Backtesting: How to Check a Strategy Before You Trade It

A trading idea looks convincing while it stays in your head. Two or three good examples on the chart, and the approach feels solid. Memory helpfully collects the confirmations and drops the cases where the same rule produced a loss. Backtesting turns an idea from a feeling into numbers. You take the written rules, run them across history and look at the result of a hundred trades instead of the three you happen to remember.

Below we go through how to run the check by hand, what to measure afterwards and which mistakes turn a good-looking report into self-deception.

What Backtesting Is

Backtesting is a check of trading rules on historical data without risking money. You scroll the chart back, apply your entry and exit conditions to past bars and collect the list of trades those rules would have produced.

Only rules written down unambiguously can be tested. The phrase "I enter when the market looks strong" cannot be tested, because a week from now you will read it differently than you do today. The wording "I enter with a limit order on a pullback to the level, stop beyond the local low, target two risks away" is already testable. If your rules are not yet collected in one document, it makes sense to put together a trader's trading plan first and test that. If you do not have an idea of your own yet, take a ready approach from the review Crypto Trading Strategies: Top 7 Approaches That Actually Work 2026 and narrow it down to specific conditions.

Checking an idea usually runs through three stages, and the backtest is only the first one.

• Backtest on history. Fast, cheap, gives the first statistics and filters out rules that clearly do not work.
• Forward test in real time. You take trades as new data arrives, with no looking ahead.
• A small real amount. Execution, fees and your own reactions appear, none of which exist in history.

What a backtest will not show is worth knowing in advance too. It does not check your discipline in the moment, does not account for execution speed or liquidity at your size, and does not reproduce the emotions that follow three losses in a row. History shows how prices behaved, not how you will behave.

How to Test

Two practical methods are available to any trader without programming. The first is manual review on the chart. The second is Replay mode in TradingView, which simulates movement bar by bar.

Manual Backtest on the Chart

The working order looks like this.

• Write the rules down before the test starts. Entry condition, stop, target or exit condition, risk per trade, filters by time and volatility.
• Pick an instrument and a period. Take a stretch with different market regimes, including a trend, an extended range and sharp volatility spikes.
• Hide the future. Scroll the chart so the right side is out of view, or switch on Replay.
• Record every trade immediately. Date, direction, entry price, stop, exit, result in R and a short comment. It is easiest to keep that record where your live statistics will later live, which is covered in detail in Trader's Trading Journal: Why You Need It and How to Keep One.
• Add up the result only after the whole sample, not after every tenth trade.

R here means the risk of a single trade, that is the distance from entry to stop in money. A result in R is more convenient than percentages, because it does not depend on deposit size.

A practical guideline for sample size is 50-100 trades across several market phases. That is not a magic threshold, just the minimum at which one lucky streak stops defining the whole result.

Account for costs from day one, otherwise the report describes a different strategy. Example. A 1000 USDT deposit, a 2000 USDT position, entry and exit at market. At Binance base rates for a regular user on USDⓈ-M futures, the taker fee is 0.05% of position value, which is 1 USDT on entry and 1 USDT on exit. A hundred trades cost 200 USDT, which is 20% of the deposit. If the average trade in the test produced +0.1% of position value (2 USDT), a hundred trades give 200 USDT gross and leave zero after fees. At an average result of +0.3% (6 USDT), gross is 600 USDT and 400 USDT remains net. The numbers are illustrative and rates depend on the exchange and user tier, but the order of magnitude explains why frequent trades with a small average win do not survive.

Add slippage to the fees, that is the difference between the expected and the actual fill price. In a manual test it is usually simply absent, since you mark the entry at the perfect price. A sensible approach is to build a fixed adjustment in ticks into every trade and see whether the statistics survive it.

A separate limitation concerns the order book and the tape. A historical chart stores only prices and bar volume, so you cannot reconstruct from candles what the density level in the order book looked like at the moment of entry or how fast the tape was running. Ideas built on large limit orders, spoofing or a reaction to a big print cannot be verified by a backtest on history. A different scheme works for them. You watch the market in real time in a terminal with market depth and the tape, write down every scenario on the spot and build the sample forward. Secret Terminal shows the order book, the tape and clusters for the connected exchanges, and its built-in journal lets you record statistics in the same window. How to read that data is shown in a free lesson on analysis through the order book and clusters.

Replay Mode in TradingView

Bar Replay plays history back bar by bar, so you make decisions without knowing the next candle. The button sits on the top panel of the chart, next to the rewind icon. Once it is on, you choose the starting point by clicking the chart, through the "Select date" menu or with the random bar button, then start playback and adjust the speed. A separate button steps forward, and the jump-to-real-time button returns you to the current market. From the keyboard, playback starts and stops with Shift and the down arrow, and Shift with the right arrow steps forward.

Drawings made in Replay mode are kept after you exit, and indicators calculate on the replayed data, so rules based on indicators can be checked as well. The session is saved automatically, and next time you can continue from the same bar.

The depth of available history depends on the TradingView plan. Data as of October 7, 2026 from the official pricing page.

For BTC on TradingView, minute history starts in August 2011 and second history in August 2022. So a daily or four-hour model can be worked through on the free plan, while a scalping one-minute model has to be tested either on a paid plan or forward, in real time.

The mode has technical limits that are better known before you start.

