Reading an Equity Curve: Drawdown, Recovery, and What They Hide

An equity curve is the most honest output a backtest produces and the most easily misread one. A curve that looks smooth and steep can hide a strategy that loses 40% in a month and recovers it in the next. A curve that looks choppy can be a robust, boring, low-drawdown system.

Here is how to read one properly.

The four numbers that matter

Maximum drawdown (MDD) — the largest peak-to-trough decline in equity.

MDD = (Peak − Trough) / Peak

MDD is quoted as a percentage, but the duration matters as much as the depth. A 15% drawdown that recovered in four days and a 15% drawdown that took nine months to recover are different experiences for the account holder and different implications for the strategy.

Recovery factor — net profit divided by maximum drawdown.

Recovery factor = Net profit ÷ MDD

This is the most informative single ratio in backtest statistics, because it answers “was the money I made worth the drawdown I endured?” A strategy with a 2.0 recovery factor earned twice its worst drawdown. Below 1.0, it did not earn back its worst loss.

Time under water — the percentage of the test period spent below a prior equity peak.

A strategy with a 50% return and 14 months underwater is psychologically different from one with a 50% return and 3 months underwater, and in practice the second one survives and the first one gets deleted at the 9-month mark and then misses the recovery.

Sharpe ratio — return per unit of volatility. Useful, but it is calculated on returns that are often smoothed or on a sample too small to be meaningful. Treat a Sharpe above 2.0 in a retail gold backtest as a red flag to investigate, not as an achievement.

What a curve hides

The single best trade

Remove the largest winning trade from the curve and recompute everything. If the strategy becomes unprofitable, the result depended on one outlier rather than on an edge. This is common and it is not automatically disqualifying — but it must be disclosed. “Best trade removed: strategy still profitable” is a stronger statement than any Sharpe ratio.

The regime it was born in

Test a strategy over 2019–2021 and you are testing a period where gold moved in a contained range for most of it. Test over 2020 Q3 and you are testing a panic spike. Neither period represents “the market”. A curve is always the curve of that history, and a strategy whose curve looks perfect over one regime is often a regime-specific artefact.

Look for this by segmenting: does the curve have a consistent slope across sub-periods, or does it have one spectacular segment doing all the work?

The averaging-in

A grid or martingale strategy can produce a beautiful smooth equity curve in backtest while accumulating an enormous hidden position. The curve is smooth because the losses are deferred, not because they are absent. Look at the position size over time, not just the equity line. A curve that rises steadily while open lots climb from 0.1 to 3.0 lots is not a low-risk strategy — it is a leveraged one with a delayed accounting system.

The closing-time bias

Strategies that only take trades at certain hours, or that avoid holding over weekends, can look stronger than they are. This is a legitimate design choice, but you have to know it is there. A “no overnight positions” rule removes weekend gap risk from the backtest and also removes the weekend gap risk from reality — you get both, but you should understand that a strategy with overnight positions has a very different weekend risk profile.

Sanity checks on any equity curve

1. Does the curve survive the worst historical period for the instrument? If the test period does not include gold’s sharpest drawdown, the MDD is not a worst-case estimate.
2. Is the curve suspiciously smooth? Real markets have gaps. A curve with no sharp vertical moves suggests the backtest may not model slippage, or may model it only in the average case.
3. How long is the sample? Two years of daily data is ~500 trades for a multi-day strategy. That is enough to distinguish a real edge from noise for a high-frequency system, and nowhere near enough for a 5:1 asymmetric system with 30 trades.
4. Is the number of trades consistent with the strategy’s design? A strategy claiming 3 trades/month over 3 years should show ~108 trades.
5. Does the drawdown cluster in one period or spread out? Clustered drawdown means a regime change you may not survive again.

Converting the curve into a position size

The curve’s job is to inform sizing. Take the maximum drawdown the strategy produced, and size so that if it happens again, you can still trade.

Account size = (MDD %) ÷ (risk % per trade)

A strategy with 20% MDD, sized at 2% risk per trade, needs the account to be able to absorb a 10× sequence of losing trades to reach that drawdown. Whether you would keep following it through that is a separate question, but the arithmetic should be explicit before you run it, not discovered afterward.

What to publish

When we publish backtest results, we publish:

• The full parameter set
• The instrument, timeframe, and test period
• The spread and commission assumption
• Maximum drawdown, recovery factor, and time under water
• The number of trades
• Whether the result survives removing the best trade
• The segment-by-segment breakdown

Anything less is marketing, not research. If a result does not include the parameters and the drawdown, you cannot evaluate it — and if you cannot evaluate it, it is not evidence of anything.

---

Educational content only. Not investment advice. Past performance is not indicative of future results. Trading forex and CFDs carries substantial risk of loss.