Performance Metrics Cheat Sheet

A single-page reference to the metrics used to evaluate a trading system, each with its formula, what it actually measures, and a caveated rule-of-thumb interpretation.

Metrics: Performance metrics fall into three families: return metrics (CAGR, expectancy, average win‑loss), risk-adjusted return metrics (Sharpe, Sortino, Calmar), and risk/consistency metrics (max drawdown, volatility, profit factor, win rate, R‑multiple, recovery factor). No single number describes a strategy — CAGR without drawdown is meaningless, and a high win rate can still lose money if the losers are large. The “rough interpretation” bands below are widely cited heuristics for reasoning about a backtest, not targets to optimise toward and never guarantees of future results; any of them can be inflated by overfitting or an unrealistic backtest.

This cheat sheet lists the metrics that matter when judging a systematic strategy. Read it alongside the caveats: every figure here is only as trustworthy as the backtest that produced it. A metric computed on curve-fit or look-ahead-contaminated data is worse than no metric, because it looks authoritative. See The Backtesting Checklist before trusting any number below.

How to read a metric

Group the metrics by the question they answer. “How much did it make?” is a return metric. “How much did it make per unit of risk?” is a risk-adjusted metric. “How much pain to get there?” is a drawdown or volatility metric. A serious evaluation quotes at least one from each group, plus the number of trades — a Sharpe of 3 over 12 trades is noise, not skill.

These bands are heuristics, not targets

The “rough interpretation” column repeats numbers commonly used by practitioners to sanity-check a backtest. They are context-dependent (a market-making book and a swing system live in different regimes), easy to inflate with overfitting, and say nothing about the future. Never treat a band as a goal to optimise toward, and never present any of these as achievable returns.

Return metrics

Metric	Formula	What it tells you	Rough interpretation (heuristic only)
CAGR	CAGR = (End / Start)^(1 / years) − 1	The smoothed annual growth rate of equity, ignoring the path taken. Compounds; comparable across time horizons.	Only meaningful next to max drawdown. A CAGR quoted alone tells you nothing about survivability.
Expectancy	E = (Win% × AvgWin) − (Loss% × AvgLoss)	Average profit or loss per trade in currency (or in R). The engine of the equity curve.	Must be positive after costs. Small positive expectancy × many trades is a viable system; negative expectancy cannot be fixed by sizing.
Average win / loss	AvgWin = Σ wins / #wins; AvgLoss = Σ losses / #losses	Typical size of a winner versus a loser. Feeds expectancy and the win/loss ratio.	Trend systems: large AvgWin, low win rate. Mean-reversion: small AvgWin, high win rate. Neither is “better” in isolation.
Win rate	Win% = winning trades / total trades	How often the system is right. Says nothing about how much it makes when right or wrong.	A high win rate with a poor payoff ratio can still be a losing system. Judge with average win/loss, not alone.

Risk-adjusted return metrics

Metric	Formula	What it tells you	Rough interpretation (heuristic only)
Sharpe ratio	Sharpe = (Rp − Rf) / σp, annualised × √periods	Excess return per unit of total volatility. The standard risk-adjusted yardstick.	Practitioners often eye <1 as weak, ~1–2 as decent, >2 as strong — but Sharpe is easily inflated by overfitting, non-normal returns, and short samples.
Sortino ratio	Sortino = (Rp − Rf) / σdown	Like Sharpe but divides only by downside deviation, so it does not penalise upside volatility.	Higher than Sharpe for right-skewed strategies. Compare Sortino to Sharpe to see how much volatility is “good” (upside).
Calmar ratio	Calmar = CAGR / \|Max Drawdown\|	Return earned per unit of worst-case peak-to-trough loss. Popular with managed-futures/CTA evaluation.	Usually computed over ~3 years. Rewards systems that avoid deep drawdowns; sensitive to the single worst drawdown in the sample.

Annualising Sharpe

Sharpe computed on daily returns is annualised by multiplying by √252 (trading days); on monthly returns by √12. Mixing frequencies, or annualising a ratio built from autocorrelated returns, is a common way to report a misleadingly high number.

