Return & risk metrics

Returns, volatility, drawdown, hit rate, P&L per trade. The basic vocabulary, and why high-frequency returns need event-aware, fat-tail-aware versions of all of them.

The idea

What is an equity curve and what does it actually tell you?

An equity curve is account value plotted over time: the cumulative sum of every trade's P&L. On its own it tells you almost nothing: its height is set by leverage and starting capital, both arbitrary. What matters is the shape (how smooth it is, how it draws down, how reliably it recovers), which is exactly what the metrics below quantify.

Two strategies can finish at the same point having travelled completely different paths: one a smooth diagonal, the other a violent saw-tooth that spent months underwater. The destination (total return) is identical; the experience, the risk and the fundability are not. Every metric on this page is a way of measuring the path, not the destination.

The honest warning that motivates the whole topic: the equity curve is the most seductive and least informative chart in trading. It is trivial to make one go up in a backtest; see backtesting and simulation. The discipline is refusing to be impressed by the line until you have seen its volatility, its drawdown and its decomposition.

How do you measure return and volatility?

Return is the mean of the return stream over a period (often annualised). Volatility is its standard deviation, the dispersion of period returns around that mean. Return without volatility is meaningless, because leverage scales both together: only their ratio (the next page's Sharpe) is invariant. Always report the pair, never the return alone.

The plain-English versions: mean return asks "on average, what did one period make?"; volatility asks "how much does the return bounce around its average?". To compare across strategies you annualise, and this is where most reported numbers start to lie. The mean scales linearly with the number of periods, but volatility scales with the square root of time.

The key reason both are reported together: leverage scales mean and volatility identically. Lever a strategy 2× and both double; the equity curve looks twice as exciting and is exactly as risky per pound. So a return quoted without its volatility is a leverage-dependent number with no information content. This is the single most repeated point in performance measurement, and the reason the risk-adjusted ratios page exists.

What are hit rate and win rate, and are they the same thing?

Hit rate (used interchangeably with win rate here) is the fraction of trades that make money: winners divided by total trades. It is necessary but wildly insufficient: a 90% hit rate can lose money and a 35% hit rate can be a fortune. Hit rate is only meaningful paired with the payoff ratio (average win size divided by average loss size).

Some desks reserve "win rate" for P&L-positive days and "hit rate" for individual trades, but the arithmetic is identical: just state your unit (per trade, per day) and be consistent. The trap is that hit rate is the most over-quoted, most misleading single statistic in retail trading. A martingale-flavoured strategy that takes tiny profits and lets losers run can post a 95% hit rate and a catastrophic expectancy, because the rare losses dwarf the frequent wins. A trend follower can be right 35% of the time and compound beautifully, because its winners are multiples of its losers.

In an HFT context, high-frequency strategies typically sit at modest hit rates per trade (often only slightly above 50%) but make the edge up in volume and a controlled payoff profile: thousands of near-coin-flip bets with a small positive expectancy each. The number that matters is expectancy per trade, defined next, not the hit rate in isolation.

What is the payoff ratio, and how does it trade off against hit rate?

The payoff ratio is the average winning trade divided by the average losing trade (in absolute size). Together with hit rate it determines whether a strategy makes money: expectancy per trade = hit rate × average win − (1 − hit rate) × average loss. A strategy is break-even when the payoff ratio equals $(1-p)/p$.

There are two ways to make money: be right often with small wins, or be right rarely with big wins. The payoff ratio measures how big your wins are relative to your losses; the hit rate measures how often you win. They are substitutes: you can buy a lower hit rate with a higher payoff ratio and vice versa, and the break-even line is the exact exchange rate between them.

The break-even frontier is the curve $b = (1-p)/p$ in (hit rate, payoff ratio) space: anything above it is profitable, anything below loses. At $p = 0.5$ you need $b \gt 1$ (wins bigger than losses); at $p = 0.35$ you need $b \gt 1.86$; at $p = 0.9$ you can survive with $b$ as low as 0.11. This single curve dissolves almost every "what win rate do I need?" argument.

The cost connection closes it: every real trade pays the cost stack (spread, impact, fees), which shifts the frontier up, so you need a higher payoff or hit rate to clear costs. A strategy that is break-even gross is loss-making net. That is the gross-versus-net point of the costs topic, viewed through the win/loss lens.

What is drawdown, and why measure both its depth and its duration?

A drawdown is the drop from a previous equity peak to a subsequent trough: how far underwater the strategy went. Maximum drawdown is the worst such drop. But depth alone is half the story: drawdown duration (how long you stayed below the prior peak) is what actually ends strategies, because capital and conviction run out in time, not in percent.

Drawdown duration has two components: the time-to-trough (peak to the bottom) and the time-to-recovery (trough back to a new high); together they give the underwater period, the total time spent below the old peak. A strategy can have a shallow −8% MDD that took fourteen months to recover: operationally fatal, because no allocator or solo trader keeps funding a flat-to-down book for over a year. Depth tells you how much it hurt; duration tells you whether you would still be in the seat when it healed.

HFT cares differently. A high-Sharpe HFT book has shallow drawdowns by construction (thousands of small bets), so MDD-in-percent looks trivial. The operationally relevant drawdown for HFT is usually measured in days of no edge or the worst run of losing days, and it feeds the kill-switch and risk-measurement logic: at what loss do you turn the strategy off? Drawdown is also the risk denominator of the Calmar ratio (annualised return divided by absolute MDD), which is why it belongs in this raw-measurements page even though it is consumed next door.

Worked example

A synthetic intraday mean-reversion strategy, 100 trades over one notional month, as of 2026; reproduce it from the trade counts. It runs 58 winners and 42 losers, so the hit rate is $p = 0.58$. The average win is $W = \pounds 120$ and the average loss $L = \pounds 150$, a payoff ratio $b = 0.80$: wins are smaller than losses, typical of mean reversion (many small wins, occasional larger losses when the reversion fails).

Over 100 trades, gross P&L is $+\pounds 660$. Now apply costs: say each round trip costs £8 all-in (spread, impact and fees from implicit costs). Net expectancy is $6.60 - 8.00 = -\pounds 1.40$ per trade, so the £660 gross becomes −£140 net over the month. This is the topic's recurring lesson in one trade ledger: real gross, fake net.

And the drawdown. Suppose the worst stretch was a run of 9 losing trades that took equity from a peak of £820 down to £370, an MDD of −£450, or −55% of peak equity, recovered 31 trades later. Depth (−55%) says the strategy is un-leverageable as configured; duration (31 trades) says you would have spent a third of the month underwater questioning it. A positive gross expectancy, a hit rate that clears its own break-even bar, and an equity curve that ends up, yet the strategy loses money net and carries a drawdown that rules out leverage. None of those failures is visible from the equity-curve height alone. That is why the raw metrics exist. All numbers synthetic.