Framework

Read the numbers as five groups, not one good figure

The starting point for how to read backtest results is to stop grading metrics one at a time. A single impression—large net profit, high win rate—misleads readers easily. In practice, you organise the many available metrics into the following five groups and read the whole picture as a balance between them.

• Return: net profit, gross profit and gross loss, profit factor, expectancy. The direction and size of what was earned.
• Efficiency: win rate, average win, average loss, payoff ratio. The quality of a single trade.
• Downside risk: maximum drawdown, recovery period, Calmar, Ulcer. How far the equity falls.
• Stability: Sharpe, Sortino, SQN, subperiod consistency. Dispersion and repeatability.
• Data sufficiency: trade count, confidence intervals, distribution, outliers. How far the conclusion can be trusted.

For the wider picture, the complete TradingView backtesting workflow guide sets out the full context. This article zooms in on one part of it—how to read the numbers. Start by mapping the five groups and their metrics.

Premises

Fix the assumptions before you read any number

The same metric means something completely different under different assumptions. Before you begin reading numbers, confirm at least the following seven items. Comparing two backtest results whose premises are not aligned—by metric size alone—produces unstable conclusions.

• Period: from when to when, and which regimes—uptrend, downtrend, range—it covers.
• Symbol: the instrument, and the nature of its volatility and liquidity.
• Timeframe: one-minute versus daily changes both trade count and distribution.
• Initial capital: the reference point for interpreting drawdown and risk of ruin.
• Position sizing: fixed lot or compounding—each shifts how the metrics look.
• Costs: whether spread, commission and slippage are reflected.
• Trade count: the sample size that governs how much to trust the conclusion.

For preparing the source data, see exporting the Strategy Tester trade list to CSV; for building in costs, see how to model spread, commission and slippage. Gross figures that ignore costs, in particular, tend to diverge sharply from live performance.

Profitability

Net profit, gross profit and gross loss

Net profit is the final change in balance after gross loss is subtracted from gross profit. First split it into gross profit and gross loss, then separate the figure before costs (gross) from the one after costs (net). Looking only at the single net-profit number hides the bias buried inside its components.

For example, even a large net profit is doubtful in repeatability if most of it comes from one or two outsized winners. Conversely, subtracting costs can flip net profit from positive to negative, so the larger the gap between gross and net, the more the cost assumptions decide the conclusion. Always decompose net profit into gross profit, gross loss and distribution before reading it.

Efficiency

Win rate, average win, average loss, payoff ratio and expectancy

The efficiency group measures the quality of a single trade. Win rate is the share of winning trades; payoff ratio is the average win divided by the average loss. Most important is the figure that unites them—expectancy, the average profit or loss per trade.

The key fact here is that win rate alone cannot rank strategies. Even a high win rate can produce negative expectancy when the average loss greatly exceeds the average win. Conversely, a win rate below 40% can still yield positive expectancy when the payoff ratio is high enough. Plotting win rate against payoff ratio on a 2×2 grid reveals a strategy’s character.

This relationship is why a strategy with both a conveniently high win rate and a high payoff ratio warrants caution about over-optimisation. For more, see detecting overfitting through parameter stability.

Profit Factor

Profit factor: meaning and pitfalls

Profit factor (PF) is gross profit divided by gross loss. It is widely used as a summary of profitability: above 1 means a cumulative gain, below 1 a cumulative loss.

Useful as it is, PF has several pitfalls. First, it is unstable on a small sample—one large winner sends it soaring and one large loser makes it crash. Second, it is sensitive to outliers: a PF propped up by a single outsized winner can have low repeatability. Third, no universal pass line exists. Do not accept a fixed rule such as “PF above X is good”; always read it alongside trade count, drawdown and distribution. PF is only one input for comparison.

Risk-adjusted

Sharpe, Sortino and SQN (risk-adjusted metrics)

Risk-adjusted metrics measure efficiency by dividing return by dispersion (risk). The Sharpe ratio divides return by standard deviation; the Sortino ratio divides by downside deviation. SQN (System Quality Number) combines expectancy, dispersion and trade count.

These are powerful, but they depend heavily on assumptions. The calculation frequency (daily or per trade), the risk-free-rate setting and the sample period all change the value, and even the annualisation method shifts how it looks. A higher Sharpe is not always better, and with few trades it can swing high or low. Interpret it from several angles—alongside Sortino, SQN and the actual return distribution.

