Risk-Adjusted Return

Return measured relative to the risk taken to earn it, not in isolation. The honest way to compare a strategy against buy-and-hold — because two strategies with the same return can have wildly different drawdowns, and the one that suffered less pain is the better one.

In plain English

A raw return number answers "how much did it make?" A risk-adjusted return answers the more useful question: "how much did it make per unit of suffering?" Earning +100% by sitting through an 80% drawdown is a very different achievement from earning +100% with a worst dip of 15% — the second strategy is far better even though the headline return is identical.

This is the central correction to "everything loses to buy-and-hold." On raw return, holding wins in a bull market almost by definition. On a risk-adjusted basis, a strategy that captured most of the upside while cutting the drawdown often wins — and that is the honest comparison.

The two common risk-adjusted measures

• Sharpe ratio = mean return ÷ standard deviation of returns (return per unit of wobble). Unitless; comparable across strategies and leverages. See sharpe ratio and sharpe significance for when a Sharpe is real vs noise.
• Calmar ratio (return-over-max-drawdown) = total return ÷ max drawdown (return per unit of worst-case pain). The most intuitive for "could I survive this?" See drawdown.

Both divide the reward by a measure of the risk.

Why it matters for this fleet

The leverage cliff (§5) is the cleanest risk-adjusted lesson in the dossier. Leverage scales the same trades, so it adds return and drawdown in lockstep — meaning it buys no risk-adjusted edge. The median worst drawdown per leverage rung:

A 50× row that "out-printed" buy-and-hold did so by stacking −98%-class drawdown — un-survivable risk for a return that holding matched at 1× with a fraction of the pain. Read edge at 2×, where liquidation sits far away; read 50×+ as the liquidation-cliff finder, never as skill.

The gap this concept exposes (the open feature)

The buy-and-hold benchmark (buy and hold) is stored as a bare return scalar with no drawdown attached, so the public "beat buy-and-hold" comparison (alpha) is raw-return-vs-raw-return — risk-blind. Buy-and-hold earned SOL +2,405% by surviving a ~90%+ peak-to-trough drawdown. A strategy that trailed on return but dodged most of that drawdown "lost" the headline while being the more deployable choice. The engine already has buy-and-hold's full equity curve and already computes per-strategy max drawdown, so computing the benchmark's own drawdown and showing a risk-adjusted comparison beside raw alpha is low-cost — captured as SEED-025.

Examples from the live fleet

• id659 (1×) vs id523 (2×) — the same SOL 1h 21/50 long: identical 436 trades, but drawdown −5.1% at 1× vs −9.9% at 2×. Same trades, scaled pain — so their risk-adjusted return is essentially flat across the rungs. Leverage moved the risk, not the edge.
• id522 (ETH 4h 50× long) — the only edge-significant hold-beater, CAGR 116.9% — but paid for with a 30–61% drawdown. On raw return it "beat hold"; on a risk-adjusted basis the 50× amplification it leaned on is exactly what makes that return un-survivable in practice.

Related

• buy and hold
• alpha
• drawdown
• sharpe ratio
• sharpe significance
• risk of ruin
• leverage

Sources

• wiki/qa-sessions/2026-06-22-session.md#q1 (first formal entry)
• growth/content/dossiers/ema-cross/1-analysis.md §5 (the leverage cliff, run 83)
• apps/backend/src/routes/benchmark.ts (computeBenchmark — the curve already exists); SEED-025