GUIDE · TESTING & COSTS
How to backtest a gold EA honestly
A backtest can make almost any robot look profitable. These are the eight checks we use on every strategy, the same ones that rejected over 40 ideas before two survived.
1. Use real ticks, not generated ones
In the MT5 Strategy Tester choose Modelling: Every tick based on real ticks. The faster modes generate artificial prices between candles and often assume a tiny spread. For a short-term gold strategy that alone can turn a loser into a winner (our momentum test went from +$16,900 to −$995).
Also set Delays: Random delay, so orders don't fill at perfect prices.
2. Test again at double the costs
Live spreads are wider than history suggests, especially around news. Re-run the test with costs doubled. A real edge survives; a fragile one collapses. Our Breakout EA kept a profit factor of 1.25 at double cost; most scalpers drop below 1.
3. Write the rules down before you look
Decide the rules, the settings to try and what counts as a pass before running the test. Otherwise you'll keep tweaking until the past looks perfect, and the future won't care. We save each plan to a file before every test.
4. Choose on old data, check once on new data
Split history in two. Pick the best settings on the older part (we use 2012–2021), then run them once on the newer part (2022–2026). If you go back and re-tune after seeing the new data, it is no longer a fair check.
One of our ideas, a failed-breakout reversal, made a profit factor of 1.50 over 2012–2021. On 2022–2026 it was 0.75. A seller who only showed the first period would have launched it.
5. Remove the five best trades
Delete the five biggest winners and recalculate. If the profit disappears, the strategy depends on luck. Our Catalyst research stayed profitable without its top five trades in every period.
6. Look year by year and side by side
Check every year, and buys versus sells separately. A strategy that made all its money in one year, or only on buys during a gold bull run, isn't robust. Expect some losing years: our Breakout research had several, which is exactly why we show them.
7. Try the same rules on other markets
Run the identical rules on markets they weren't designed for. If they 'work' everywhere, be suspicious of the test. If they work only where there's a reason, that's a good sign. Our breakout rules passed on gold and failed on 14 other markets.
8. Run a demo with the rules frozen
Finally run the bot on a demo account for weeks or months without changing anything, and compare each live trade with what the backtest says should have happened. Our own demo started in November 2026, with results published in the research log.
Red flags in other people's backtests
- "99.9% quality" with a fixed, very low spread
- Only the last few months tested
- A perfectly straight equity line: often a grid or martingale hiding losses
- Huge win rates (90%+) with tiny wins and no visible stop loss
- No live signal, or a signal much shorter than the backtest