Metrics5 min read

How to Read an Equity Curve

An equity curve is the time series of a strategy's account value, plotted from the first bar of the backtest to the last. It is the single most information-dense diagnostic in systematic trading: drawdowns, regime shifts, leverage scaling, and overfitting all leave visible fingerprints on it. Reading it well means treating it as a forensic artifact, not a marketing chart.

Computation

The equity curve at time t is the cumulative product of period returns, scaled by initial capital. For a strategy producing returns r_1, r_2, ..., r_t:

E(t) = E(0) × Π (1 + r_i) for i = 1 to t

Equivalently, in log space — which is how you should usually view it for multi-year backtests:

log E(t) = log E(0) + Σ log(1 + r_i)

Log scaling converts geometric growth into a straight line, so deviations from linearity become visually meaningful instead of being swallowed by compounding at the right edge of the chart.

How to interpret it

Start by checking linearity on a log axis. A robust strategy produces an equity curve that is approximately linear in log space across the full sample, with consistent slope across non-overlapping subperiods. A curve that is flat for years then explodes at the end is the signature of a strategy concentrated in one regime — usually a sign the edge does not generalize.

Examine the drawdowns. Measure both their depth (peak-to-trough percentage) and their duration (time from peak to recovery). A clean curve shows drawdowns that are shallow relative to the trailing return and that recover within a time horizon shorter than the recovery would take under a buy-and-hold benchmark. Long flat periods at all-time highs are not problematic; long flat periods below the high-water mark are.

Look at the noise structure. Healthy equity curves have stochastic-looking texture — small wiggles in both directions at the bar level. Curves that ascend in unnaturally smooth diagonal lines almost always indicate lookahead bias, survivorship bias, or in-sample fitting. Real edges are noisy.

A useful exercise: cover the right third of the equity curve with your hand. If the visible left two-thirds would not have convinced you to deploy capital, you are extrapolating from a sample that includes information you would not have had in real time.

Compare against a benchmark plotted on the same axes — typically a buy-and-hold of the underlying or a risk-free accrual line. The spread between the strategy curve and the benchmark, not the strategy curve in isolation, is what you are actually trading. A 12% CAGR strategy during a decade when the underlying returned 14% has negative alpha regardless of how the equity curve looks standalone.

What it does not capture

An equity curve hides path dependency at the trade level. Two strategies with identical equity curves can have radically different position sizing, holding periods, and execution requirements. The curve also obscures leverage: a 20% annual return on 5x leverage and a 20% return on 1x leverage produce visually identical lines, but the first has five times the tail risk.

Transaction costs, slippage, and borrow fees may or may not be embedded — the curve looks the same either way. A backtest equity curve is only as honest as the cost model behind it, and most published curves omit market impact entirely.

Equity curves cannot reveal selection bias from strategy search. The curve you are looking at is the survivor of however many variants you tested. The smoother and more attractive the curve, the more skeptical you should be that you are looking at a discovered edge rather than an in-sample artifact.

The curve also says nothing about capacity. A strategy whose backtest assumes infinite liquidity will produce a beautiful equity curve up to the point real capital touches it. Volume-conditioned fill assumptions are a separate diagnostic the curve cannot replace.

In Kestrel Signal

Kestrel Signal renders equity curves on a log axis by default, overlays the chosen benchmark, and shades drawdown periods directly on the chart so depth and duration are read together. The in-sample and out-of-sample segments are visually separated, and walk-forward folds are stitched with markers at each refit boundary — making it harder to mistake a fitted curve for a discovered one.