Backtesting Prerequisites

How to Backtest a XAUUSD Strategy:
What You Must Know Before You Start

Most backtest failures happen before a single trade runs. Five specific setup errors invalidate everything that follows — fix these first, or your results mean nothing regardless of how carefully you analyse them.

Quick Answer

Before backtesting any XAUUSD strategy, you need 99% tick data quality (not simulated ticks), a realistic spread set to your broker's ECN average (10–15 pips), a date range chosen before viewing results, separate periods for optimisation and testing, and at minimum 3 years of data covering multiple market regimes. Miss any one of these and the results are statistically meaningless.

The 5 Backtest Killers

Each of these errors invalidates your entire backtest before a single trade runs. Click any card to see the full breakdown.

1

Killer #1: Wrong Tick Data Quality

Running with 'Open prices only' or 1-min bars on a scalping strategy. Every result is fiction.

2

Killer #2: Zero Spread Setting

If your strategy tester has 0 for spread, every result is 20–40% better than real life.

3

Killer #3: Cherry-Picked Date Range

Starting your test from a date that happens to be the beginning of a favorable period — without knowing it.

4

Killer #4: Testing on Optimised Data

In-sample optimisation + in-sample testing = guaranteed good results that mean nothing.

5

Killer #5: Test Period Too Short

Under 1 year of data catches maybe 2–3 different market conditions. You need at least 3 years.

Backtest Validity Checker

Answer honestly — these are pass/fail checks. A single 'No' in the first four can invalidate all results.

1

Is your modelling quality 99% (Every tick based on real ticks)?

2

Is your spread set to your real ECN broker's average spread (not 0)?

3

Did you choose your date range BEFORE looking at any results?

4

Are you testing on different dates than you used for optimisation?

5

Is your test period at least 3 years covering multiple market regimes?

Why Prerequisites Matter More Than Methodology

Most backtesting guides jump straight into the step-by-step process: open Strategy Tester, select the EA, click Start. But the methodology is only valid if the foundation is correct. Running a perfect 8-step process on bad tick data produces the same result as running a sloppy process — meaningless numbers.

The five killers above are ordered by how early they corrupt the results. Tick data quality is the foundation — everything else is built on top. If you have 99% quality data but set spread to 0, at least the trade timing is real and you have something to correct. If you have 0% quality data with perfect spread, you cannot fix the results at all.

Think of it as a lab experiment. You can analyse contaminated samples using perfect laboratory technique — your technique will be flawless and your results will be completely wrong. Backtesting is the same: perfect methodology, wrong inputs, wrong output.

The Psychological Traps That Create These Mistakes

Most backtest mistakes are not technical errors — they are psychological ones. Traders want the strategy to work, so they unconsciously make choices that bias results toward positive. The zero-spread default exists in MT5 and most traders leave it because they do not know about it — but some leave it because the results look better with 0 spread and they do not investigate why.

Cherry-picking date ranges is almost never deliberate. It happens because you glance at the chart, see a period that looks suitable, and use it. The fact that you glanced at the chart means you already introduced selection bias before the test started.

The discipline required for valid backtesting is deciding every parameter — date range, spread, modelling quality, in/out-of-sample split — before seeing any results. Write them down. Commit. Then run. Any decision made after seeing results is contaminated by what you saw.

What Bad Setup Actually Costs: Real Numbers

Zero Spread Impact

100 trades at 0.01 lot:

$120

Hidden cost at 12-pip spread
vs 0-spread backtest result

Simulated vs Real Ticks

Typical win rate inflation:

+15–25%

Simulated ticks vs real ticks
for intrabar strategies

Regime Coverage

1 year covers:

1–2

market regimes
vs 5–6 in a 5-year test

Building a Valid Backtest Environment from Scratch

If you are starting fresh, here is the exact sequence to ensure all five killers are avoided before you run a single test:

01

Get real tick data first

Visit dukascopy.com/swiss/english/marketwatch/historical/ and download XAUUSD data from January 2018 (or earlier if available) to present. The .csv format works with most MT5 tick importers. Alternatively, download from your broker's server via MT5's History Center — but verify the data goes back at least 5 years and check for gaps in 2020.

02

Write down your parameters before testing

On paper or a text file, record: date range (e.g. 01/01/2019 – 31/12/2024), spread (e.g. 12 pips), modelling (Every Tick Based on Real Ticks), in-sample cutoff (31/03/2024), out-of-sample period (01/04/2024 – 31/12/2024). These do not change after this point regardless of what results show.

03

Set up MT5 Strategy Tester exactly

Symbol: XAUUSD. Model: Every Tick Based on Real Ticks. Period: your EA's chart timeframe. Date: your in-sample range only. Click Expert Properties → spread: your committed number. Confirm 'Use date' is checked. Do not change any of these settings after you see results from the in-sample run.

04

Run in-sample first, record metrics without optimising

Run with default EA parameters first to establish a baseline. Record profit factor, max drawdown, and consecutive losses. Then optimise if needed — but the optimisation target must be set before running. If you see results that tempt you to change the optimisation criteria, write down the temptation and ignore it.

05

Run out-of-sample exactly once

Take your final settings and run them on your held-back period. This is a one-time read-only test. If out-of-sample profit factor is within 30% of in-sample, the strategy has passed. If it is worse by 50% or more, the settings are curve-fitted and should not be traded live regardless of how good the in-sample looks.

Reference Setup: What Valid Backtesting Looks Like

As a practical reference, when backtesting an M15 breakout strategy on XAUUSD (such as Goldie Razor V2.8.4), the valid setup parameters are: date range January 2019 – December 2024 to cover five distinct market regimes, spread set to 12–15 pips matching typical ECN conditions during London and New York sessions, Every Tick Based on Real Ticks modelling, and an out-of-sample validation on 2024 data only with optimisation performed exclusively on 2019–2023 data.

The same framework applies to any XAUUSD strategy. The parameters are not strategy-specific — they are the minimum valid environment for the instrument and timeframe combination. If you are evaluating an EA from any source, ask whether these exact conditions were met. If they were not, the numbers you are looking at cannot be trusted as performance predictions.

What Comes After Prerequisites

Once all five killers are addressed, the actual methodology becomes the focus: downloading and importing tick data correctly, running the optimiser with appropriate genetic algorithm settings, interpreting the metrics (profit factor, Sharpe ratio, consecutive losses), and conducting proper out-of-sample validation. These are covered in the full methodology guide linked below.

If you are evaluating a backtest someone else has run — an EA vendor's backtest, a signal provider's performance record — the same five prerequisites apply as filters. A clean-looking backtest that fails even one of these checks should be treated with significant skepticism until the gap is explained.

Frequently Asked Questions

Goldie Razor V2.8.4

M15 breakout + H4 EMA filter — built for XAUUSD on MT5

View Goldie Razor →