Can Expert Advisor backtests be manipulated? How fake results are made and how to spot them

What a backtest actually is

A backtest replays historical price data and applies a strategy's rules to it, recording the trades that would have occurred. The output depends entirely on the quality of the data, the assumptions about costs, and the inputs chosen. Change any of those and the same EA can produce a very different report.

That flexibility is the reason backtests are useful for research, and also the reason they are easy to misuse in marketing. Every number in a report is the product of settings that someone chose.

Curve fitting and over-optimisation

The most common distortion is not fraud at all. If you run thousands of input combinations and keep the best one, you have simply found the settings that fit that particular stretch of history. The curve looks beautiful because it was selected for looking beautiful. On new data it usually falls apart.

Signs of curve fitting include a tiny change in one input turning a great result into a poor one, a profit that depends on a handful of trades, and a strategy that works on only one symbol or period. A robust edge tends to survive small changes to the inputs and a change of period.

Unrealistic spread, commission and slippage

Backtests are often run with a low fixed spread, no commission and no slippage. For a scalping or high-frequency EA these omissions can be the entire profit. A system that earns a fraction of a pip per trade before costs can show a steady rise in a test and lose money live.

A fair test uses a variable or realistic spread, includes the commission of the intended account type, and accounts for slippage. If a report does not state its spread and commission assumptions, assume they were flattering.

Poor or modified tick data

Tests that use modelled ticks interpolate price movement inside each candle, which can hide how a strategy would really have traded. Using real ticks from a broker gives a more faithful result. Some third-party data sources are cleaned, which removes the gaps, spikes and bad ticks that cause losses in real life.

A test can also use a data set from a broker with unusually favourable pricing. The result is real for that data and misleading for every other broker. Always check the data source, the modelling quality and whether it matches the broker you will use.

Cherry-picked periods, symbols and screenshots

A seller can show only the best symbol, the best year, or the best of many runs. A three-month window selected after the fact can look excellent for almost any strategy. Screenshots can also be edited or taken from a different test than the one described.

Ask for a long period that includes trending, ranging and high-volatility phases, and for results across several symbols and several brokers. A strategy that only works in one narrow window deserves scepticism.

Look-ahead bias and hidden errors

Some backtests contain a coding mistake, intentional or not, that lets the strategy use information it would not have had at the time, such as reading the close of a candle that has not finished. The test then shows impossible accuracy. Real trading cannot reproduce it.

Warning signs are an unrealistically smooth equity curve, an extremely high win rate with tiny losses, and results that collapse the moment the EA runs on a live or demo feed.

How to check results yourself

The strongest defence is to test independently. Run your own backtest with real ticks, your broker's data and realistic costs, then forward test on a demo account for several weeks on the same set-up you will use live.

• Compare your own test with the published one and ask why any differences exist.
• Stress the inputs slightly and see whether performance holds.
• Check the number of trades and the longest losing streak, not only the profit.
• Prefer a verified live track record from an independent tracking service over screenshots.

What a trustworthy result looks like

A believable result states its data source, spread, commission and period, shows drawdown and losing streaks honestly, works across a range of settings, and is supported by a live or forward track record. No result, however well documented, guarantees future performance. Treat any backtest as a hypothesis to be confirmed, not as evidence of future profit.