Trading Performance Metrics Guide

At a glance

Most cited: Sharpe ratio; also the most misleading for skewed strategies
Most practical: Expectancy in R, plus drawdown depth and duration
Most neglected: Time underwater and return concentration
Rule: Never judge a strategy on a single metric

Key takeaways

Sharpe penalises upside volatility equally with downside, which systematically favours strategies that sell insurance and disfavours trend following.
Expectancy in R multiplied by trade frequency is the most direct measure of what a strategy produces.
Drawdown duration predicts abandonment better than drawdown depth, and it is rarely reported.
Return concentration matters: if the top five trades produced the entire result, the strategy is far more fragile than the average suggests.
Metrics computed over a short sample are estimates with wide error bars, and should be reported as ranges rather than as facts.

The main metrics and what each misses

Metric	Formula	What it measures	What it hides
Expectancy (R)	(Win% x AvgWin) - (Loss% x AvgLoss)	Average profit per trade in risk units	Frequency; variance; tail events
Profit factor	Gross profit / gross loss	Ratio of wins to losses in aggregate	Distribution; one large win can dominate
Sharpe ratio	(Return - risk free) / volatility	Return per unit of total volatility	Skewness; penalises upside volatility
Sortino ratio	Excess return / downside deviation	Return per unit of downside volatility	Still assumes a stable distribution
Calmar ratio	Annual return / max drawdown	Return per unit of worst loss	Depends entirely on one historical event
Maximum drawdown	Largest peak-to-trough decline	Worst historical loss	Duration; and it is one sample path
Time underwater	Longest period below a prior peak	How long you would have endured	Depth
Win rate	Winners / total trades	Frequency of success	Everything about size; meaningless alone
Ulcer index	RMS of drawdown depths over time	Combined depth and duration of declines	Less intuitive; rarely reported

The problem with the Sharpe ratio

Sharpe divides excess return by standard deviation, treating all variability as risk. That is reasonable for symmetric return distributions and actively misleading for the skewed distributions that most trading strategies produce.

Strategy A: premium selling
   Monthly returns: +1.2% eleven times, -8% once
   Annual return    +5.2%
   Volatility       9.1%
   Sharpe           about 0.4
   Skew             strongly negative
   Real risk        a single event can exceed the annual gain

Strategy B: trend following
   Monthly returns: mostly -0.5% to +0.5%, two months +7%
   Annual return    +11%
   Volatility       13%
   Sharpe           about 0.7
   Skew             strongly positive
   Real risk        long flat periods, no catastrophic event

Sharpe penalises Strategy B for its large positive months
and treats Strategy A's steady accrual as low risk,
which inverts the actual tail risk ranking.

Two strategies with very different risks and similar Sharpe ratios.

What to report instead

1
Expectancy in R and trades per year
Together these give the annual edge in risk units, which is the most direct measure of what the strategy produces.
2
The full trade distribution
Not just the average. Show the histogram, the largest win and loss, and how much of the total came from the top five trades.
3
Drawdown depth and duration
Both the maximum and the distribution from a Monte Carlo resampling, plus the longest time underwater.
4
Performance by year and by regime
A single exceptional year hidden in an average is the most common way a mediocre strategy looks good.
5
Skewness and the worst month
These reveal whether the strategy is long or short tail risk, which Sharpe conceals.
6
Sensitivity to costs and start date
Re-run with double the assumed costs, and shift the window by a few months. Report both results.

Statistical significance of a track record

t-statistic = (mean trade return / std dev) x sqrt(N)

For a t-statistic of 2 (roughly 95% confidence):
   N required = (2 x std dev / mean)^2

Strategy with mean 0.3R and std dev 1.2R per trade:
   N = (2 x 1.2 / 0.3)^2 = 64 trades

Strategy with mean 0.1R and std dev 1.5R per trade:
   N = (2 x 1.5 / 0.1)^2 = 900 trades

The weaker the edge relative to its variance, the longer
the record needed.  And this assumes only ONE strategy
was tested.  Testing 50 variants requires a far higher
threshold to claim significance.

How long a track record needs to be to distinguish skill from luck.

This calculation explains why three profitable months proves nothing and why even three profitable years may not be conclusive for a low-frequency strategy. It also explains why testing across many instruments is so valuable: it multiplies the effective sample without requiring more calendar time.

Which metrics to prioritise by strategy type

Strategy type	Primary metrics	Metrics that mislead
Trend following	Expectancy, time underwater, return concentration	Win rate; Sharpe understates it
Mean reversion	Worst single trade, skewness, tail risk	Win rate and Sharpe both flatter it
Premium selling	Worst month, skewness, maximum loss	Sharpe and win rate severely flatter it
Market neutral	Correlation to market, drawdown, leverage used	Absolute return without leverage context
High frequency	Cost ratio, capacity, latency sensitivity	Backtested returns before impact
Long-term allocation	Drawdown duration, real return after inflation	Short-period Sharpe

Frequently asked questions

What is a good Sharpe ratio for a trading strategy?

For an individual trader, a live Sharpe of 0.5 to 1.0 is respectable and 1.5 or above is excellent. Backtested Sharpe ratios above 2 usually indicate a bias rather than an exceptional edge. Always report skewness alongside, because a high Sharpe with negative skew describes hidden tail risk rather than low risk.

Is profit factor a useful metric?

It is intuitive and reasonable as a summary: above about 1.3 suggests a meaningful edge. Its weakness is that it is dominated by outliers, so a single very large winner can produce an attractive figure from a strategy that otherwise loses. Always check it alongside the trade distribution.

Why does win rate matter so little?

Because it says nothing about size. A 90 percent win rate with losses ten times the size of wins is a losing strategy, and a 35 percent win rate with wins three times the size of losses is a good one. Expectancy combines both and is the number that determines returns.

How many trades do I need before my metrics mean anything?

It depends on the ratio of your edge to its variance. A strong edge may be distinguishable in 60 to 100 trades; a weak one may need several hundred. The t-statistic calculation gives a specific answer for your numbers, and it should be computed before drawing conclusions from a track record.

What metric best predicts whether I will stick with a strategy?

Time underwater, which is rarely reported. Traders abandon strategies during long flat periods more often than during sharp drawdowns, because there is no visible progress and no way to distinguish a normal stretch from a broken edge. Knowing your strategy has historically spent, say, 18 months underwater is essential preparation.

Test this idea before you trade it

Describe the rules in plain language and AlgoTrader AI turns them into a structured strategy blueprint with a configurable historical backtest, cost assumptions, and exportable code.

Build a backtest

Keep reading

Referenced by

Educational use only. This guide explains how a strategy works. It is not investment advice, not a recommendation, and no result described here is a forecast. Test any approach on historical and out-of-sample data, size positions conservatively, and never risk money you cannot afford to lose.