How the MetaTrader Strategy Tester Works

The Strategy Tester takes an expert advisor, a symbol, a timeframe and a date range, and replays the historical prices of that range through the EA one event at a time. Every time the EA opens, modifies or closes an order, the tester fills it against the simulated price stream, tracks balance, equity and margin along the way, and compiles the whole run into a report.

The scale is easy to underestimate. Two years of EUR/USD on H1 is roughly 12,400 hourly bars — but in every-tick mode the tester expands those into tens of millions of simulated price updates. That is why the choice of modelling mode changes both how accurate a test is and how long it takes to run.

MT4 downloads bar history — typically 1-minute bars — from the broker’s history centre, and everything finer than a minute has to be generated. MT5 requests history from the trade server too, and where the broker publishes it, that includes real tick history: every recorded quote with its bid and ask.

• Data is broker-specific. Server timezone, spread model and tick filtering all shape the bars, so the “same” symbol differs slightly between brokers.
• Old history is often patchy — missing minutes or whole sessions — and the tester silently works with whatever it has.
• Tick history is large (often gigabytes per symbol-year) and not every broker provides it for every instrument or period.

A 1-minute bar stores only four prices: open, high, low and close. Unless real ticks are available, the tester has to invent the path between them. Each modelling mode makes a different trade-off between fidelity and speed:

The cost of interpolation is concentrated in intrabar logic. Generated ticks follow a predetermined pattern through the bar, so the tester is guessing whether the high or the low came first. An EA with a 5-pip stop and a 7-pip target inside a 15-pip M1 bar can be filled in either order depending on that guess — and the tighter the levels, the more often the guess decides the trade.

With generated ticks there is no recorded spread, so the tester applies a stand-in. MT4 uses a fixed spread in points (or a snapshot of the current one) across the entire history; MT5 with generated ticks uses the spread stored alongside each minute bar or a fixed custom value. Only real-tick testingreplays the bid/ask that actually existed — including the widening around news and rollover.

The shorter the average trade, the bigger this effect. A strategy targeting 60 pips per trade barely notices half a pip of extra spread; a 6-pip scalper lives or dies by it.

However the prices are modelled, the tester’s execution layer is friendlier than any live server:

• Market orders fill instantlyat the current modelled price — no latency between decision and fill.
• There are no requotes or rejections; every order the EA sends is accepted.
• Stop and limit orders trigger exactly at their level, with no slippage through fast markets.
• Liquidity is unlimited — a 0.1-lot and a 50-lot order fill at the same price.

These simplifications are individually small and collectively significant. If live execution costs an average of 0.2 pips of slippage per side, that is 0.4 pips per round trip — 8% of the average win for a strategy that makes 5 pips per winning trade, and effectively zero in the backtest.

MT4 prints a modelling quality percentage, capped at 90% because it never has real ticks: 90% means complete M1 data covered the range and ticks were generated from it. It also counts mismatched chart errors— inconsistencies between the timeframes used to build the test. MT5 reports a history quality figure with the bar and tick counts behind the run.

The tester can also run the same EA hundreds or thousands of times with different parameter values — exhaustively or with a genetic algorithm, and in MT5 across multiple CPU cores or remote agents. Three parameters with 20 values each is already 8,000 candidate runs.

The trap is built into the method: the best combination is the best on that one sample of history, by construction, and part of its edge is fitted to noise that will not repeat. MT5’s forward-period option, which re-checks the top results on data the optimizer never saw, is one defence; the guide to more reliable backtests covers this workflow in detail.

Put the pieces together and the gap between a report and a live account stops being mysterious: modelled ticks instead of real ones, one spread instead of a changing one, perfect fills instead of real execution, and parameters tuned to the tested period. How each of those gaps biases results — and how to spot them in a report — is the subject of why backtests can be misleading.

A practical first step is reading your own reports more carefully: the free MetaTrader Backtest Analyzer parses an MT4/MT5 tester report in your browser and surfaces the statistics — drawdown, streaks, distribution — that a headline profit number hides.