A backtest is a story, not a promise
Every automated strategy begins with a beautiful chart: an equity curve climbing smoothly up and to the right, produced by running the strategy over years of historical prices. It feels like proof. It is not proof. It is a story the past tells about a set of rules — and like any story, it depends enormously on who collected the data, what was left out, and how many times the author rewrote the ending before showing it to you.
This guide covers the four traps that do most of the damage, and the tools on our shelves that help you avoid them. None of this requires a maths degree. It requires a suspicious mind.
Trap 1: bad data in, beautiful curve out
Most free price data is sampled — one price per minute, per hour, per day. A strategy that trades inside the bar (which most scalping and grid strategies do) cannot be tested honestly on sampled data, because the test has to *guess* what happened between the samples. Guess wrong and a losing strategy looks like a winner.
The honest fix is tick data: every single price change, in order. That is what the Data shelf is for:
- Dukascopy Data is free historical tick data covering 60+ pairs over 15+ years — the classic starting point, and the source behind most of the workflows below.
- Tickstory moves Dukascopy tick data into MT4 and MT5, with a free Lite tier and a Premium tier listed at around $99/year.
- Tick Data Suite is the dedicated MT4 tick-data backtesting tool, listed at $99 initial plus $10/mo in the inventory's older quote — worth rechecking at purchase, since that quote is dated.
Tick data is slower and heavier than sampled data. That is the point. A test that takes an afternoon and tells the truth beats a test that takes a minute and flatters you.
Trap 2: slippage and spread, the invisible tax
A backtest usually assumes you buy at the price you see. Real markets charge you for the privilege of trading: the spread between bid and ask, plus slippage when the price moves between your signal and your fill — worst exactly when the market is moving fast, which is exactly when your strategy most wants to trade.
A strategy that makes 2 pips per trade on paper and pays 1.5 pips in spread and slippage is not a 2-pip strategy. It is a 0.5-pip strategy with a rounding error standing between it and ruin. Any backtest that does not let you set spread and slippage explicitly should be treated as a rough sketch, not a result. Run every promising strategy twice: once with generous assumptions, once with harsh ones. If it only survives the generous version, it does not survive.
Trap 3: look-ahead — cheating without meaning to
Look-ahead bias is when a test accidentally uses information that would not have been available at the time of the trade. The classic version is a strategy that reads the close of the current bar to decide a trade "during" that bar. In real life, the close does not exist until the bar ends. In a backtest, it is sitting right there in the data file, quietly making your strategy look psychic.
Other versions are subtler: indicators computed over the full dataset (including the future), survivorship in the symbol list (only testing pairs that still exist), and "I noticed this pattern in 2021, then tested it on 2021." The defence is boring and effective: decide the rules first, then test on data you have not stared at. Out-of-sample testing — rules built on one period, verified on a later period the rules never saw — is the single cheapest honesty upgrade in trading.
Trap 4: overfitting — the strategy that memorised the past
Give a strategy enough adjustable parameters and it will eventually fit any historical period perfectly, the way a curve can be forced through any set of points if you add enough bends. That strategy has not discovered a market edge. It has memorised one particular stretch of history, noise included, and it will fall apart the moment the market changes character — which markets do constantly.
Warning signs: dozens of optimised parameters, spectacular results on one pair and one period, and performance that collapses the moment you change the test window by a few months. The antidote is the same as for look-ahead: fewer parameters, simpler rules, and a result that survives across multiple periods and, ideally, multiple symbols. A strategy that makes a modest profit in five different years is worth more than one that makes a fortune in one.
The tool that makes this practical
Generating strategies by hand and testing them one at a time is slow, which is why people skip the honest steps. StrategyQuant X, on the Backtesting & Optimisation shelf, exists for exactly this job: it generates and stress-tests strategies for MT4, MT5, cTrader and TradeStation, with a Starter licence listed at $990 and Professional at $1,490 one-time. Its value is not the generation — it is the stress testing: running the same strategy across varied data, with randomised slippage and spread, to see whether the edge survives or was a mirage. Whether you use it or not, steal its workflow: generate, stress-test, verify out-of-sample, and only then believe anything.
The honest backtest checklist
Stackbot's house rule: if a backtest result makes you excited, run it again with worse assumptions — excitement that survives pessimism is the only kind worth acting on.
- Tick data or it didn't happen for any strategy trading inside the bar.
- Spread and slippage set explicitly, and tested at harsh values too.
- Rules decided before the test, with a separate out-of-sample period the rules never saw.
- Few parameters, many periods. Simplicity that survives beats complexity that shines once.
- Written down, warts and all. The losing tests are data too — a journal (see our journals guide) is where backtest honesty turns into live honesty.
A backtest cannot tell you the future. It can tell you whether your idea survived the past under honest conditions — and that, plus a healthy suspicion of your own excitement, is the best starting point any trader gets.