การตรวจสอบ 6 ข้อ
A backtest result must clear all six checks before we cite it. Checks are applied in order; failure at any stage ends evaluation.
- 01 ความโปร่งใสของซอร์สโค้ด
We review the strategy logic and its settings for hidden martingale logic, undisclosed averaging-down, or hard-coded broker parameters. We only cite results whose logic we can read in full.
- 02 ประวัติขั้นต่ำ 5 ปี
Backtest window covers at least 5 years, ensuring exposure to both the 2022 risk-off regime and the 2023–2024 recovery.
- 03 คุณภาพการสร้างแบบจำลอง 92–100%
Every-tick backtests at 92–100% modelling quality. Lower quality settings are not accepted.
- 04 การทดสอบซ้ำนอกกลุ่มตัวอย่างหลังล็อก
Settings are frozen at a cutoff date. Every day after that cutoff is re-tested on market data released later — data the strategy was never fitted to — and the result is kept as a separate window, never merged into the backtest figures. This is a rolling out-of-sample re-test, not a live account: the tester does not simulate slippage, requotes or rejected orders.
- 05 บันทึกความเข้ากันได้ของสเปรด
How the result changes at different brokers' typical spreads is documented wherever it is cited. Multi-broker live/forward verification is not part of this method, and we do not present it as one.
- 06 Drawdown ภายใน 20%
Max historical drawdown ≤ 20%. Where we hold the tester's equity drawdown — the deepest the account sat with open positions included — that is the figure we judge, not the shallower balance one. Exceptions are flagged with an explicit warning wherever the result is cited.
กระบวนการ backtest
Five sequential steps from raw data to the figures we cite in our research.
- 01 Symbol & history selection
We run on real tick data. Which feed depends on the test — a broker's own tick history, or Dukascopy ticks loaded as a custom symbol — and the exact source, symbol and contract specification are named wherever the result is cited. History length is a minimum of 5 calendar years, capturing the 2022 Ukraine/USD risk-off shock and the 2023 banking stress.
- 02 Modelling quality
All backtests run on the every-tick model at 92–100% modelling quality (tick-level simulation). We do not publish results from open/close or 1-minute OHLC interpolation, which can produce systematically optimistic drawdown figures for scalping EAs.
- 03 Realistic spread & commission
We apply the broker's actual floating spread observed during the backtest period where available, plus realistic commission per lot. Variable spread is not collapsed to a fixed average.
- 04 Full-window test & in-sample disclosure
We do not run a separate held-out walk-forward optimisation. Instead, each strategy is tested across the full 2021–2026 window — which spans the 2022 risk-off shock rather than a flattering sub-period — and any parameter chosen by looking at the test window (an in-sample decision) is disclosed plainly wherever the result is cited.
- 05 Spread sensitivity
Performance is measured on Exness real retail-spread tick data, and how the result changes at other brokers' typical spreads is noted wherever it is cited. Live, multi-broker forward verification is not part of this method, and we do not present it as one.
ข้อมูลอ้างอิง
นิยามตัวชี้วัด
Every figure we cite from these backtests — in guides, research and glossary examples — uses these exact definitions. No metric is renamed or recomputed differently between pages.
| Metric | Definition | Why it matters |
|---|---|---|
| Max Drawdown | Peak-to-trough equity decline during the backtest period, expressed as % of peak equity. | Primary measure of catastrophic risk — governs position sizing. |
| Worst Streak | Worst uninterrupted sequence of net-negative trades (not individual losing trades). | Stress-tests the psychological durability of a live operator running the EA. |
| Recovery Days | Calendar days from max-drawdown trough to new equity high. | Distinguishes fast-recovery profiles from 'stuck under water' patterns. |
| 12-Month Return | Most recent 12 months of backtest performance, not annualised from a longer period. | Recent regime relevance — markets change; a 2017 EA may be misadapted for 2024. |
| CAGR 5Y | Compound annual growth rate over the full 5-year window. | Long-term baseline normalises for lucky short windows. |
| Sharpe Ratio | Mean daily return divided by std deviation of daily returns, annualised. Risk-free rate = 0 (conservative). | Single number that captures return-per-unit-of-volatility. |
| Sortino Ratio | Like Sharpe but divides by downside deviation only (ignores upside volatility). | More relevant than Sharpe for EAs with asymmetric upside. |
| Win Rate | Percent of closed trades with positive net P&L. | Context only — a 40% win rate with a 3:1 R:R is fine; we always show alongside average R:R. |
ข้อจำกัดของขอบเขต
We document what our methodology does not cover so readers can calibrate confidence correctly:
- Live forward test results are not part of the data. Backtest data is the evidence base.
- Black-box EAs without source access are not reviewed. We cannot verify the absence of hidden logic.
- Crypto CFDs and metals are outside the scope of this data. Coverage is limited to major and minor Forex pairs on MT5.
- Optimised parameters are tested for robustness but not guaranteed future-proof. Market regimes shift; re-test any EA — including one you build — before relying on parameters chosen more than 12 months ago.
Questions about methodology specifics can be sent via the contact form.
นโยบายบรรณาธิการ