• Cluster charts (Volume footprint), Renko, Kagi, Point & Figure, Range, Line Break and Time Price Opportunity are not replayed.
• Spread charts and tick-based intervals are not supported by the mode.
• A bar cannot be replayed in smaller segments.
• Orders in the Paper Trading simulator are filled at real quotes rather than replayed ones, so record test trades manually in your own table.
• New server-side alerts cannot be created during replay.

The main risk of the mode is not technical. You remember what happened to bitcoin last March, and that knowledge nudges your hand. The cure is picking a random bar, an unfamiliar instrument, and refusing to rewind after a losing trade.

What to Measure

The result of a test is not a single profit figure but a set of metrics, each answering its own question. Taken separately, they mislead easily.

Win Rate, R:R, Drawdown

Win rate alone proves nothing. A system with 70% winning trades loses money if the average loss is three times the average profit. Calculate expectancy with the formula: average result = win rate × average profit in R − (1 − win rate) × average loss in R. With a 40% win rate and an average profit of 2.5R against a 1R loss, you get 0.4 × 2.5 − 0.6 × 1 = +0.4R per trade. Such a system makes money even though more than half of its trades close in the red.

Maximum drawdown shows how much capital is lost in the worst streak. It is measured from the peak of the equity curve. A simple calculation gives the scale. Ten losing trades in a row at 2% risk per trade produce a drawdown of about 18% of the deposit, and returning to the previous mark takes a gain of roughly 22%. If such a streak already appeared in the test, it will appear in live trading too, and the only question is whether you will sit through it as planned.

Account separately for the funding rate if the strategy holds futures positions longer than a few hours. It is charged or credited on the exchange schedule, usually every eight hours, though the interval and the size depend on the venue and the contract. For day trading it is a minor item; for holding a position over several days it becomes a visible part of the result.

A sample of twenty trades proves nothing. A winning streak of that size happens to a losing strategy as well, and the other way round. The closer the win rate is to 50% and the smaller the edge per trade, the more trades you need to tell skill from luck.

Mistakes

Five mistakes show up more often than the rest and damage the result the most.

• Vague rules. Conditions open to different readings turn the test into confirmation of your own expectations.
• Looking ahead. Any decision made with the right side of the chart visible understates how hard the entry really was.
• Ignoring costs. Fees, slippage and funding rate eat the edge of small trades entirely.
• Curve fitting, that is tuning to history.
• Survivorship bias.

Curve Fitting

Fitting starts when parameters get adjusted until the equity curve looks beautiful. An indicator period of 14 gives a modest plus, 17 gives an excellent result, and 17 goes into the strategy. The problem is that you described a specific stretch of history with its randomness rather than a market regularity.

The signs are easy to spot. More than three or four parameters, an equity curve with almost no drawdowns, and a one-step shift of any parameter that drops the result into the red. A stable rule does not behave that way; at neighboring values it gives a similar result, even if slightly worse.

The cure is splitting the data. Half of the history is used for tuning, the second half stays untouched until the very end and the final version of the rules is checked on it. If the result falls apart on the untouched stretch, the idea failed the check, whatever it promised on the first half.

Survivorship Bias

Survivorship bias appears when the test runs only on assets that made it to today. In crypto the effect is especially strong. Coins get delisted, projects die, and the list of available pairs holds the winners. A "buy altcoin dips" strategy tested on the current top 50 will show an excellent result, because the projects that went to zero along the way never entered the sample.

The same trap applies to the period. A test on the 2023-2024 stretch of a rising market makes almost any long idea profitable. An honest check includes an extended decline and a range, where most signals turn out to be false.

The practical solution is simple. Include instruments that traded back then and are now illiquid or delisted, and look at the result by market phase separately, not only at the overall total.

FAQ

• How many trades does a backtest need?

  The guideline is 50-100 trades across different market phases. That is not statistical proof, just the minimum volume at which one lucky streak stops defining the outcome. The smaller the edge per trade, the larger the sample you need.

• Does a positive backtest guarantee profit in live trading?

  No. The test checks only the logic of the rules on past data. It does not include slippage at your size, execution delays, liquidity changes or your own behavior during a losing streak. A positive result is a reason to move to a forward test, not a reason to raise risk.

• How does a backtest differ from a demo account?

  A backtest runs the rules across ready history and gives statistics in an hour or two. A demo works in real time and is closer to the live market, but a simulation has no real money, and without it there is no real reaction to a loss. So after a demo the sensible step is trading the minimum amount you are prepared to lose, 10 USDT for example, allowing for the minimum order size and fees. There is no need to raise leverage to work around a small deposit; that changes the risk, not the quality of the check.

• Do you need a paid TradingView plan for Replay mode?

  On the free Basic plan, Replay works on the daily timeframe and higher. Replaying minute data requires a paid plan, the depth depends on its tier, and second data is available on Premium and above. Plan terms change from time to time, so check the current ones on the TradingView site.

• How do you test strategies based on the order book and the tape?

  Historical candles do not store the state of the order book or the tape, so a chart backtest does not work here. The statistics are collected forward, in real time. The trader watches the scenario in the terminal, records the condition, the entry, the price reaction and the result, and after a few weeks has a sample of their own.

• The backtest showed a loss, what now?

  First find out where exactly the money is lost. Break the trades down by market phase, by time of day, by direction. The minus often comes from one group of conditions while the rest of the rules work. If the loss is spread evenly, the idea was not confirmed, and that is a normal outcome of a check that just saved your deposit.

Checking an idea starts with the data you can see on screen. Download Secret Terminal from the official site, connect your exchange and build a workspace with the order book, the tape and clusters, so that trades and statistics are recorded in one window instead of three separate services.