GUIDE · TESTING & COSTS
Why most gold scalpers lose: the spread maths
Gold scalping robots look incredible in screenshots. We built and tested several ourselves, with real tick data, and every one of them lost money. Here is the simple maths behind it, and what to look for instead.
Every trade starts at a loss
When you open a gold trade you buy at the higher ask price and close at the lower bid. That gap, the spread, plus a little slippage, is paid on every single trade. On XAUUSD it is typically about 25–40 points ($0.25–0.40 per ounce) on a standard account, and often far more in the seconds around big news.
A 0.01 lot is 1 ounce, so the cost is about $0.35 per trade. Small, but scalpers make hundreds of trades.
| Trades per month | Cost per year at 0.01 lot | % of a $1,000 account |
|---|---|---|
| 9 (our Breakout EA) | $38 | 3.8% |
| 21 (one a day) | $88 | 8.8% |
| 200 | $840 | 84% |
| 600 | $2,520 | 252% |
The win rate a scalper really needs
Take a typical scalp: target $2, stop $2. It looks like a fair 1:1 bet. With a $0.35 cost, a win pays +$1.65 and a loss costs −$2.35.
To break even you now need 59% winners, not 50%. Just to stand still.
The smaller the target, the worse it gets: with a $1 target and $1 stop you need about 68% winners. That's why scalpers chase tiny targets with wide stops, and why one bad trade wipes out a week.
What happened when we tested scalpers
Before building our current bots we tested the strategies most gold robots sell, using MetaTrader 5's "Every tick based on real ticks" mode:
- Trend-pullback scalper (Dec 2025 – Oct 2026): profit factor 0.98 on M1, 0.89 on M5, 0.73 on M15. All losing.
- Machine learning with 145 indicators on M5: profit factor 0.96 on unseen data. Losing.
- Every 5-minute candle, 2011–2026: trading in either direction was roughly break-even before costs, and after costs buys lost money in 16 of 16 years, sells in 15 of 16.
The patterns weren't the problem. At this speed there's very little edge in gold, and the spread takes what little there is.
The tester trick: +$16,900 vs −$995
We also rebuilt a popular style of momentum scalper that places buy-stop and sell-stop orders around price. Same rules, same months, two test modes:
| Test mode | Result | Profit factor |
|---|---|---|
| Tester default (generated ticks, low spread) | +$16,900 | 1.90 |
| Real tick data with real spreads | −$995 | 0.94 |
The only difference was the cost per trade. Many sales pages show the first kind of test. Always ask which modelling mode and spread were used. See how to backtest a gold EA honestly.
What works better on gold
If costs are fixed per trade, the answer is to trade less often, when the expected move is much bigger than the cost:
- Big US data surprises, when gold often moves tens of dollars in an hour. That's what Catalyst EA trades, about 2–3 times a month.
- Daily breakouts, when gold picks a direction and runs for hours. That's Breakout EA, about 9 times a month.
At those speeds the $0.35 cost is a few percent of a typical move instead of a third of it.
Checklist before you buy a gold scalper
- Is the backtest done with real ticks and a realistic spread (25–40 points), not a fixed 5–10?
- Multiply its trades per month by $0.35 per 0.01 lot. Is the profit much bigger than that?
- Is there a live signal running for months, not just a backtest?
- Does it use a stop loss on every trade, with no grid or martingale? See martingale & grid explained.