Risk warning: software only, not investment advice. Trading involves substantial risk, including loss of capital. Results shown are simulated and not reliable indicators of future performance. Full disclosure
← All articles

Why Backtests Differ From Live Trading Results

9 November 2026 · 10 min read

Key takeaways

  • MT5's 'Every tick' mode synthesises ticks from M1 bars, so any EA sensitive to intra-bar price paths is tested against invented data.
  • 'Every tick based on real ticks' is only as accurate as your broker's tick history, which varies between brokers and is patchy before around 2017.
  • Fixed-spread backtests overstate profit on indices and metals; rerun on historical spread with real-tick data before trusting a result.
  • Latency barely affects H4 swing strategies but can consume a meaningful slice of every winner on 5-pip scalping targets.
  • Expect live drawdown to exceed the backtest figure: a 15% tested drawdown plausibly becomes 20-25% live once real spread, slippage and execution variance land.

Backtests diverge from live trading because MT5 simulates execution imperfectly: spread modeling, slippage, tick data quality, latency and broker execution policy each distort results differently depending on which backtest mode you run. An M15-H4 indices, FX or metals EA that shows a strong profit factor on 'Every tick' can lose most of that edge live. This article maps each divergence source to its backtest mode and gives you a diagnostic for your own strategy.

Why does my MT5 backtest profit disappear in live trading?

Your backtest profits vanish because the Strategy Tester simulates execution conditions that live trading replaces with messier ones. MetaQuotes' own documentation on Strategy Tester modeling modes shows the tester approximates ticks through several modeling levels, and each level trades accuracy for speed. A backtest fills every bar at the modeled price; a live broker widens spreads around news, requotes, and applies slippage your.ex5 never saw during testing.

Add curve-fitted parameters tuned to one historical window and the gap widens further. The result: a profit factor of 2.0 in the tester can degrade to breakeven or worse on a live account. That divergence is not bad luck; it is a predictable set of mechanisms, each diagnosable, which the rest of this article maps one by one. The same honesty is why every bot listed at BotGenomix publishes losing months and unseen-data results rather than tester screenshots.

How does tick data quality change backtest accuracy?

Where MT5's tick generator comes from

In "Every tick" mode, the Strategy Tester doesn't replay recorded ticks. It synthesises them from M1 bars using a built-in model that interpolates OHLC points and fills in plausible shapes. Your EA trades a simulation of a minute, not the minute itself, so any entry logic sensitive to intra-bar sequence is being tested against invented data.

Why broker tick history differs between brokers

"Every tick based on real ticks" pulls recorded ticks from your broker's server, and those ticks are broker-specific. One broker's feed may post 40 ticks per minute on EURUSD while another posts 8, with different spreads at the same timestamp. A backtest run on Broker A's history is not portable evidence for how the same EA behaves on Broker B's feed, especially on metals where spread variation between feeds is wider.

How missing ticks hide slippage and spread spikes

Thin or gappy tick history smooths out the worst moments: the spread blowouts around the NY open, the requotes during news, the seconds where price jumps 30 points with no fills in between. Your backtest fills at clean prices that never existed. On M15-H4 strategies the damage is smaller than on scalping, but it still inflates profit factor on fast entries.

This is why we publish month-by-month backtests including drawdowns and losing months at https://botgenomix.com/track-record, labelled simulated, rather than a single flattering equity curve.

What is the difference between 'Every tick' and 'Every tick based on real ticks' modeling?

'Every tick' is synthetic: MT5's tester generates ticks from M1 OHLC bars using an internal algorithm. 'Every tick based on real ticks' replays recorded ticks from your broker's servers, so each tick is one that actually traded. The second is closer to live, but only as close as the tick history your broker provides (MetaQuotes documentation on tick generation and real tick import).

How the tick generator interpolates between M1 bars

In synthetic mode, the tester only knows four prices per minute: open, high, low, close. It invents the path between them, and the path it invents is deterministic, not what the market did. On M15-H4 charts this rarely matters for direction, but it matters for any EA that triggers on intra-bar price movement, tight stops, or limit fills.

What real-tick mode does and does not fix

Every tickReal ticks
Tick pathInterpolatedRecorded
Spread per tickFixed or syntheticHistorical, varies
Data before 2017-ishFineOften patchy
Run timeFastSlow

It does not fix slippage, latency, or your broker's future spread behaviour.

