What Is Backtesting in Trading? Plain English—Plus Where Ours Fits

What Is Backtesting?

Backtesting is replaying a precise trading rule over past prices and counting what would have happened. You take an idea — say, “buy when a stock breaks out of a bull pennant, sell at the target or the safety exit” — turn it into exact instructions, and walk it through history one day at a time, recording every trade it would have made.

A plain example: apply that pennant rule to ten years of daily prices for 500 stocks. The computer finds every qualifying pennant the detector found in the supported history, “buys” at each eligible breakout, “sells” wherever the rules say, and writes each result down. At the end you have a pile of simulated trades: how many won, how many lost, and how big each result was.

The output isn't a prediction. It's a historical track record for an idea. Before risking a real dollar, you can see how the rule behaved across thousands of past market situations, including the ugly ones.

Why Backtesting Beats Opinions

Human memory is a terrible statistician. You remember the spectacular chart where a pattern nailed the move, and quietly forget the four charts where it fizzled. Every trading forum is full of patterns that “work” — in someone's memory.

A backtest has to keep the losers. A pattern working in your memory might be three lucky charts; a rule applied across thousands of counted trades has at least faced a much harder exam. That still does not guarantee the future, but it is more useful than a highlight reel.

Counting is boring, and boring is the point. It replaces “do I believe this?” with a better question: “What happened when we applied the same rule every time?”

How a Backtest Works, Step by Step

A useful rule-based backtest usually follows the same four moves:

The Mistakes That Make Backtests Lie: Look-Ahead Bias and Friends

A backtest is only as honest as its builder. These are the three traps to look for before trusting a beautiful equity curve:

• Look-ahead bias: letting the test peek at information that did not exist yet. A test that “buys at the day's low” is cheating because nobody knows the low until the day is over.
• Survivorship bias: testing only stocks that survived. A universe made from today's winners can quietly omit failed or delisted companies that would have hurt the result.
• Curve fitting (overfitting): adding conditions until the rule fits one history perfectly. A rule with fourteen carefully tuned exceptions may have memorized the past instead of finding something repeatable.

A Strong Defense: In-Sample vs. Out-of-Sample

One strong defense against curve fitting is to split the history before you tune the rule. The first pile — the “in-sample” data — is the practice set you may study. The second — the “out-of-sample” data — is the exam: years and stocks the finished rule did not see while it was being designed.

A fitted rule often looks best on the practice set because that is where it was designed. The useful question is how much of the result remains on the locked-away data. The edge may hold up, weaken, change shape, or disappear. Less degradation is stronger evidence that the rule learned something repeatable, but no single split proves a real edge by itself.

TradingPal uses held-out tests when a filter, gate, or policy choice is being tuned under a frozen research protocol. A candidate earns its keep on one slice of history, then has to hold up on symbols or years it was not tuned on. A good in-sample score is not enough. The complete methodology explains which parts use held-out validation, which use walk-forward ranking, and which limitations still remain.

One honest limit: out-of-sample data is still the past. Passing the exam isn't a promise about next year—the future is the one set of prices nobody can hold out in advance. It is simply stronger evidence than an in-sample score from the same history used to tune the rule.

How TradingPal Backtests

After the market closes, the pattern pipeline scans 500+ stocks and refreshes the stored results the product serves. Here is how one simulated pattern trade works; for portfolio caps, 25-run medians, cost scenarios, and live drift checks, continue to How TradingPal backtests.

After the pattern existed on a prior bar, an entry can fill at the confirmed trigger—the boundary plus its close-confirm threshold. When a stock gaps through that trigger overnight, the simulation uses the worse opening price instead, because the planned trigger price was no longer available.

Every trade carries a declared safety exit (the “stop-loss”) beyond the relevant recent swing—a dip for a long or a peak for a short—and every simulated trade follows its mechanical exit rule. Targets come from the pattern's own height (the “measured move”); some upward breakouts can keep riding above the 10-day average after reaching that target because that rule was declared in advance.

One limitation matters: the published pattern group contains shapes that ultimately passed the detector's checks after the first crossing. The first eligible fill uses only information available then, but the published group is not proof that every pattern was already known to be confirmed at its first tick. The result is tens of thousands of simulated trades, winners and losers included, summarized in the latest completed stored result.

The Latest Nightly Results

This table comes from the latest completed nightly result — each pattern's win rate, its average return in R, and the number of simulated trades behind the row. The article renders the stored result immediately instead of starting a new calculation.

Where the Backtest Shows up in TradingPal

The key outputs from the latest completed stored result flow into the app. The screener's left rail carries the track-record panel—the real component below—summarizing the current family's win rate, the cumulative return and CAGR of a capital-constrained account trading it, its drawdown, and profit factor. Qualifying setup tiles and the copilot can also show the relevant historical context.

That's the point of doing all this counting: not a research paper hidden in a folder, but historical context beside a qualifying setup when you're deciding what to study next.

What a Backtest Can't Promise

The past isn't a promise. Markets change, and a rule that thrived in one era can fade in the next. A long test and a held-out exam make the evidence stronger; they do not turn it into a forecast.

Real fills can also differ from simulated ones, and today's supported stock universe does not include every company that disappeared from history. The honest way to hold all of this in your head: a backtest tells you what happened under stated rules and assumptions. It does not tell you what the next trade will do.

What Is Backtesting?: FAQ

Keep Learning