Make MT5 backtests more realistic by auditing tick data, spread, commission, swap, execution delay, contract specifications and internal Renko assumptions.
A robot can produce different results on two brokers without changing a single input. The brokers may supply different quote paths, spreads, sessions and contract specifications. The tester may also use different cost or execution assumptions. Before searching for a better preset, audit the environment.
This guide builds a realism checklist for MT5 backtests, including robots that construct Renko internally. Its numerical examples demonstrate measurement. They do not prescribe a particular broker, instrument or trading configuration.
Separate the data model from the execution model
The data model determines which price events the EA receives. The execution model determines how orders, stops and costs are represented. Better input data can improve a test, but it cannot reproduce every live liquidity event, queue position or operational failure.
Think of the experiment as four layers: historical quotes, strategy decisions, order simulation and account accounting. A problem in any layer can change the result. A “real ticks” selection primarily addresses the first layer; it does not certify all four.
Record these layers separately in your research notes. “Real ticks, costs unknown” is more informative than “99% accurate.” Avoid assigning an overall accuracy percentage that the experiment has not measured.
What real ticks do—and do not—provide
Real-tick mode uses broker-supplied historical ticks where available. Generated-tick modes reconstruct an intraminute path from bar data. The real path matters when order matters: a high before a low can produce a different trade from a low before a high, even if the finished candle is identical.
MetaQuotes documents that missing real-tick intervals can be supplemented from minute data and that consistency with available bars matters. Inspect the journal and history coverage rather than assuming every event came from an uninterrupted real-tick feed. [1]
For a once-per-bar strategy, a simplified mode may be useful for limited development questions. For intrabar trailing stops, short-lived confirmations or internal Renko, the price sequence is central to the algorithm. Compare modes deliberately and explain any differences before relying on a faster run.
Spread is a cost and a path dependency
Spread is the difference between Bid and Ask. A purchase and its later sale are not evaluated against a single frictionless chart price. Around rollovers, session openings or unusual volatility, spreads may differ substantially from quieter periods.
Inspect costs where the strategy actually trades. An average all-day spread can conceal the fact that most entries occur during a wider-spread window. A scalping strategy with small expected trade outcomes may be particularly sensitive to this mismatch.
In real-tick histories, Bid/Ask behavior can vary within a minute. Do not describe the native tester as having one universal “spread multiplier” field. For deliberate spread perturbation, a controlled custom-symbol dataset or a suitable test harness may be needed. Document how you altered the data and ensure the change did not accidentally invalidate the symbol specification.
Commission and swap need explicit checks
Commission can be charged per side, per lot or through another broker schedule. Swap depends on the instrument, direction, holding period and broker rules. A preset that avoids overnight positions in one server timezone may cross rollover on another.
Inspect actual commission and swap columns in the tester’s ledger. If a test includes neither, do not silently call the result “after all costs.” State what is represented and what is not. Reconcile the assumptions with the broker’s specification and the EA’s intended use.
Consider a purely illustrative 200-trade record earning 6 currency units per trade before an additional 2-unit round-trip cost. The added cost removes 400 units and reduces the average by one third. If the original result already included that cost, subtracting it again would also be wrong. Keep a clear cost ledger.
Execution delay is not a full slippage model
MT5 provides execution-delay emulation in tester settings. This can help examine timing sensitivity, but it should not be treated as a complete reconstruction of broker liquidity or all live fill behavior. MetaQuotes describes delay settings as part of strategy optimization and testing. [2]
Different order types also have different triggering and execution rules. Non-exchange and exchange instruments may use different price references. A signal visible on a Last-price chart does not imply that the resulting order trades at Last. Check Bid, Ask and the relevant order type. [3]
Perform a small diagnostic comparison with documented delay assumptions. If a slight timing change removes the apparent advantage, that is a research finding. Do not optimize a replacement preset on every delay case and then call the whole exercise independent validation.
Audit points, ticks and contract size
A point is the platform’s quoted price increment associated with the symbol’s digits. The minimum tradable price increment is the tick size, which may differ. Tick value and contract size determine the monetary effect of a move at a given volume.
For a hypothetical symbol with point size 0.01, a 200-point setting represents 2.00 price units. For another with point size 0.10, 200 points represents 20.00. These examples are arithmetic only. Use the actual symbol specification before transferring any preset.
Check minimum lot, lot step, maximum volume and stop restrictions. If the EA rounds volume, the actual exposure may differ from a simple theoretical risk calculation. If an order is blocked because the volume or stop is invalid, retain that finding instead of editing the report to remove “bad” runs.
Internal Renko adds another audit layer
A Renko strategy does not remove dependency on quotes. Brick construction depends on box units, reversal rules, the chosen price stream, treatment of wicks and event ordering. Two implementations using the word “Renko” may build different structures from the same input.
For EA #001, document its internal fixed-box construction and its host chart separately. Treat the supplied US100 preset as one broker-specific research configuration. Changing the host timeframe does not necessarily change the internal signal timeframe; changing tick history or box size can.
Check the warm-up period and what happens when one quote crosses several box boundaries. A large gap can produce multiple internal events. Visual inspection should verify how confirmation and orders behave during those events rather than assuming a brick picture completely explains execution.
Design a broker comparison that answers a question
Keep EA version and strategy rules fixed. First compare symbol specifications, then align the economic meaning of inputs. Where you must translate point-based distances, record the translation. After that, compare history coverage, costs, sessions and execution assumptions.
If results differ, identify the earliest differing signal or fill. Work backward to the quote stream and configuration at that time. This is often more useful than comparing only final net profit. A difference near a session boundary suggests a different investigation from a difference caused by volume rounding.
Realism checklist
- Exact broker server and symbol suffix are recorded.
- Data coverage and journal warnings have been reviewed.
- Modeling mode fits the strategy’s event sensitivity.
- Spread, commission and swap assumptions are explicit.
- Delay and fill assumptions are not described as guarantees.
- Point size, tick size, volume steps and stops are checked.
- Renko construction and server-time filters are documented.
Frequently asked questions
Are real ticks enough for a trustworthy backtest?
They improve the representation of the historical quote path where available. Trust also requires correct software behavior, suitable costs, execution assumptions and disciplined validation.
Can I use the same SET file at every broker?
Only after checking its meaning on each symbol. Point-based distances, sessions and volume constraints can differ. A portable strategy idea does not imply a portable unchanged preset.
Does a zero-commission report mean no trading costs?
No. Spread, swap and execution effects may still exist, and the commission setting itself needs verification. Label the actual assumptions rather than generalizing from one ledger.
Sources and further reading
Original educational guide by MT5 Robot Lab. Platform details are checked against the linked documentation; worked examples and diagrams are identified as illustrative. Last updated: .