Downside

Drawdown and recovery (max DD, Calmar, Recovery, Ulcer)

The downside-risk group looks at how far the equity curve falls from its peak. Maximum drawdown (max DD) is the largest such decline in the past data. Related measures include Calmar—annualised return divided by max DD—Recovery Factor—net profit divided by max DD—and the Ulcer Index, which evaluates both the depth and the duration of declines.

The point to remember is that max DD is “the worst so far,” not “the ceiling from here.” The future can exceed it. There is no single answer to what percentage is acceptable—it changes with capital and objectives. The length of the recovery period matters too: for the same depth, a strategy that takes longer to recover imposes a heavier psychological and financial burden. For detailed drawdown calculation, recovery period and how to read Calmar and Ulcer, see maximum drawdown explained. For sequence risk and the risk of ruin from losing streaks, Monte Carlo risk-of-ruin testing is a useful companion.

Sample

Trade count, confidence intervals, distribution and subperiods

However good the metrics look, a small sample makes the conclusion untrustworthy. With few trades, the confidence interval around each metric widens, and you mistake a lucky sequence for skill. In a sample of a few dozen, one large winner or losing streak dominates the whole view.

• Trade count: the sample size that sets confidence. If small, keep the reading provisional.
• Confidence interval: win rate and PF carry a range—do not treat a point estimate as precise.
• Distribution and outliers: look beyond the average at the spread and the extreme values.
• Subperiods: check whether the tendency holds when the full period is split.

Whether the edge survives when you split the period into halves—or finer slices—is a clue to overfitting. The idea of validating across time slices is covered in depth in walk-forward and out-of-sample analysis. Read numbers with an eye to “how far can I trust this,” not “how many decimals of precision.”

Worked example

Read the metrics together across three example strategies

Metrics only gain meaning in combination. Compare three fictional strategies—A, B and C—whose assumptions (period, symbol, timeframe, initial capital and cost conditions) are aligned. Every number below is fictional, illustrative data, and all profit-and-loss figures are stated in account-currency units.

Comparing profit factor with the mini-bars, Strategy C is highest and Strategy B sits below 1.

The reading points are as follows. Strategy B has the highest win rate at 61%, yet its average loss exceeds its average win, so expectancy is slightly negative and PF dips below 1—the classic case you would miss by watching win rate alone. Strategy A has a low 38% win rate, but the 2.40 payoff ratio drives the highest expectancy and net profit; because the win rate is low, though, it must withstand losing streaks. Strategy C balances expectancy, PF and max DD well, but with only 100 trades the confidence in its conclusion needs a caveat. For an integrated view of several strategies under aligned assumptions, see multi-strategy portfolio backtesting.

Dependency

Why you must not judge a metric in isolation

As we have seen, the metrics depend on one another. Win rate gains meaning only with payoff ratio, PF only with trade count, and Sharpe only with the assumptions and the sample. Pulling out one number to pass or fail a strategy is the reading most likely to mislead.

In short, rather than finding one good number and stopping there, the heart of how to read backtest results is a mindset that scans across all five groups to ask, “where is the weakness?”

Self-check

Self-check calculator (win rate, payoff ratio, expectancy, PF)

Enter wins, losses, average win and average loss, and the tool estimates win rate, payoff ratio, expectancy and profit factor for educational purposes. Values are computed only in your browser and are never sent anywhere. First confirm how the formulas move with the worked example (Table 2) below, then swap in your own numbers in the tool beneath it.

Checklist

Practical checklist and summary

Finally, here is a practical checklist for how to read backtest results. The conclusion is simple: do not hunt for one good number—scan across the five groups for weaknesses.

• Have you aligned the assumptions (period, symbol, timeframe, initial capital, sizing, costs, trade count)?
• Have you decomposed net profit into gross/net, gross profit/loss and distribution?
• Have you read payoff ratio and expectancy together, not just win rate?
• Have you read PF and Sharpe with the backing of trade count and assumptions?
• Have you checked the shape of downside risk with maximum drawdown and recovery period?
• Have you estimated confidence from trade count, confidence intervals and subperiods?
• Have you judged relatively, without taking any universal pass threshold at face value?

When you want to check these on your own trade history, the next step is validation with real data. Once you need saving, comparison, KPI-reliability assessment or deeper dives into return distributions and rolling statistics, the Pro features help. Start free with the basic analysis first.

Related

Related reading and read next