When each mode is honest enough for M15-H4 strategies

On M15-H4 with stops wider than typical spread noise, synthetic mode is a reasonable first pass and real-tick mode is the confirmation. If the two diverge materially, treat it as a red flag before going live. We publish both for every BotGenomix bot, labelled simulated, alongside the licence tiers on our pricing page.

How do spread and slippage get modeled in a backtest versus real execution?

Fixed, current and historical spread in the tester

The Strategy Tester gives you three spread choices: fixed, current (whatever your broker quotes right now), or historical spread, which only works with real-tick data. Most curve-fitted results you see online were run on a fixed spread of 10 points on EURUSD. On indices and metals, where spread varies far more, a fixed assumption can overstate profit on every trade. Always rerun on historical spread before believing a result.

Why news-time spread widening never appears in most backtests

Historical spread data records what the broker quoted, minute by minute, so a US CPI release showing 30-80 point widening on gold will appear if your data source captured it. But most free tick downloads and any fixed-spread run contain none of it. An EA that trades the release itself can look profitable in the tester and stop out live, purely on spread. This is one of the core gaps we document per bot at BotGenomix, including which strategies avoid news windows entirely.

Slippage: why the tester fills at price and your broker does not

The tester fills every order exactly at the requested price unless you add a custom slippage model. Live, your stop orders on M15 indices can slip several points in fast markets, and a trailing stop fills at whatever the broker can get. A strategy with a 15-point target and 5-point stop on US30 is far more slippage-sensitive than an H4 swing setup with a 200-point target. Quantify it: run the same EA with a simulated 1-3 point adverse fill and see whether the profit factor survives. If it collapses, the edge was execution, not strategy.

Why does latency matter for scalping strategies but not swing strategies?

Latency is the round-trip time between your terminal (or VPS) sending an order and the broker filling it. Every EA eats some of it; the question is whether that delay is large relative to your trade's edge.

VPS-to-broker distance and order round-trip time

A VPS in London filling orders on a UK broker might see 5-20ms round trips; a VPS on the wrong continent can hit 150-300ms. The MetaQuotes Strategy Tester documentation models spread and slippage but not network latency, so a backtest cannot show you this gap at all.

Why H4 entries barely notice 100ms

On an H4 strategy targeting 80-200 pips, a 100ms delay costs a fraction of a pip in expectation. It is noise, and it disappears inside your spread assumptions.

Where latency compounds: trailing stops and fast exits

Scalping inverts the ratio: a 5-pip target with 100ms of delay can lose a meaningful slice of every winner. Trailing stops suffer twice, because the trail fires on a price that has already moved. Our own bots at BotGenomix run M15-H4 entries partly for this reason, and we publish the latency assumptions behind each licence tier at https://botgenomix.com/legal rather than pretending backtests capture execution.

How does broker execution policy change the same EA's results?

Market vs instant execution, requotes and stop levels

Market execution fills at the current price with no requote; instant execution sends a fixed price the broker can reject or requote. An EA that assumes instant fills will behave differently the moment it runs on a market-execution broker, especially around news. Stop levels matter too: a broker with a 30-point minimum stop distance silently rejects stops your backtest never simulated, and rejected modifications compound into missed exits.

Swap, commission and symbol naming differences

Backtests often ignore swap, or apply the tester's swap rather than your broker's triple-swap Wednesday charge. Commission modeling varies per symbol, and metals or indices may carry different contract sizes across brokers. On holding strategies (M15-H4 swings run for days), swap and commission can flip a profitable backtest negative.

Why the same.ex5 gives different results on two brokers

Symbol suffixes (EURUSD.a vs EURUSDm), different digits, tick history quality and fill policy all feed into results. Run the same.ex5 on a demo of your actual broker before buying any licence, including ours.

How can I tell which divergence source is hurting my specific strategy?

A diagnostic checklist by timeframe and style

Rank the divergence sources by how much they can hurt your setup. On M15-H4 indices, FX majors and metals, tick data quality and spread modeling dominate; latency and requotes matter far less. Work the list in this order: real-tick data availability for your symbols, spread behavior at your actual trading hours (indices widen badly at the US open), stop levels and execution policy on your broker, then latency only if you hold positions under a few minutes. Broker policy is the one you can check directly: IC Markets publishes its execution policy documentation, covering market versus instant execution, stop levels and requotes, and your broker's equivalent page tells you whether your EA's market orders can be rejected or re-quoted at all. If your EA uses instant execution with fixed stops, a market-execution broker will produce different fills on the same logic.

