Retired: 2026-07-12
48.18%
break-even 56.7% at its realized average win/loss
re-measured (top 10 coins · 1H · 2026-10-06): 50.43%
Percentage of trades that were profitable
0.71
re-measured (top 10 coins · 1H · 2026-10-06): 0.74
Gross profit / gross loss ratio
-100% (account ruin, top-10, fixed nominal sizing)
Net profit after all fees and losses
100%
Largest peak-to-trough decline
as published (10 coins · 1H · 2026-08-14)
Marks next to a metric: ~ = borderline (PF between break-even and 1.5; win rate ≤5 pts above break-even, or 50–55% if none shown) · ! = below the bar, or the strategy is already retired or shelved — a closed verdict outranks its number. No mark = clears it and is still standing.
Loses in both the bull and the bear window of our backtest. 'Retired' in the strictest sense — no regime where it worked.
Card re-measured: 2026-10-06 · 1,511 trades analyzed
Status: RETIRED (2026-08-14 live re-measure). At the preset defaults this configuration ruins the account (-100% under fixed nominal sizing). The page documents the actual code logic and the measured outcome — it is not a trading recommendation.
Overview
Mean Reversion LONG is the textbook dip-buy: when price stretches far below its short-term average and an oscillator confirms “oversold”, buy the snap-back. The engine implements exactly that — a standard-deviation band around a simple moving average, gated by RSI. The measurement shows what happens when the textbook meets a trending crypto tape at 1H granularity.
How It Works
- Band — SMA(20) with a ±2.0σ band; σ is the population standard deviation (ddof=0), the same band convention as the repo’s Bollinger-family code
- Oscillator — RSI(14) with Wilder’s smoothing
- LONG — close below the lower band and RSI below 30
- The code defines a mirrored SHORT side (upper band + RSI above 70); this preset runs LONG only
- Entry — next bar’s open after the signal candle
- Exit — TP 8% / SL 10% (preset defaults)
Results (live engine re-measure, 2026-08-14)
Scope: top-10 coins, 1H candles, leverage 5x, fee 0.05% per side, window 2023-12-10 → 2026-08-14, fixed nominal sizing.
| Metric | Value |
|---|---|
| Profit factor | 0.71 |
| Total return | -100.0% (account ruin) |
| Win rate | 48.18% |
| Max drawdown | 100.0% (account equity) |
| Sharpe | -1.57 |
| Trades | 1,455 |
Why It Loses
- In a trending decline, “oversold” re-arms itself. The band is anchored to SMA(20), which falls with the market — so a sustained downtrend keeps producing fresh “2.0σ below a falling mean + RSI under 30” entries, each one a lower dip than the last.
- The exit asymmetry runs against you. SL 10% is wider than TP 8%; with a 48.18% win rate the average loss outweighs the average win, and PF 0.71 follows.
- Leverage converts a losing edge into ruin. At 5x with 0.05% fee per side, a profit factor below 1 across 1,455 trades grinds the equity to 0 — the -100% is the account dying, not a large-but-survivable loss.
Caveats
- -100% is this repo’s ruin notation: under fixed nominal sizing the account equity reached 0 inside the window.
- The verdict is for the LONG preset at its defaults on the stated scope; the SHORT side in the code was not part of this measurement, and no claim is made about it.
- Retired status: reproducible in the simulator, presented for education only.