7 Backtesting Mistakes

01 Peeking at the future (hindsight bias)

What it looks like

You're scrolling a static chart, looking at candles that already happened, and you draw a trendline or spot a pattern that "would have" triggered an entry. You already know the candle after your entry went up, because it's sitting right there on screen.

Why it misleads

Your brain uses information from after the entry to decide the entry looks good. Every trade you find this way is contaminated by knowledge a real trader in that moment never had. The strategy isn't profitable — your hindsight is.

The fix

Test on a replay that reveals candles one at a time, so you make each decision with only the information that existed at that moment, and the next candle is genuinely unknown until you advance to it.

02 Testing too few trades

What it looks like

Eight trades, six winners, and a conclusion: "this works." The sample is small enough that a handful of lucky entries can carry the whole result.

Why it misleads

A small number of trades can look great or terrible almost at random. A coin flip can land heads six times in eight tries without being a biased coin — the same is true of a mediocre strategy hitting a short lucky streak.

The fix

Run the strategy across enough trades, and across enough separate time periods, that a handful of outliers can't decide the outcome by themselves. If you can't say roughly how many trades your test covered, you don't have a result yet — you have a story about eight trades.

03 Changing the rules mid-test

What it looks like

A trade goes against you, so you move the stop "just this once." A setup you'd normally skip looks tempting, so you take it because the chart "feels" right. By the end of the test the rules you started with aren't the rules you actually followed.

Why it misleads

You're no longer testing a strategy — you're testing your own improvisation, on data where you already know what happens next. That's not a repeatable system; it's a set of favorable exceptions you granted yourself with hindsight.

The fix

Write the entry, exit, stop-loss and position-size rules down before you start the test, and follow them exactly. If a rule needs to change, that's the end of one test and the start of a new one — don't blend the two.

04 Ignoring fees and slippage

What it looks like

The backtest counts pure price movement: bought at exactly the candle's price, sold at exactly the candle's price, no cost in between.

Why it misleads

Every real trade pays an exchange fee both ways, and a real order can fill at a slightly worse price than the one you saw. Those costs are small per trade and add up fast for a strategy that trades often — a system that's barely profitable on paper can be a loser once real costs are subtracted.

The fix

Know your strategy's trade frequency, and mentally (or on paper) subtract a fee for every entry and exit before you trust the result. A strategy that only looks good with zero costs isn't ready for a live account.

05 Testing only one market regime

What it looks like

The whole test runs on three months of a strong uptrend, because that's the period that was easy to find or easy to remember.

Why it misleads

A strategy that buys dips and holds can look brilliant in a trending market and lose money the moment the market turns sideways or drops. One regime tells you how the strategy handles one kind of market — nothing about the others.

The fix

Run the same rules across a range of conditions — a strong trend, a slow chop, a sharp drop — using history that goes back further than the last few months. If the strategy only survives in one of those, that's the actual result: a strategy for one regime, not a strategy.

06 No stop-loss, or sizing by feel

What it looks like

There's no stop-loss on the trade, "because it'll probably come back," and the position size was whatever felt right in the moment rather than a number you calculated.

Why it misleads

Without a stop, a single bad trade in the test can hide the strategy's real downside — in a backtest a losing trade can quietly recover before you close it, which isn't guaranteed in a live market. Sizing by feel means your risk per trade isn't consistent, so the test isn't measuring one strategy — it's measuring a different bet size every time.

The fix

Decide the stop-loss level as part of the entry rule, not after the trade is already open. Size every position the same way, based on a fixed percentage of the account you're willing to risk — a position size calculator does the arithmetic for you.

07 Trusting one great result (overfitting)

What it looks like

You adjust the strategy's parameters over and over — a slightly different indicator length, a slightly different entry filter — until one combination produces an outstanding result on that exact stretch of history.

Why it misleads

With enough small adjustments, you can eventually fit almost any set of rules to almost any known price history — that's what makes it "known" history. The fit describes the past you tuned it on, not any pattern likely to repeat.

The fix

Be suspicious of a result that only appears after many rounds of tweaking. Test the final rules, unchanged, on a separate stretch of history you didn't use while tuning — if the result falls apart there, the original result was overfitting, not an edge.

→ Next

New to backtesting entirely? Start with how to backtest a crypto strategy for the full process this article assumes. Sizing every trade consistently is mistake #6 above — the position size calculator handles the math.