The textbook trendline-break setup, quantified and tested
Trendline breakouts followed by a retest and the subsequent "first wave" move initially seemed like a solid strategy.
Breakout entry example (XAUUSD daily, real data): buy when price breaks above the recent high.
Trendline breakouts followed by a retest and the subsequent “first wave” move initially seemed like a solid strategy. My hypothesis was that by connecting the last two swings to form a trendline, I could capture the momentum after the price returns to that line. I tested four variants: Trend-Following and Trend-Reversal, both in long-only and hedged configurations. I used a three-stage state machine to track the breakout, the retest, and the entry. To avoid look-ahead bias, I used only confirmed fractal points for the lines, verifying the data against synthetic prefixes.
The Initial Promise
The “Trend-Following/Hedged” variant actually cleared my first two rigorous gates. It showed a total return of +44%, was profitable in 5 out of 6 years, and had a worst-day loss of -3.0%. It also passed my H1 parameter robustness test, maintaining profitability across all 15 tested parameter combinations. Notably, its correlation with my core systems was only 0.62, suggesting it wasn’t just rediscovering the same old trends. The Trend-Reversal variant, however, vanished during walk-forward testing, confirming my previous findings that mean-reversion strategies often lack a lasting edge.
Where the Strategy Collapsed
Despite that early success, the system failed three critical later-stage stress tests:
- Lack of Timeframe (TF) Robustness: While it performed well on H1, it collapsed on higher timeframes. It returned -1.9% on H4 and -6.9% on D1, where it was profitable in 0 out of 6 years. This suggests the strategy was merely overfitting to H1 noise rather than capturing a genuine market trend.
- M1 Intraday Failure: When I reconstructed the account equity using 1-minute bars, the strategy fell apart. While the daily bar evaluation looked fine, the M1 data showed a -6.66% loss, with four separate days exceeding my -5% daily loss limit. The hedged positions were amplifying intraday correlation reversals; this is a classic example of why bar-based testing often hides risks.
- Over-reliance on Annual Re-selection: When I stopped re-selecting parameters annually and instead used a fixed selection from 2015 to 2019, the performance plummeted. The PF (profit factor) dropped to 1.04, with a drawdown of -24.9% and a 35.7% failure rate in maximum loss simulations. The system’s “success” was just chasing past winners rather than having a stable, inherent edge.
Final Verdict
I have decided not to deploy this system. Even after isolating a long-only version (which did improve the M1 daily loss from -6.66% to -4.44% and cleared the -5% daily limit breaches), the system remains fundamentally flawed. The D1 timeframe continued to fail, and the drawdown remained at -22.8%, which violates my -10% risk threshold. If I lower the risk to keep the drawdown within limits, the annual return shrinks to a negligible 1.3%. This process served as a useful stress test for my verification framework. It successfully caught a “false positive” that passed the early stages but lacked the structural integrity to survive deeper analysis. I am stopping development here to avoid data-dredging on a foundation that clearly lacks multi-timeframe robustness. The code for the trendline logic will be archived for potential use in future, unrelated hypotheses.
How this connects
This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).