Practice17 May 2026 · 6 min read

Reading an Equity Curve: What Smooth Actually Means

Equity-curve smoothness is at least five distinct properties; conflating them is how overfit systems pass review and fail in production.

A smooth equity curve is the most seductive artifact in backtesting, and the most poorly understood. Traders staring at a near-linear ascent from lower-left to upper-right tend to read it as evidence of edge, when in fact smoothness is a property with at least five distinct definitions, most of which are unrelated to whether the strategy will survive next quarter. Conflating them is how overfit systems get funded.

This post separates the meanings of "smooth" so you can audit your own equity curves with the right metric for the right concern.

Smoothness is not a single quantity

When a trader says an equity curve looks smooth, they could mean any of: low return volatility, low drawdown depth, short drawdown duration, monotonic local behavior, or low autocorrelation in returns. These are mathematically distinct and frequently in tension. A strategy with very low daily volatility can still produce long, grinding drawdowns. A strategy with rare but small drawdowns may have fat-tailed returns that volatility-based metrics flatter.

The Sharpe ratio collapses the first definition into a single number and stops there. That is why it lies to you about path. Two strategies with identical Sharpe can have wildly different drawdown profiles depending on the serial correlation of their returns.

The R-squared interpretation

The most defensible single-number measure of equity-curve smoothness is the coefficient of determination from a linear regression of log equity against time. If your log equity sits on a straight line, R² approaches 1. If returns are noisy around the trend, R² falls.

R² = 1 − Σ(log E_t − (a + b·t))² / Σ(log E_t − mean(log E))²

Two cautions. First, R² rewards trend stability, not magnitude — a strategy returning 2% annually with vanishing noise scores higher than one returning 30% with mild noise. Second, R² is computed on a single realized path. Bootstrapping the trade sequence and recomputing R² across resamples reveals how much of the smoothness is structural versus accidental ordering.

Ulcer index and drawdown geometry

If your concern is path pain rather than statistical fit, the Ulcer Index captures both depth and duration of drawdowns in one quantity. Define the drawdown at time t as D_t, the percentage decline from the running maximum.

UI = sqrt( (1/N) · Σ D_t² )

Unlike standard deviation, UI ignores upside volatility entirely and penalizes long underwater periods quadratically. A strategy that spends 18 months recovering from a 12% drawdown produces a far higher UI than one that takes the same loss and recovers in 6 weeks. For traders running real capital, UI tracks psychological survivability more honestly than Sharpe.

The Martin ratio — CAGR divided by Ulcer Index — is the underused cousin of Sharpe. It answers a different question: how much trend am I getting per unit of underwater pain, rather than per unit of total wiggle.

Autocorrelation: the smoothness that should worry you

Some equity curves look smooth because returns are positively autocorrelated. Today's gain raises the probability of tomorrow's gain. This is common in strategies that hold positions for extended periods, in mean-reverting strategies during regimes that favor them, and in any system with embedded lookahead bias.

Positive autocorrelation inflates Sharpe by suppressing the realized variance of the equity path relative to the variance of independent returns with the same mean. The naive annualization factor sqrt(252) assumes zero autocorrelation. When ρ₁ is materially positive, the corrected scaling factor is:

σ_annual ≈ σ_daily · sqrt( 252 · (1 + 2·Σ (1 − k/N)·ρ_k) )

If your strategy has ρ₁ above roughly 0.15 in daily returns, your reported Sharpe is overstated by tens of percent. Audit autocorrelation before you trust any risk-adjusted metric. Smoothness from autocorrelation often evaporates the moment the underlying regime changes.

How to read a curve in practice

Open the equity curve with four numbers next to it: R² against a linear log-trend, Ulcer Index, lag-1 return autocorrelation, and the ratio of the largest drawdown to the second-largest. The fourth ratio is the cheapest tail-risk diagnostic available. If it exceeds 2, your worst loss is not representative — it is the tail showing through a small sample, and your distribution of drawdowns is almost certainly being underestimated by the visible curve.

Then look at the curve segment by segment. A genuinely robust system shows comparable local smoothness across non-overlapping thirds of the sample. A curve that is glassy for two years and ragged in the third year is almost always a system whose edge decayed, or one whose smooth period reflects a regime that has ended.

In Kestrel Signal, these statistics are computed alongside the curve by default, with bootstrap confidence intervals on R² and UI. The point is not that smoothness is bad — it is that the word means at least five different things, and pattern-matching on the visual impression of a line conflates all of them. Decompose first, then judge.