Trading statistics that matter — and how each one lies alone

Win rate

The fraction of trades that made money. It is the most quoted number and the least informative: a 70% win rate with small wins and large losses loses money; a 35% win rate with 3R winners makes it. On its own it answers only "how often am I right?", which is not the same question as "am I profitable?". It is also the number most abused by small samples — see the interval table below.

Expectancy

What one trade is worth on average, in R: (win rate × average win) − (loss rate × average loss). This is the number that says whether a method makes money. Positive expectancy over a large sample is the only thing a trading record needs to show; everything else describes how it makes it. Expectancy is also the baseline against which any edge is measured (how to find your edge).

Trader A: 60% wins, avg win +0.8R, avg loss −1.4R → 0.6×0.8 − 0.4×1.4 = −0.08R per trade
Trader B: 40% wins, avg win +2.2R, avg loss −1.0R → 0.4×2.2 − 0.6×1.0 = +0.28R per trade

Profit factor

Gross profit divided by gross loss. A profit factor of 1.0 is break-even; 1.5 means you made $1.50 for every $1 lost. It is a good summary of efficiency and a poor one of sustainability: a factor of 3.0 over eleven trades is a coin that came up heads, and a factor of 1.3 over four hundred trades is a business. Read it with the sample.

Maximum drawdown

The largest peak-to-trough decline in your equity, in R or percent. It answers the question expectancy cannot: how bad did it get on the way? Two records with identical expectancy can have drawdowns of 5R and 25R, and only one of those is survivable at a given risk per trade. Drawdown is also what prop-firm rules are written about (the rules explained). Always note how many trades the drawdown lasted as well as how deep it went — a long shallow one is harder to sit through than a short deep one.

Sharpe and Sortino

Both divide average return by a measure of its variability. Sharpe uses the standard deviation of all returns; Sortino uses only the downside deviation, so it does not penalise large winning trades. A high ratio means the returns were steady relative to their size. For a discretionary trader the useful reading is relative: a rising Sharpe over successive months means the method is getting more consistent, a falling one means it is getting lumpier, and a ratio that is high only because of one huge trade is a warning rather than a compliment.

SQN — system quality number

SQN scales expectancy by its consistency and the sample size: (mean R ÷ standard deviation of R) × √(number of trades). It rewards records that make money predictably, and it explicitly grows with sample size, so a method needs both an edge and enough trades to score well. Rough bands are often quoted — around 2 is tradeable, 3 is good, 5 is excellent — but treat them as orientation, not judgement. What SQN is genuinely good at is comparing two versions of the same method over similar samples.

Kelly fraction

Kelly is the bet size, as a fraction of capital, that maximises long-run growth if your win rate and payoff are exactly what your record says. They never are, and full Kelly produces brutal drawdowns, so it is used as a ceiling: traders who calculate it typically risk a quarter to a half of it. Its real value is as a sanity check — if Kelly says 6% and you are risking 3%, you are aggressive; if it says 0.4% and you are risking 1%, your record does not support your size.

Monte Carlo: your record, reshuffled

Your equity curve is one ordering of your trades. Shuffle those same trades thousands of times and you get a distribution of the curves you could have had with the same results in a different order — and therefore a distribution of drawdowns and a probability of ruin (hitting a loss you could not continue from) at a given risk per trade. This is the number that should set your position size. If 5% of resamples ruin you at 1% risk, that is the risk you are running, whatever the actual curve looked like.

The caveat every number needs

All of these are estimates from a sample, and a sample of trades is usually small. A win rate of 58% from 30 trades has a 95% confidence interval of roughly 41% to 74% — wide enough that the true rate could be losing. Every rate you read should come with its interval, and every average with its sample size. When a tool prints a win rate as "58%" with nothing next to it, it is presenting a guess as a fact.

Reading them together

1. 1Is expectancy positive, and is the sample large enough that its interval clears zero?
2. 2Is the drawdown survivable at my risk per trade — and what does Monte Carlo say about the drawdowns I have not had yet?
3. 3Is profit factor stable across months, or carried by a few trades?
4. 4Are Sharpe/Sortino/SQN improving as the sample grows, or decaying?
5. 5Does my actual risk per trade sit well under the Kelly ceiling?