How to Backtest a Trading Strategy (Without Fooling Yourself)

By Nummius · October 18, 2026

Backtesting means applying a trading strategy’s rules to historical price data to see how it would have performed. Done well, it tells you whether an idea deserves real money. Done badly, it produces a beautiful equity curve that falls apart the week you start trading it. This guide walks through a process that keeps the test honest, from writing the rules to forward testing.

Step 1: write rules a computer could follow

“Buy strong stocks on pullbacks” is an idea, not a strategy. A testable strategy removes discretion from every decision:

• Universe: what you trade, for example large-cap US stocks or the ES futures contract
• Entry: the exact condition, for example “the daily close crosses above the 20-day moving average while the 50-day moving average is rising”
• Exit: the stop loss, the profit target, a time stop, a trailing rule, or a combination
• Position sizing: fixed shares, a fixed dollar risk, or a percentage of equity
• Filters: time of day, market regime, minimum volume or price

A good test: if two people read your rules and would take different trades on the same chart, the rules are not finished yet.

Step 2: get data that matches how you trade

Your backtest can only be as good as its data. Check these before you run anything:

• Timeframe. Test intraday strategies on intraday bars. A daily bar does not tell you whether your stop or your target was hit first on a day that touched both.
• Adjusted prices for stocks. Splits and dividends create fake gaps in unadjusted data.
• Survivorship bias. A list of today’s index members leaves out the companies that were delisted, acquired or dropped. Testing only on today’s survivors flatters the result.
• Futures rolls. Continuous futures contracts are stitched together from expiring contracts. Know how your data handles the gap at each roll.
• Enough history. Include different regimes: trending and ranging markets, calm and volatile periods, rising and falling rates.

Step 3: choose manual or automated backtesting

There are two broad ways to backtest, and they answer different questions.

Manual backtesting (bar-by-bar replay) hides the future and steps through the chart one bar at a time while you take trades by your rules. It is slow, but it suits discretionary setups and builds pattern recognition that a spreadsheet never will. Its main weakness is accidental peeking at what comes next.

Automated backtesting (systematic testing) applies coded rules to all the data at once. It is fast, repeatable and can produce thousands of trades, but only for rules that can be written precisely.

Many traders use both: replay to develop and understand a setup, then an automated test to measure it over a large sample.

Step 4: model costs realistically

Include commissions, exchange fees and slippage, which is the difference between the price you expected and the price you got. Strategies that trade often with small average gains are the most sensitive.

Be conservative with slippage on market orders, stop orders and thinly traded instruments.

Step 5: read the right metrics

A single net profit number hides most of what matters. Read these together:

If R and expectancy are new terms, our guide to the risk-reward ratio explains them with worked examples.

Step 6: guard against the classic biases

• Look-ahead bias: using information that was not available when the decision was made. Common forms are entering at a bar’s open based on that bar’s close, or using economic data that was revised later.
• Overfitting (curve fitting): tuning parameters until the past looks perfect. Warning signs are many parameters, results that collapse when a parameter changes slightly, and oddly specific values, such as a 17-period average beating both 15 and 20 by a wide margin.
• Data snooping: testing dozens of ideas on the same data and keeping the best. Some of them will look good by pure chance.
• Survivorship bias: see step 2.

Step 7: validate out of sample and walk forward

Split your data. Develop and tune the strategy on one period (in-sample), then test it once on a period you never touched (out-of-sample). If performance collapses out of sample, the edge was probably never there.

Walk-forward analysis repeats that idea in a loop: optimize on a window, for example two years, test on the following window, for example six months, then roll both windows forward and repeat. Stitching the out-of-sample pieces together gives a far more realistic picture than one optimized backtest.

When choosing parameters, prefer a stable plateau, where neighboring values perform similarly, over a single sharp peak. A plateau suggests the strategy captures something real; a peak suggests it captured noise.

Step 8: forward test before risking real money

Paper trade the strategy or trade it at a very small size for a while. Compare live fills, slippage and trade frequency with the backtest; every difference tells you something the backtest missed. Also check the human side. A 15% drawdown looks harmless on a chart of ten years and feels very different when you live through it day by day.

A backtesting checklist

• The rules are unambiguous and written down
• The data matches the timeframe and is adjusted, with survivorship handled
• Commissions and slippage are included
• There are enough trades, across different market regimes
• Results hold up when parameters change slightly
• Out-of-sample or walk-forward results are acceptable
• Forward-test results resemble the backtest

Backtesting in Nummius

Nummius Core includes manual and systematic backtesting for stocks and futures, with 80+ indicators and walk-forward validation; plan details are on the pricing page. If you want to practice reading real historical charts without money on the line, the free Duelist plan includes replay of real market data for paper trading and the Trading Arena, where you trade 1v1 duels on historical charts. Backtested and live trades can then be reviewed the same way in the trading journal.

Key takeaways

• A backtest is only as good as its rules, its data and its cost assumptions.
• Read drawdown, trade count and expectancy alongside net profit.
• Distrust results that depend on precise parameter values.
• Validate out of sample, walk forward, then forward test small.

Once a strategy passes, keep a trading journal of the live trades so you can see whether reality matches the test.

This article is for educational purposes only and is not financial advice. Past performance, whether real or simulated in a backtest, does not guarantee future results. Trading involves risk, including the loss of principal.