Direct answer
EA backtesting is the process of running an expert advisor (EA) through historical market data to estimate how it might have performed in the past. The main limitation is uncertainty: what worked historically may fail later because future prices, liquidity, volatility, and trading frictions can differ, and because the test setup may not reflect real execution.
Mechanics and what gets assumed
Backtesting typically takes (1) an EA’s rules, (2) historical price data, and (3) a set of modeling assumptions about how trades would be filled. The assumptions can include how spreads and commissions behave, whether orders are filled at the shown price or after delays, and how slippage is handled. Even if the EA logic is stable, the test is only as realistic as the test environment.
A useful way to think about it is separation of stable mechanics from variable conditions:
- Stable mechanics: the EA’s decision rules and order-generation logic.
- Variable conditions: market regimes, liquidity, bid-ask spreads, execution speed, and the trading costs actually incurred.
If the backtest treats variable conditions too ideally (for example, using the same spreads as the chart shows, or ignoring slippage), the simulated outcomes can be systematically different from what would happen when the EA trades with real constraints.
Evidence or example of why it can fail
Consider a backtest that shows strong returns using historical bars. In live trading, at least three categories of differences can appear:
- Data differences: the historical feed may not represent the exact price path that existed tick-by-tick.
- Execution differences: fills may be worse due to slippage, partial fills, or latency.
- Environment differences: the EA may rely on volatility or trend conditions that later shift.
Even without any “wrong” coding, these differences can change trade timing and the sequence of entries/exits, which matters a lot for strategies sensitive to order placement and market microstructure.
Limitations and risks
Material failure modes include:
- Overfitting to history: an EA can appear optimized to past patterns, while future markets follow different dynamics.
- Look-ahead bias and hidden assumption leaks: test settings may accidentally use information the EA would not have had at the time.
- Survivorship and selection bias: if the tested period was chosen because it looked favorable, results may not generalize.
- Cost and execution underestimation: small differences in commissions, spreads, and slippage can materially affect profitability, especially for strategies with frequent trading.
A key limitation is the mismatch between “historical relationships” and “future results.” Historical performance can reflect many transient conditions; it is not evidence that those conditions will persist.
Verification or next question
To independently verify what a backtest can and cannot tell you, focus on whether the test setup matches realistic trading constraints: the data quality, the treatment of spreads and commissions, and how slippage and order-fill timing are modeled. If an EA’s performance depends heavily on highly specific assumptions, then the backtest is less useful as evidence.
A good next question is: “Which inputs and execution assumptions would most change the trade outcomes, and how sensitive are results to reasonable variations of those assumptions?”