How to backtest an EA properly: a 7-step checklist before real money

Beginners backtest to prove the bot will win. Experienced traders backtest to find how the bot will die. That mindset difference decides whether you keep your account. Here is a 7-step checklist for any EA before real money.

1. At least one year of data, right timeframe, right symbol

Run Strategy Tester on 12+ months of the exact pair and timeframe the bot was designed for. Testing an M15 bot on H1 "Open prices only" is meaningless — use "Every tick" (or real-tick mode if your broker offers it).

2. Read drawdown before profit

The first number to check is not Total Net Profit — it is Maximal Drawdown. Rule of thumb: future live drawdown tends to be 1.5–2× the backtest. A 30% backtest dip means prepare for 50%+. Can you sit through that?

3. Study the losing streaks

Open the trade history and find the longest losing streak. Eight consecutive losses in a backtest will happen live — usually right after you increase size. If seeing that streak tells you that you would switch the bot off halfway, this strategy is not for you, even if the total is positive.

4. Stress-test spread and slippage

Re-run with spread at 1.5× the average (gold: try 40–50 points). Short-TP scalping strategies usually die right here — better to learn that now than with real money.

5. Out-of-sample test

If you tuned parameters on 2024 data, run unchanged on 2025. A mild degradation is normal; flipping to a loss means you curve-fit — the parameters were only ever beautiful in hindsight.

6. Two to four weeks on demo, compared to backtest

Run demo and compare live fills against the backtest over the same period. Missing trades or entries far from expected prices → tick-data or running-candle logic issues — fix those before thinking about real funds.

7. Go live at "mildly painful" size

Start at 0.5–1% risk per trade or the minimum lot. The goal of the first month or two is not profit — it is confirming the bot behaves as expected at the lowest possible tuition.

---

Bottom line: a pretty backtest is necessary, not sufficient. A bot worth running is one where you understand how it wins, how it loses, and what its worst streak looks like — before it happens.