Re-running the same test across modes and brokers

Run the identical EA and settings through "Every tick based on real ticks" and "1 minute OHLC" on your symbol. If the equity curves diverge sharply, tick modeling is your gap. Then run the real-tick test against two brokers' data if you can export it; a spread-sensitive strategy will show the difference immediately.

Red flags that mean curve-fitting, not modeling error

Win rate above 85% with a profit factor above 2 on H1 and below, a single dominant parameter that shifts results wildly, or a backtest with almost no losing months all point to overfitting rather than simulation error. Walk-forward and out-of-sample tests exist to catch exactly this before live money does.

What realistic expectations should I set before running a strategy live?

Assume the live result will be worse than the backtest, and quantify by how much before you commit capital. A reasonable working rule: if your backtest shows a 15% max drawdown, expect 20-25% live once real spread, slippage and execution variance land. Size your risk per trade so that worse-case drawdown stays survivable on your £5k-£50k account.

Demo-first and small-capital staging on your own terminal

Run the EA on demo or a fraction of intended size for at least one to two months before scaling. Your VPS latency, broker spread at your session times, and execution policy are all specific to your setup, so your live numbers are the only ones that matter. Compare month one live against the backtest month by month; divergence beyond roughly 30% on profit factor is a signal to stop and diagnose.

Why losing months in a published backtest are a good sign

A backtest with no losing months is almost always curve-fitted. Real strategies on M15-H4 indices and metals hit flat or negative stretches when volatility regimes shift, and a vendor who publishes them is showing you where the strategy breaks, not just where it works.

How we document simulated results at BotGenomix

Every bot we sell shows its full month-by-month backtest including drawdowns and losing months, robustness checks, and a separate result on unseen 2025+ data, all labelled simulated. We charge a flat monthly licence, Super (£149/mo) or Expert (£249/mo), with no performance fees, so our incentive is retention rather than overselling a backtest. If a live month diverges badly from the published numbers, we would rather you see it early and cancel than discover it at max drawdown.

FAQ

Is 'Every tick based on real ticks' always the most accurate MT5 backtest mode?

It is the closest to live execution, but only as accurate as the tick history your broker recorded. Early data is often patchy, feeds differ between brokers, and the mode still ignores slippage, latency, and future spread behaviour. Treat it as confirmation, not truth: if it diverges sharply from synthetic mode on your symbol, diagnose before risking capital.

How much spread should I add to an MT5 backtest to make it realistic?

There is no single number to add. Run the test on historical spread with real-tick data, which captures news-time widening on gold and indices that a fixed 10-point EURUSD assumption never shows. Then stress it further by simulating 1-3 points of adverse slippage per fill. If profit factor collapses under that treatment, the edge was execution, not strategy.

Why does my EA backtest differently on a demo account than in the Strategy Tester?

A demo account is live trading with real spreads, real execution policy, and real latency; the tester simulates all of those. Demo fills happen at whatever the broker quotes, requotes and stop-level rejections included, none of which exist in a backtest. Divergence between the two usually points to spread modeling or execution policy rather than bad logic.

Can a backtest ever predict live drawdown accurately?

No, but it brackets it. Expect live drawdown to exceed the backtest figure once real spread, slippage and execution variance land; a 15% tested drawdown plausibly becomes 20-25% live. Use the backtest as a floor, size positions so the worse case stays survivable, and compare your first live months against published months to measure the actual gap.

Do I need a VPS if my EA trades on H4 charts?

Usually not for latency reasons. On an H4 strategy targeting 80-200 pips, even 150ms of extra round-trip time costs a fraction of a pip, noise inside your spread assumptions. A VPS still helps for uptime and power cuts if you hold positions for days. Latency only becomes decisive when targets shrink to a few pips.

Important: BotGenomix provides downloadable automated trading software for independent use. It does not provide personalised investment advice or manage customer trading accounts. Trading involves substantial risk, including the potential loss of capital. Past performance and backtested results are not reliable indicators of future performance. Full risk information