How to Backtest CRT
JustWolves is an educational platform and is not a SEBI-registered investment advisor. Trading financial instruments involves significant risk of loss and may not be suitable for all investors. Past performance is not indicative of future results. Read our full disclaimer →
What you'll learn
- CRT is well suited to backtesting because its core rule (sweep and close back inside) is objective
- Write every rule before testing: C1 timeframe and filters, entry model, stop, targets, management and skips
- Use bar replay so you only see candles as they would have appeared
- Log every valid setup, including skips and missed entries, and measure results in R
- Check robustness: different periods, markets and small rule changes
In this article · 16 sections
CRT's strict candle rules make it one of the easier price action methods to test. Either C2 swept and closed back inside, or it didn't. That objectivity is only useful if the rest of the strategy is equally precise and the test is done without hindsight. This lesson walks through a manual CRT backtest from rules to results, and how to check whether the results are robust.
What You Will Learn
- How to write testable CRT rules
- How to run a manual backtest without hindsight
- What to log and how to measure results
- How to check robustness
Before You Start
Read CRT Stop-Loss, Targets and Invalidation and, from the Risk course, How to Backtest.
Step 1: Write the Rules
| Rule | Example |
|---|---|
| Market and timeframes | Nifty futures; 1-hour C1; 5-minute entries |
| Bias | Daily higher highs and lows for longs only |
| C1 filter | Range between 0.5× and 2× the 20-candle average range |
| C2 rule | Sweeps CRL and closes above it |
| Entry | Limit order at CRL plus 2 points during C3 |
| Stop and buffer | Below C2's low minus 10% of average range |
| Targets and management | Half at the midpoint, half at CRH; stop to breakeven after the midpoint |
| Skip rules | Scheduled major news during C3; target below 1R |
Step 2: Replay Without Hindsight
Manual backtest process
Step 1: Choose the period
In advance, including trends, ranges and volatile phases.
Step 2: Use bar replay
Reveal candles one at a time using your charting platform's bar replay.
Step 3: Mark as you go
C1 when it closes, C2 when it closes, entry during C3.
Step 4: Apply every rule
Including skips; never adjust rules mid-test.
Step 5: Log
Every valid setup, whether traded, skipped or missed.
Practice data: EUR/USD 4H
Loading historical candles…
Step 3: What to Log
- Date, time and direction
- CRH, CRL, C2 extreme and close
- Entry, stop and targets (or 'missed' or 'skipped' with the reason)
- Result in R, after estimated costs
- Context tags: trend alignment, level, session, news
Step 4: Measure Results
- Number of setups and how many were traded, skipped or missed.
- Win rate and average win and loss in R.
- Expectancy: (win rate × average win) − (loss rate × average loss).
- Maximum drawdown in R and the longest losing streak.
- Breakdown by tag: for example, trend-aligned versus counter-trend.
Look Beyond the Average
An average result in R is the headline, but it can hide a lot. Report at least these alongside it:
- The median trade. In our own CRT test, the average trade was about −0.07R but the median trade was about −0.35R. Most trades lose a little, and a minority of large winners pull the average up. A strategy like that feels much worse to trade than its average suggests.
- The largest single winner. If removing one or two trades changes the conclusion, the conclusion is not robust.
- Results by group. By market, by direction, by time of day. Our test found gold performing clearly worse than EUR/USD over the same period, which a single pooled average would hide.
- Win rate together with average win and loss. A 61% win rate can still lose money if the wins are small, as our test of strong C2 closes showed.
Step 5: Check Robustness
A result from one market and one period can be luck. Test the same rules on a different period and, if relevant, a different market. Then change one parameter slightly at a time (the C1 size filter, the buffer, the entry offset) and check that results don't collapse.
Rules that only work with one precise setting are likely overfitted. Rules that work reasonably across a range of settings are more likely to hold up live.
Estimating Costs
Include realistic costs for your market. For Nifty futures, that means brokerage, securities transaction tax, exchange charges, stamp duty and GST, plus slippage on stops. For forex, it means the spread and any commission. Subtract them from each trade's result before calculating expectancy.
CRT trades on low timeframes often have small targets, so costs can take a meaningful share of each win. A strategy that looks profitable before costs can turn negative after them, which is exactly what a backtest should reveal before you trade it.
How Big Are Costs, Really?
Costs feel small per trade but are large relative to a CRT stop. In our test, the median 4H CRT stop was about 10 pips on EUR/USD, 13 on GBP/USD and 15 on USD/JPY. A spread of 1–1.5 pips is therefore about 0.1R per trade, more than the raw pattern's average result. On gold, wider stops made the same kind of cost only about 0.02R. Express costs in R for your own market before you trust any result.
