Why Live EA Results Differ From Backtests

A backtest and a live account run the same code through entirely different machinery. The MetaTrader Strategy Tester replays stored price history: every order fills instantly, at the recorded price, with whatever spread the model assumes. A live account sends each order across a network to a broker server, where it fills against the liquidity that actually exists at that moment.

Each section below is one place where those two paths separate. Individually the differences look small; across hundreds of trades they compound into the gap most EA users eventually notice.

The tester’s biggest simplification is cost. Live spreads breathe: EUR/USD might average 1.0–1.5 pips through the London session, widen to 3–4 pips around rollover, and jump far higher into news — while many backtests apply a single fixed spread to years of trades. Slippage is the second cost: a market order fills at the price available when it reaches the server, not the price the EA saw when it decided. On instant-execution accounts the same gap surfaces as a requote instead — a rejection the tester never issues.

Latency sits underneath both effects: the longer an order travels, the further price can drift before the fill. How slippage arises is covered in the slippage guide, and the network side in the latency guide.

A backtest runs on stored history — downloaded bars or, at best, the broker’s recorded tick archive. Your live account runs on the feed streaming right now, filtered and aggregated by your broker in its own way. Even the most realistic tick mode replays one broker’s past feed, which is not guaranteed to match the ticks your terminal receives this week: highs, lows and tick counts all differ slightly, and an EA that triggers on exact price touches will fire on different bars.

Server time is the quiet one. Daily candles open and close on the broker’s server clock, and many EAs carry time filters such as “trade only 09:00–18:00”. Backtest on data stamped in one timezone, then run live on a server using another — or one that handles daylight saving differently — and the EA is literally trading different hours and reading differently shaped daily bars.

Around scheduled releases — central bank decisions, inflation prints, employment data — liquidity thins out exactly when price moves fastest. Spreads that normally sit near 1 pip can widen to 5–10 pips for a few minutes, and live fills can land several pips from the requested price. The tester replays those same minutes with its usual assumptions: recorded prices, modelled spread, instant fills.

The result is asymmetric. An EA that happens to trade through news collects its most generous fills in the backtest at precisely the moments live execution is worst. Breakout entries and tight stops placed around volatility are the most exposed.

A simulation has perfect uptime. A live setup does not: the VPS reboots for an update, the terminal loses its connection, the broker server has a maintenance window, or ticks simply stop arriving for a stretch. Each interruption edits the live track record in a way no simulation reproduces:

• A missed entry removes a trade the backtest counts — the two trade lists stop matching from that point on.
• A missed exit is worse: the EA holds a position it would have closed, turning a few offline minutes into a materially different outcome.
• Even a few minutes of downtime per week is enough to make live results diverge trade-by-trade from any test run.

Run the identical EA, settings and period on accounts at two different brokers and the results will not match — sometimes dramatically. Each broker assembles its own price feed, applies its own spread or commission model, runs its own server clock and enforces its own execution rules.

None of this is a malfunction — the EA is responding to genuinely different inputs. It is also why a results screenshot from someone else’s account, even running the same EA, says little about what yours will record.

Every backtest ends where its data ends, and live trading begins exactly there — in conditions the test never contained. Volatility regimes shift, ranging pairs start trending, spread structures change. An EA whose parameters were tuned on the test window, deliberately or just by keeping “what worked”, is to some degree fitted to a market that no longer exists.

This source of divergence has nothing to do with execution: even with perfect fills and identical data, a strategy can simply meet a different market. It is the main reason some degradation between backtest and live performance is the base case to plan for, not a surprise to investigate.

A fair comparison removes every mismatch you can control, so that whatever remains is real divergence:

• Re-run the backtest over the exact live window — same symbol, timeframe, settings and deposit — and compare against that, not the original multi-year test.
• Give every EA a unique magic number and filter the live history by it, so one strategy's results are not blurred by anything else on the account.
• Compare distributions, not headlines: trade count, win rate, average win and loss, profit factor. A different trade count means missed or extra trades, not just different outcomes.
• Measure the friction directly: average spread paid and slippage per trade on the live account versus what the backtest assumed.
• Decide in advance how much degradation is acceptable — for example, a review trigger if live expectancy falls below half of the backtested figure.

Both halves of the comparison come from your own records: the free Backtest Analyzer reads the Strategy Tester report, and the live side comes from your own MetaTrader account history, filtered per magic number over the matching window. When the same metrics sit side by side, the question changes from “why is live worse?” to the answerable one: which specific difference — cost, data, downtime or market — accounts for the gap.