How to Backtest a Trading Strategy Without Coding

Sam's table from Lesson 79 had 10 trades in it, and he knew ten trades couldn't prove anything. Waiting for 50 more live ones would take months of mornings, and every one of them would cost real money to learn from.

But here's the thing about the market: it keeps a diary. Every candle from the last few months was still sitting on his charts, waiting to be read.

So on Saturday morning he opened a spreadsheet, typed his pullbackPullbackA pullback is a temporary counter-trend price movement within an established trend — in an uptrend, a pullback is a short-term decline that offers a lower-risk entry point before the trend resumes, as opposed to a reversal which signals theRead definition → rules from Lesson 77 across the top and started stepping through old charts one candle at a time. By Sunday night he had 50 trades and one uncomfortable discovery. We'll get to that.

In this lesson you'll learn what backtesting is, a five-step manual method that needs no code, what to record, which numbers matter, how easily a backtest can lie (DayTradingToolkit simulated it) and what Sam's 50-trade backtest showed. It builds on Lesson 56: Win Rate vs. Risk/Reward and Lesson 74: Your First Trading Plan.

Backtesting is the process of applying a trading strategy's rules to historical price data to see how the strategy would have performed if you'd traded it in the past.

Think of a flight simulator. A pilot crashes a hundred times on the ground so they never have to crash in the air. A backtest lets your rules "crash" on past charts, so you find out where they break while it only costs you an evening.

A backtest isn't a prediction, though. It tells you how your rules would have done during one stretch of the past, under the conditions of that time. Treat it as a filter that removes bad ideas, not a certificate that proves good ones.

Backtesting shows you whether a set of rules has any chance of working, how often it loses and how deep its losing streaks run, in days instead of months.

It does three other jobs that beginners don't expect:

A manual backtest takes five steps: write exact rules, pick a fair sample, replay the charts, record every trade and analyze the results.

Write the entry trigger, the stop, the target, the position size and the conditions under which you skip a trade. Here's the test: if two people could read your rules and take different trades, the rules aren't exact yet.

Keep the rules few. A rule of thumb is three to five entry conditions. Each extra condition is another place for the test to quietly fit itself to the past.

Choose your stocks and your dates before you look at a single trade. Include different kinds of days: trend days, range days and big news days (Lesson 79 shows how to label them). If your sample is one strong month, your result is a result about that month.

Most charting platforms have a bar replay mode: you pick a past starting point and step forward one bar at a time. TradingView's replay lets you press "Forward" to advance a step or play at an adjustable speed, though how much intraday history you can replay depends on your plan and the symbol (TradingView Support). If your platform has no replay, cover the right side of your screen with a sheet of paper and scroll slowly. It looks silly, and it works.

At each candle, decide with only what was visible then. If the setup matches every rule, log it. If it matches most of them, skip it. No "it would have worked."

Log each trade the moment you take it in the replay, including the ones you'd rather forget. Here's what one row might look like:

Record results in R (your planned risk as one unit), so the test doesn't depend on how many shares you'd have bought. In that row, the risk was $0.32 and the reward was $0.55, which is +1.72R before costs and +1.6R after about two cents per share of slippage and the spread. Deduct a realistic estimate of costs on every trade, because costs are the first thing that disappears from a backtest and the first thing that shows up in real trading.

Once all the trades are in, calculate:

Easily, and usually in your favor, which is exactly why it's dangerous. Here are the biases to build your test around.

Overfitting happens when you keep adjusting rules after seeing results, so the strategy learns the noise in your sample instead of a real pattern. It sneaks in one reasonable-sounding tweak at a time: "Let me skip Mondays." "Let me raise the volume filter." Each one makes the past look better.

DayTradingToolkit simulated how powerful this is. DayTradingToolkit created rule variations that have no edge at all (they win 33% of the time with 2R winners and 1R losers, so the true expectancy is exactly zero). For each run, the simulation tested N variations on 50 trades each, picked the one that looked best, and then tested that winner on 50 brand-new trades. That was repeated 4,000 times.

DayTradingToolkit simulation: 4,000 runs per row, 50 trades per test, fixed seed, independent variations. Not real market data.

Read that table twice. Even a rule with a true edge of zero showed +0.16R or better in about one 50-trade test out of five. Try 20 variations and the best of them almost always looks like a winner, and then it falls back to nothing on new data. (In real life, variations of one rule on the same charts are correlated, so the inflation is smaller, but it points the same way.)

Researchers have put numbers on this. Bailey, Borwein, López de Prado and Zhu showed that with only five years of daily data, no more than about 45 variations of a strategy should be tried before the best one is likely to look good by chance alone. And a paper by Sullivan, Timmermann and White built a bootstrap method to adjust for this "data snooping" on the same rules Brock, Lakonishok and LeBaron had studied, using 100 years of Dow Jones data.

Real edges also shrink once they leave the lab. A study of 97 stock-return predictors found average monthly returns of 0.582% in the original samples, 0.402% in the out-of-sample period before publication and 0.264% after publication (McLean and Pontiff, 2016, summarized by the CFA Institute). That's about stock factors, not day trading, but the lesson carries over: expect the live version of anything to be weaker than the backtest.

Simple arithmetic on Sam's hypothetical backtest result.

A thin edge can vanish into a cost you never put in the spreadsheet. That's why a modest, honestly costed backtest beats a spectacular one.