Strategy research, with real backtests
Most strategy write-ups on the internet show you a curve and skip the numbers. We have our own backtest engine, so every article here runs the strategy across multiple markets and configurations on years of minute-level data — and we publish whatever comes out. When a popular strategy does not work, that is the article.
We backtested the classic RSI 30/70 mean-reversion Expert Advisor (EA) across 8 markets on H1 and H4. Seven of eight lost money on H1; BTCUSD gave back 40.82%. Full numbers, plus whether a trend filter or tighter levels actually help.
2026-08-10Supertrend EAWe backtested Supertrend(10,3) as an Expert Advisor (EA) across 8 markets on H1 and H4. Six of eight lost on H1; the H4 winners mostly failed out of sample. Plus why the popular ADX filter made every single result worse.
2026-08-10Multi-strategy portfolioWe ran our 10 template strategies together on one account. Average pairwise correlation of daily returns: 0.014. At the same drawdown as the average single strategy, the portfolio returned +272.59% versus +28.72%. Full method, leave-one-out test, and where they still fail together.
2026-08-13Adding losing strategiesWe backtested 322 strategies with no filtering, then picked winners on one year and measured them on the next. Only 42% were still profitable; in-sample and out-of-sample Sharpe correlated -0.034. What did persist: correlation (+0.707) and drawdown, not return.
2026-08-17Grid / martingale EAWe ran the standard retail grid EA over 55 months of M1 data on six markets. 67,633 baskets closed, every one at a profit — and 33 of 54 accounts ended dead. 0 of 9 configurations survived on gold or Bitcoin. Full method and data.
2026-08-15MA crossover EAWe backtested the classic EMA 50/200 crossover Expert Advisor (EA) across 8 markets and 13 configurations. Nothing graded above C; EURUSD H1 lost 19.83%. Full results, plus whether an ADX filter or faster averages actually help.
2026-08-05How we run these
- Same rules across every market — no per-symbol curve fitting.
- Minute-level stop and target checks, with an out-of-sample segment scored separately.
- Grades and every underlying number are shown, negative results included.
- Snapshots are dated: the backtest window rolls forward, so figures move over time.