What We Learned Testing CRT Ourselves
While preparing this course, JustWolves backtested the CRT model on 845 real 4H setups across EUR/USD, GBP/USD, USD/JPY and gold (May–November 2026). The results are in What Is Candle Range Theory? and the lessons that follow it. Here we share what went wrong along the way, because every one of these problems will appear in your own testing too.
| Problem | What happened | What we did |
|---|---|---|
| Bad candle data | Free daily forex candles had open almost equal to close on about 90% of days, making every daily candle look like a doji. | Rebuilt daily and 4H candles from hourly data, and checked every dataset for odd bodies and gaps before testing. |
| Candle alignment | 4H and daily candles depend on the platform's time zone. Different alignments produce different C1, C2 and C3 candles. | Aligned all candles to the 5 pm New York close and stated it in the method. |
| Which came first? | Inside one candle, you cannot tell whether price hit the stop or the target first. | Resolved trades on hourly candles. When both fell in one hour, we counted a loss. |
| One trade distorting everything | A setup where C3 opened beyond C2's extreme left almost no risk and recorded a +152R "win". It turned one result from negative to positive. | Excluded setups invalidated before entry, applied that rule to every variation, and checked the largest trades in every test. |
| Bad ticks | A single gold print, 37 points away from surrounding prices, would have filled a limit order that probably never filled live. | Treated fills that depend on one isolated wick as suspect and checked them on a lower timeframe. |
| Costs | Spreads were about 0.1R per forex trade, larger than the raw average result. | Reported results before costs, stated so clearly, and estimated cost in R separately. |
The +152R trade deserves a second look, because it is the kind of error that produces impressive screenshots. On paper, one variation of our test, a slightly wider stop buffer, turned the whole CRT model from losing to winning. In reality, a single setup with a tiny stop accounted for almost all of the difference. If a result looks too good, find the trades responsible before you believe it.
Practical Example
Hypothetically, Deepa tests the rules above on six months of Nifty 1-hour data using bar replay. She logs 64 valid C2s: 41 traded, 15 skipped by her rules and 8 missed when the limit entry didn't fill.
She calculates expectancy after estimated costs, then repeats the test on a different six-month period and with the C1 filter loosened slightly. Results are similar in both, so she moves to a month of paper trading before risking money, and keeps logging the same fields.
Common Mistakes
- Testing on a finished chart instead of replaying.
- Changing rules mid-test.
- Logging only traded setups, not skips and misses.
- Ignoring costs.
- Small samples treated as conclusive.
Professional Insight
The most useful output of a backtest is often not the headline result but the breakdown: which contexts worked and which didn't. Experienced traders use that breakdown to simplify their rules, removing conditions that don't help, rather than adding more. Fewer, well-understood rules are easier to follow live and much easier to review.
Risk Considerations
- Past results don't guarantee future results.
- Manual tests are slow and still prone to bias.
- Overfitting produces strategies that fail live.
Important Terminology
| Term | Meaning |
|---|---|
| Bar replay | Revealing historical candles one at a time. |
| Expectancy | Average result per trade in R. |
| Robustness | Results holding across periods and small rule changes. |
| Overfitting | Rules tuned too closely to past data. |
| Tag | A label recording a setup's context. |
How do I backtest CRT?
Write objective rules, replay historical charts candle by candle, log every valid setup and its result in R, then check robustness.
Why is CRT easy to backtest?
Its core rule, a sweep beyond C1's range and a close back inside, is objective.
What should I log?
Levels, entry, stop, targets, result in R, and context tags, for traded, skipped and missed setups.
How many trades do I need to backtest CRT?
At least 50–100 setups per market and timeframe before drawing conclusions, and more if the average result is small. In our test of 845 setups, an average of −0.06R was still too close to zero to call the raw pattern clearly good or bad.
What comes after a backtest?
Robustness checks, then paper trading or very small live size.
Why do my CRT backtest results look better than my live trading?
Common reasons are testing on finished charts instead of replaying, skipping costs, counting fills that only happened on a single bad tick, and assuming the target was hit before the stop inside one candle. Each of these inflates backtest results.
Should I include costs when I backtest CRT?
Yes. Deduct a realistic spread, commission and some slippage from every trade. CRT stops are often small, so even a one-pip cost can be a large share of the risk on each trade. A strategy that only works before costs is not a strategy you can trade.
Practical Exercise
Run your first CRT backtest.
Your task
Step 1: Write the rules
Using the table in this lesson.
Step 2: Replay three months
On one market and timeframe.
Step 3: Log everything
Including skips and misses.
Step 4: Calculate
Expectancy, drawdown and results by tag.
Check Your Understanding
Bar replay helps avoid:
Related Lessons
Previous: CRT Stop-Loss, Targets and Invalidation. Next: Common CRT Mistakes. See also How to Backtest.