What a backtest really is
A backtest replays an EA's rules over historical prices and adds up the trades. It is useful: it shows whether an idea ever worked and how big the losing streaks were. It is also easy to get wrong, and almost every error makes the result look better than reality.
Ten reasons live results differ
| Cause | What goes wrong | Effect |
|---|---|---|
| Spread and commission | Tests use a fixed or average spread; live spreads widen at news and rollover | Profit per trade shrinks, scalpers suffer most |
| Slippage | In a test every order fills at the exact price | Live fills are worse, especially in fast markets |
| Low-quality data | Gaps and interpolated ticks give trades that never existed | Phantom profits |
| Overfitting | Settings tuned until the past looks perfect | Fits noise, fails on new data |
| Look-ahead errors | A rule accidentally uses information from the future | Impossible, perfect-looking results |
| Execution delay | Live orders take time and the price moves | Entries and exits land worse |
| Broker differences | Another broker, different prices, spreads and stop rules | A strategy can work on one and not another |
| Market change | The regime the EA was tuned for ends | Edge fades |
| Position sizing | The test used different lot sizing from your account | Drawdowns are not comparable |
| Short samples | A few dozen trades cannot show the true win rate | Luck looks like skill |
How to check an EA properly
- Demand a live record on a third-party tracker such as Myfxbook, on a real account, with the broker named. We publish ours: see our live account.
- Compare like with like. Same broker, same account type, same symbol. Our EA is supported only on Fusion Markets and IC Markets Raw accounts for this reason.
- Run it on a demo account first for at least a month and compare trade by trade with the back-test over the same dates.
- Check the trade count. A back-test of 30 trades says almost nothing; look for hundreds.
- Look at drawdown, not only profit. Ask how deep it went and how long it lasted, in testing and live.
- Look for a stable curve across years, not one lucky period.
- Ask what happens when it stops working. There should be a rule for that.
- Start small. Use the smallest plan or lot size until live results match expectations.
Why we show everything in public
We trade gold with our own EA, publish every trade on the website, and stream it live. Our back-test is on the home page, labelled as a simulation. When you read any EA seller's numbers, ask the same question we ask of ours: is it something I can check without trusting the seller?
Frequently asked questions
Why is my EA live result worse than the backtest?
Usually a mix of wider spreads, slippage, execution delay, imperfect test data and settings tuned too closely to the past. Compare trade by trade on the same broker to find which.
How long should I forward test an EA?
At least a month on a demo account, and ideally until it has produced a few dozen trades, so you can compare with the back-test over the same dates.
What is overfitting?
Tuning an EA's settings until it looks perfect on past data. It then fits the past's random noise and tends to fail on new prices.
Is a backtest useless?
No. It shows whether an idea ever worked and how bad the losing streaks were. It is just not enough on its own.