What a real gold EA backtest should include

A backtest is not a track record. It is the seller's own simulation, so it can only rule an EA out, never prove it works. For live results, see how to verify a gold EA's track record. For why good tests still fail on a real account, see why gold EAs fail live. This page is about what a test must contain before it is worth reading at all.

What makes a gold EA backtest trustworthy?

Realistic data, full trading costs, a long period with different gold conditions, results on data the EA was never tuned on, settings that still work when nudged, honest drawdown and enough trades. A backtest missing any of these can look far better than the EA really is.

Ask for the full report, not a cropped equity curve. In MT5 the Strategy Tester report shows the modelling mode, the period, the deposit, the costs and the trade list. If a seller won't share it, you have learned something.

What data and modelling quality should the backtest use?

Real tick data or the most detailed modelling your platform offers, from a broker feed close to the one you'll trade on. Testing on bar open prices or coarse bars can hide what happens inside each bar, which matters for tight stops and short holding times.

In MT5, "Every tick based on real ticks" is the most detailed mode. Coarser modes guess the path the price took inside a bar. For an EA that holds trades for hours, the guess may not matter much. For one that scalps a few points, it can decide whether a trade hit its stop or its target first. Also check which broker's data was used. Gold prices, spreads and session times differ between brokers, so a test on one feed can mislead on another.

Which trading costs must be in the backtest?

Spread, commission, swap and slippage. Gold spreads widen at quiet hours and around news, commission is charged per lot, swap is charged on overnight positions and slippage hits fast fills. A test with a fixed tight spread and no commission flatters the result.

Check each one in the report:

• Spread: variable from real data, or a fixed value that is at least as wide as your broker's typical spread.
• Commission: set to your account type's rate, per lot, both sides.
• Swap: applied to any trade held past the daily rollover.
• Slippage: some allowance for fills worse than the quoted price, especially on stops and news.

A good habit is to rerun the test with costs made harsher. If the result collapses, the edge was thinner than the costs.

How long should a gold EA backtest be?

Long enough to include different kinds of gold market: strong trends, sideways stretches, quiet periods and sharp volatile ones. A test covering a single calm or single trending stretch only tells you how the EA does in that one condition.

Gold behaves differently from one stretch of years to the next. A trend-following EA tested only on a strong trend will look brilliant, and a range-trading EA tested only on a quiet range will look brilliant too. Ask which years are covered and why the test starts where it does. A start date chosen just after a bad stretch is a common way to clean up a curve.

What are out-of-sample and walk-forward testing?

Out-of-sample means checking the EA on data it was never tuned on. Walk-forward repeats this in steps: tune on one window, test on the next, then move forward. Only the results on unseen data say much about the future.

Any EA can be tuned until it fits the past. The more settings it has, the easier that is. So ask which part of the period was used to choose the settings, and which part was kept back. If the answer is "all of it", the whole test is in-sample and shows only how well the settings fit history.

Do nearby settings still work?

A robust EA keeps working when its settings are nudged a little. If one exact value works and its neighbours lose, the result is probably a fit to noise. Ask to see results for a range of settings, not just the best one.

How do I read drawdown and trade count in a backtest?

Read the maximum drawdown on equity, not just balance, and how long it took to recover. Check the number of trades too. A handful of trades, however good, can be luck. More trades across more years make the result harder to fake.

Equity drawdown includes open losses, which matters for any EA that holds losing trades. If balance drawdown and equity drawdown are far apart, read our guide on how to spot a martingale gold EA. Also look at how long the worst drawdown lasted. A drop you could survive on paper may be one you would quit during in real life.

Quick checklist: a gold EA backtest worth reading

• Full MT5 report, not a screenshot.
• Real tick data, or the most detailed mode available.
• Spread, commission, swap and slippage all charged.
• Years of data with trending, ranging, quiet and volatile gold.
• Clear split between tuning data and unseen data.
• Walk-forward or out-of-sample results shown.
• Nearby settings also tested.
• Equity drawdown and recovery time reported.
• Enough trades spread across the whole period.

How BaroqTech tests its own ideas

We run every idea through the same kind of checks: full costs, data the idea never saw, and tests for luck and consistency. Every result, pass or fail, goes on the record.

No strategy has passed yet; every result is public at baroqtech.com/record.

The Monthly Pass and the Lifetime Pass are software licences for your own MetaTrader account. They score every robot and strategy you run the same way, with costs inside every number. See the plans.

FAQ

An educational tool. It is not investment advice. Trading leveraged products such as gold CFDs carries a high risk of losing money, and you can lose all of it.

BaroqTech makes software. It is not investment advice and it does not manage money. Trading leveraged products such as gold CFDs carries a high risk of losing money, and you can lose all of it. Results from a test or from a demo account do not predict results on a real account.