Learn › Backtesting without code

How to Backtest a Trading Strategy Without Code

A backtest applies a fixed set of trading rules to historical market data and records what those rules would have done: every buy and sell, the value of the account along the way, and what it cost. Code is one way to write the rules down. A visual builder is another — the same conditions, joined with AND, OR and NOT, wired together instead of typed.

What a backtest measures

A backtest answers a narrow question: what would these exact rules have done on this data, over this period, at these costs? Its output is a record of trades and an equity curve — the account's value at every point in the test.

It describes the past. A backtest is not a forecast: a rule that worked on one decade of one stock is evidence about that decade and that stock. Its value comes from making an idea specific enough to measure, so it can be compared, repeated and questioned.

1. State the idea as a question

Start from something a chart, a book or a post suggests, and write it as a question with a comparison in it. For example: did buying when the 14-day RSI fell below 30, and selling when it rose back above 70, do better than simply holding the same stock over the same years?

The comparison is part of the question. "It made money" is a weak result when holding the stock made more with less effort; the question decides what the result is measured against.

2. Turn it into exact rules

A rule is testable when a computer could apply it without asking you anything. Each strategy needs:

• An entry — the condition that opens a position, e.g. RSI(14) crosses below 30.
• An exit — a signal (RSI crosses above 70), a target or stop measured from the entry price, a number of bars held, or a time such as the end of the session.
• A size — all available cash, a fixed amount, or a percentage per buy.

Vague words are where tests go wrong. "When RSI is low" has no threshold; "a strong breakout" has no definition. Each has to become a number, a crossing or a comparison between two lines. A rule must also be computable on the candle it fires: using a value that only becomes known later (the day's close, before the close has happened) is look-ahead bias, and it makes any strategy look better than it could ever have been.

3. Build the rules without code

In a visual strategy builder the rule is a diagram. Each condition is a block — RSI < 30, Close crosses above the 50-day SMA, a candlestick pattern — and blocks are joined through logic gates:

• AND is true when every input is true; OR when any is; NOT inverts one.
• The combined condition is wired into a BUY, SELL or TRIM (partial sale) terminal.
• Exits that depend on the open position — a percentage gain, a stop loss, a number of bars held — and exits on the clock, such as the end of the session, are blocks of their own.

The diagram is the rule: what is wired is what runs, so there is no second description of the strategy to keep in step with the code.

4. Choose the instruments, the period and the timeframe

• Instruments. One stock tells you about one stock. Running the same rules across several — ideally a list chosen before seeing any result — shows whether the result is general or belongs to one lucky ticker.
• Period. A test covering only a rising market says little about a falling one. Longer periods that include different conditions give the rule more to prove.
• Timeframe. Weekly, daily, hourly and five-minute candles are different tests. A rule "of 50 bars" spans about a year on weekly candles, about ten weeks on daily candles and about four hours on five-minute ones.
• Gaps in the data. A test whose period spans more than five missing trading sessions in a stock's prices is refused, with an option to start the test after the gap.

Survivorship bias belongs here too: testing only on companies that exist today leaves out the ones that failed or were delisted, which flatters most strategies.

5. Include costs

Every trade pays something: commission, the bid–ask spread, and slippage between the price you expected and the price you got. A rule that trades often can turn a small gross gain into a net loss. Include a fee per trade at a level you consider realistic; a zero-cost run is useful as a comparison, to see how much the costs changed the result.

6. Read the results

A single number rarely describes a strategy. Read these together:

• Total return — the change in the account over the whole test.
• Maximum drawdown — the largest fall from a previous peak. A 33% drawdown means the account was worth a third less than its best earlier value and had to regain that ground first.
• Time in the market — the share of the period a position was open. A strategy invested 30% of the time carries different risk from one invested throughout.
• Number of trades — five trades is an anecdote; results rest on more.
• Win rate — how often a trade closed in profit. It says nothing about how large the wins and losses were: a strategy can win 80% of trades and lose money, or win 35% and make money.
• Fees paid — how much of the gross result went on costs.

A high return on its own does not establish that a strategy is useful. It has to be read against a benchmark.

7. Compare with a benchmark

The benchmark is whatever the question compared against — most often, buying the same instrument at the start and holding it to the end. Build that as its own strategy and run it over the same period, at the same costs. Then compare return, drawdown and time in the market side by side.

The Test Bench's first study is a worked example. It ran three Donchian Channel breakout rules on 179 US stocks over eight years and compared each with holding the stock; its central finding is that profitable is not the same as better.

The RSI mean reversion backtest does the same for four RSI entries under one rule, each set beside buying and holding and beside its own time in the market.

8. Check that the result holds up

Overfitting is what happens when a rule is adjusted, again and again, until it fits the history it is being tested on. Each change that improves the backtest is kept; the final rule describes the past very well and the future no better than chance. The more variations tried, the more likely one looks good by luck — the problem of multiple comparisons.

Common checks against it:

• Out-of-sample testing — fix the rule on one period, then run it unchanged on another.
• Other instruments — run the same rule on stocks it was not developed on.
• Different market regimes — rising, falling, calm and volatile periods, separately.
• Few, round parameters — a rule that only works at 17 bars and not at 15 or 20 is likely fitted to noise.
• Write the rule down before the run — and keep the result either way.

Common mistakes

• Using information the rule could not have had at the time (look-ahead).
• Testing only on stocks that survived.
• Leaving out costs.
• Judging on total return alone, without drawdown, time in the market or a benchmark.
• Reading a strong result from a handful of trades as a pattern.
• Choosing the ticker or the period after seeing which one worked.

Common questions

Doing this in GU Analyser

GU Analyser's strategy canvas is the visual builder described above: Signal blocks for conditions on any of its 60+ indicators and on price; Logical blocks for AND, OR and NOT; BUY, SELL and TRIM terminals; Position exits for a percentage gain, bars held or a stop loss; and Time exits for the end of the session or the week. Premium adds candlestick-pattern conditions and timing windows for conditions that do not line up on the same candle.

• Each trade fills at the close of the candle on which its rule fires.
• Fees are set as a percentage per buy and per sell. Slippage and the spread are not modelled separately, so a fee that includes them is the conservative setting.
• The result shows the return, maximum drawdown and its recovery, time in the market, fees and every trade, with the entries and exits marked on the chart.
• A buy-and-hold benchmark is a second strategy — buy on the first candle, never sell — run on the same ticker and period.
• The free plan runs everything on the S&P 500 ETF (SPY), one strategy at a time. Premium runs up to five tickers or five strategies side by side across 400+ US stocks, ETFs and indices.

For the indicators most often used in first strategies, the reference pages explain what each measures: RSI, MACD, Simple Moving Average, Bollinger Bands and Donchian Channel. To study what followed a single condition rather than a full strategy, see event analysis.