#70 — Trailing-stop bake-off (SuperTrend vs PSAR et al.)

2026-06-18

Verdict: mixed · 🟡 SuperTrend wins, PSAR among the worst

Holding a Donchian breakout entry fixed and varying only the exit, SuperTrend's per-trade ATR-band ratchet beats PSAR by ~20 pips/trade and is the only scheme positive IS+OOS at both H4 and H1 — but the win did not transfer to the live 010 entry.

Hypothesis — what it tests

Whether a better trailing exit (non-accelerating, volatility-aware) materially improves a fixed trend entry — and which trailing-stop family is best for retail FX.

Description

The user asked for PSAR alternatives, so a controlled exit bake-off was run: a fixed Donchian(20) breakout trend entry on 12 pairs at H4 and H1, net of spread, IS/OOS split, with ONLY the exit varied across trailing-stop families (SuperTrend, PSAR, MFE-giveback, chandelier, n-bar, ATR-ratchet, fixed SL/TP). A first run had a bug — globally-computed PSAR/SuperTrend series gave fake above-price exits and survivorship, caught via an impossible 99% win rate — fixed with per-trade-anchored stops and mark-to-market closes. SuperTrend (a per-trade ratcheting hl2±m·ATR band with close-based exit) won decisively: positive IS+OOS at both H4 (+9.7/+17.5 p/trade) and H1 (+9.9/+7.4), the only scheme positive at both, beating PSAR by ~20p/trade; PSAR was among the worst (−3.3 H4 / −3.0 H1, biggest drawdown) because its acceleration factor rushes the stop into noise. Since the raw entry was ≈breakeven, the exit alone produced the edge. The actionable follow-up — an A/B swapping SuperTrend for PSAR on 010's real SMA-stack entry — came out a tie and PSAR was kept: the ~20p gap came from the breakout trend entry's let-winners-run profile, another instance of validating exits only against the exact live entry.

Key result

SuperTrend IS/OOS: H4 +9.7/+17.5, H1 +9.9/+7.4 p/trade (only scheme positive at both TFs), ~+20p/trade over PSAR; PSAR −3.3 (H4) / −3.0 (H1) with biggest DD; giveback +7.9 H4 but IS-negative H1; raw entry ≈ breakeven; 010 A/B = tie, kept PSAR

Indicators

SuperTrend (ATR band ratchet)PSARDonchian channel (20)ATRchandelier stopMFE giveback

Algorithms

controlled exit bake-off (grid over exit families)IS/OOS validationper-trade anchored trailing-stop simulation

Code

research/experiments/trailing_bakeoff/trailing_bakeoff.py research/experiments/sma_exits/supertrend_ab.py

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

← #69 ESCMA exit-learner (shock-entry + CMA-NN exit)#71 Regime-conditioned entry Phase A (triple-barrier, against-drift) →

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