Risk and consistency metrics

Metric	Formula	What it tells you	Rough interpretation (heuristic only)
Max drawdown	MaxDD = max over t of (Peakₜ − Equityₜ) / Peakₜ	The largest peak-to-trough equity decline. The single most important survivability number.	Determines whether you (or your capital) can stay in the system. A strategy you exit at the drawdown low has no CAGR.
Volatility	σ = std dev of period returns, annualised × √periods	Dispersion of returns; the denominator of Sharpe. A proxy for how bumpy the ride is.	Not the same as risk-of-loss — volatility treats upside and downside symmetrically. Pair with drawdown.
Profit factor	PF = gross profit / \|gross loss\|	Rupees won per rupee lost across all trades. A compact summary of edge.	PF around 1.0 is break-even before costs; practitioners often view ~1.3–1.6 as workable. Very high PF on few trades usually signals overfitting.
R-multiple	R = trade P&L / initial risk (1R)	Expresses each trade in units of the amount risked, making outcomes comparable across position sizes.	A “+3R” winner returned three times the risk. Expectancy in R (E = mean R per trade) is the size-independent edge.
Recovery factor	RF = net profit / \|max drawdown\|	How many times the strategy earned back its worst drawdown over the test.	Higher means drawdowns were recovered quickly relative to profit. Like Calmar, dominated by the single worst drawdown.

Common ways these numbers lie

No cost model. Brokerage, STT, exchange fees, and slippage can turn a positive expectancy negative. A frictionless backtest overstates every return metric.
Too few trades. Sharpe, win rate, and profit factor need a meaningful sample. On 20 trades they are noise.
Single-path drawdown. Max drawdown is one realised path. Monte Carlo reshuffling shows the drawdown you could have seen, which is usually worse.
In-sample only. Metrics computed on the data you optimised on are upper bounds. Trust out-of-sample and forward-test figures more.
Non-normal returns. Sharpe assumes roughly Gaussian returns; strategies that sell tail risk (e.g. short-option books) can show a lovely Sharpe right up to the loss that ends them.

Related concepts

See Maximum drawdown, Position sizing, and the Risk Management Formulas reference for the calculations that turn these metrics into decisions.

Frequently asked questions

Which single metric best summarises a strategy?

There isn't one. The most defensible minimal set is CAGR paired with maximum drawdown (or the Calmar ratio that combines them), plus Sharpe or Sortino for risk-adjustment and the trade count for statistical weight. Any lone number can be gamed.

What is the difference between Sharpe and Sortino?

Both divide excess return by a measure of variability. Sharpe uses total standard deviation, penalising upside and downside swings equally. Sortino uses only downside deviation, so it rewards strategies whose volatility is mostly to the upside. Comparing the two indicates how skewed the returns are.

Why is max drawdown so important?

Drawdown is what actually forces traders and allocators to abandon a system. A strategy with excellent long-run CAGR is worthless if its worst drawdown exceeds your tolerance, because you will exit at the low and never realise the recovery. It defines survivability.

Is a high win rate good?

Not by itself. Win rate ignores the size of wins and losses. A system that wins 80% of the time but loses far more on its 20% losers can have negative expectancy. Always read win rate together with average win, average loss, and profit factor.

What is an R-multiple?

An R-multiple expresses a trade's outcome as a multiple of the amount initially risked on it. If you risked 5,000 rupees (1R) and made 15,000, the trade was +3R. Because it normalises for position size, R lets you compare trades and compute a size-independent expectancy.

What profit factor is acceptable?

Profit factor of 1.0 is break-even before costs. Many practitioners look for something in the 1.3 to 1.6 range as workable on a realistic, cost-inclusive backtest with enough trades. A profit factor of 3+ on a few dozen trades usually indicates curve-fitting rather than edge.

How do I annualise the Sharpe ratio?

Multiply the per-period Sharpe by the square root of the number of periods in a year: √252 for daily returns, √52 for weekly, √12 for monthly. This assumes returns are roughly independent across periods; autocorrelation makes the annualised figure overstated.

What is the recovery factor?

Recovery factor is net profit divided by the absolute maximum drawdown. It tells you how many times over the strategy earned back its deepest decline during the test. Higher is better, but like the Calmar ratio it is dominated by the single worst drawdown observed.

Can these interpretation bands be treated as targets?

No. The bands are heuristics for sanity-checking a backtest, not objectives. Optimising a strategy to hit a Sharpe or profit-factor band is a direct route to overfitting, and none of these figures predict future results.

Last reviewed 11 July 2026. Educational content only — not investment advice.