lives almost entirely in gold — and one instrument carrying a portfolio is the opposite of robust.

This is the third archetype to fail the bar (basket-grid PF 0.51; Fury/CORE PF 0.73–0.90 net; trend-D1 PF 0.94). The consistent lesson: simple published edges, costed honestly on the major-FX basket, do not clear PF 1.2. Trend-D1 retired.

Per-pair attribution

Gold is the only green bar. Strip it and the seven-pair FX book loses ~$3,000 (−60% of capital) on its own.

Pair	Net P/L	of which swap	Trades	Win rate
XAUUSD	+$2,157	−$389	112	43%
EURUSD	−$123	−$36	105	39%
USDJPY	−$258	+$8	113	36%
AUDUSD	−$268	−$37	132	39%
GBPUSD	−$361	−$56	116	34%
USDCAD	−$445	−$92	125	42%
USDCHF	−$701	−$132	120	30%
NZDUSD	−$853	−$55	111	30%

A 36.7% win rate is normal for trend-following — a few big winners pay for many small losers. The problem is the winners weren’t big enough: gross profit $12.9k vs gross loss $13.7k. Gold’s secular uptrend (≈$1,300 → ≈$3,300/oz) is the entire edge.

Parameter-robustness sweep

If an edge is fragile to small crossover changes, it’s overfit. Here the opposite is true — the strategy is robustly absent. Nothing clears 1.0, let alone 1.2, and there is no parameter island of profitability to overfit to.

Crossover	PF	Sharpe	CAGR	Max DD	Trades
50 / 200 (spec)	0.94	−0.02	−1.9%	−42.3%	934
100 / 200	0.95	0.01	−1.3%	−30.5%	745
50 / 100	0.99	0.13	+0.8%	−50.2%	1030
20 / 100	0.91	−0.04	−2.7%	−51.3%	1354

Both out-of-sample halves also lose independently: train 2018–2022 PF 0.96, validate 2023–2026 PF 0.97. There is no regime in the 8.5-year span where the literal spec produced a positive PF.

Structural, not cost

To separate implementation from premise, the canonical golden/death-cross variant — always in-market, hold until the cross flips, stops removed — was tested as the most charitable, textbook implementation. Multi-month holds, no whipsaw, no stop-churn: it reaches PF 1.01 in-sample (break-even) with a 129-day median hold — recovering ~7 PF points by killing whipsaw and the −3 ATR stops. But it collapses to PF 0.72 / Sharpe −0.37 out-of-sample (2023–2026). Even its in-sample break-even is entirely gold (+$2,679) papering over NZD −$1,027 and CHF −$1,210.

The diagnosis: better implementation confirms the literal rule does leak value to whipsaw and stops — but fixing all of that still leaves a strategy that only breaks even in-sample and loses out-of-sample, because the underlying premise — that the 8 majors trend — was false for this decade. Costs make a structurally edgeless strategy worse; they are not the reason it has no edge.

Verdict: FAIL (PF 0.94). Do not deploy. Don’t run 50/200 (or any tested variant) on the 8-major basket — it loses across every cut. The only signal worth a follow-up is that daily trend-following on gold carried real edge here, but that is one instrument, not a portfolio, and out of scope for “forex.” The transferable lesson: pivot the instrument set (FX majors are efficient and range-bound), not the signal.