Backtest vs live results: why your EA trades differently live
Live results almost always trail a backtest. The backtest fills orders at the price it asked for, often with simplified ticks and no delay, and its settings were usually chosen on the same data it is scored on. A small gap is normal; a live drawdown far deeper than the backtest’s, or a profit factor below 1, means the test was too optimistic or the EA is running differently.
This guide explains the gap between backtest vs live results, how big a gap is normal, how to make an MT5 backtest closer to live trading, and how to compare a live account with its backtest trade by trade.
Why live results trail the backtest
| Cause | In the backtest | Live | How to narrow the gap |
|---|---|---|---|
| Price data | Ticks can be generated to fill in between bars | Every real tick, including spikes | Test with “Every tick based on real ticks” |
| Spread and commission | May not match your account | Your broker’s real costs | Test on your broker’s data and set its commission |
| Execution | “No Delay” fills every order at the requested price | Delays, requotes and slippage | Test with “Random Delay” |
| Fitted settings | Inputs optimised on the test period | New data the settings never saw | Keep a forward period and optimise few inputs |
| Account and settings | The set file and risk you tested | Whatever is loaded live | Check the set file, lot size, symbol name and account mode |
| Running conditions | Runs without a break | Restarts, disconnections, Algo Trading switched off | Run it on a VPS and read the journal |
| Market conditions | The years you tested | Whatever the market does next | Test several years with different conditions |
MetaQuotes describes “No Delay” as ideal conditions, “Random Delay” as the mode closest to real trading, and real-tick testing as close to real conditions. A backtest run with no delay and generated ticks is a best case.
How big a gap is normal?
There is no fixed ratio. Total profit is the noisiest number to compare, so start with the per-trade figures, which should be close if the EA runs as tested:
- trades per month
- win rate
- average win and average loss
- profit factor: gross profit divided by gross loss
Expect the live maximum drawdown to pass the backtest’s at some point: a long enough live record meets losing runs the test period didn’t contain. Be concerned when live trades per month are far from the backtest’s, the average loss is clearly larger, or the profit factor stays below 1 over a few hundred trades.
How to make an MT5 backtest closer to live
- Use real ticks. Choose “Every tick based on real ticks” on the broker and account type you will trade.
- Set the costs. Add your account’s commission per lot, and keep swaps on.
- Add delay. Test with Random Delay, or a fixed delay equal to your VPS’s ping to the broker.
- Keep a forward period. Hold back the latest part of the data and judge the settings on it. MetaQuotes notes that forward testing helps avoid fitting parameters to particular stretches of history.
- Optimise less. The more inputs and combinations you try, the more likely the best result is luck rather than an edge.
- Test at your risk. Run it at the lot size or risk you will actually use, over several years.
Our step-by-step guide to backtesting an EA in MT5 shows where each setting is, and the position size calculator turns your risk into a lot size.
How to compare a live account with its backtest
- Backtest the same dates as the live account, with the same set file, symbol and starting balance.
- Export both trade lists and line them up by open time.
- Missing live trades point to a filter, a disconnection or Algo Trading being off; extra live trades point to different settings or a second EA on the account.
- For trades that match, compare entry and exit prices: the difference is your real spread, slippage and delay.
- Re-run the backtest with those costs to see what the EA earns in live conditions.
To read the live side properly, see our guide to checking results on Myfxbook and FX Blue.
When the gap means a problem
- The backtest shows almost no losing months but the EA averages down or raises lot sizes after losses: see martingale vs grid EAs.
- A scalper’s backtest used “Open prices only” or “1 minute OHLC”, which can’t show what happens inside a bar.
- The seller shows backtests but no live account: one of our forex EA scam red flags.
- Live drawdown passes the backtest’s worst within the first few weeks.
Backtest vs live results FAQ
Why is my EA losing live when the backtest was profitable?
Usually because live costs and fills are worse than in the test, the settings were fitted to the test period, or the EA isn’t running as tested. Compare the two trade by trade to find which.
Which MT5 modelling mode is closest to live trading?
“Every tick based on real ticks”, which replays ticks recorded by your broker. Combine it with Random Delay and your real commission.
Is forward testing better than backtesting?
They answer different questions. A backtest shows how settings fit the past; a forward period or a demo run shows how they hold up on data they weren’t tuned on. Use both.
How long should I demo test an EA?
Long enough to see a losing run, and a few hundred trades if it trades often. For EAs that trade a few times a month, that means several months.
Can a backtest be trusted?
As a check of the logic and a rough guide to risk, yes, if it uses real ticks, real costs and a forward period. As a forecast of profit, no.
Risk warning. Trading forex and CFDs on margin carries a high risk of losing money. Past results, live or backtested, don’t guarantee future returns.
Sources, checked 6 November 2026: MetaTrader 5 Strategy Tester: modelling, delays and forward testing; MetaTrader 5 testing report.