Path Stability

What Path Stability measures inside the Robustness Score — where your backtest lands in its own resampled cloud, plus a dynamic, scale-invariant width penalty — and how to interpret a low score in plain English.

Updated 2026-06-15
9 min read
intermediate

TL;DR. Path Stability answers two questions: was your backtest a typical draw of its own edge, and how much does trade order swing the outcome? It rewards an original result that lands near the middle of its resampled cloud, penalizes one that sits in the lucky top tail, and adds a width penalty only when the fan's downside is large relative to the strategy's own typical gain.

What you'll be able to do

  • Read a Path Stability score and know whether your equity curve was a lucky draw.
  • Understand why position in the cloud — not raw fan width — is what the score leans on.
  • Spot when "great backtest" actually means "great backtest, low Path Stability" — the most common warning sign on the platform.

Why this matters

Every backtest is a single path. The trades happened in one specific order, at one specific time, in one specific market regime. The strategy could have produced wildly different equity curves if the same trade list had landed in a different order, or if a few key trades had landed on adjacent bars.

Most of those alternate orderings are just as valid as the one you saw. None of them is the future, but the spread across them is a direct estimate of how much your backtest was edge versus how much was luck.

Path Stability turns that into a number. It runs from 0 to 100 inside the Robustness Score and carries 25% of the weight — tied for the heaviest.

The core idea

The Robustness Lambda takes your strategy's trade list, runs it through a stationary block bootstrap ten thousand times, and gets ten thousand alternate equity curves. Each one is a plausible alternate history of the same edge, and each ends at a total return — so you get a distribution of ten thousand possible outcomes.

Path Stability reads two things from that distribution:

  1. Position — was your backtest lucky? Where does your original total return fall within the cloud? Sitting near the middle (or below) means the result you reported is a typical draw of your edge — trustworthy. Sitting in the lucky top tail means a favorable trade order inflated it: the typical outcome is lower than what you saw, so the headline number overstates the edge.

  2. Width — does order dominate? How wide is the fan's downside relative to the strategy's own typical gain? This is a ratio, not a dollar figure, so it's invariant to trade count, contract size, and instrument. A fan that's wide but entirely profitable isn't penalized — there's no fragility if every ordering still makes money. A fan whose worst cases dwarf the typical gain is dampened, because where you land then leans on luck.

The two combine by multiplying: position sets the ceiling, and width can only pull it down (and only so far). Downside risk itself — ruin, deep drawdowns — is owned by Tail Discipline, so Path Stability never double-counts it.

Placeholder · Diagram

Path Stability — where the original lands in its resampled cloud

Author hint: A fan/cloud of resampled equity outcomes (percentile bands) with the ORIGINAL total return drawn as a horizontal marker line. Left panel 'Robust': the original line sits near the median of the cloud. Right panel 'Lucky': the original line sits in the top 5% of the cloud. X axis = trade number, Y axis = total return. Save to /public/images/guide/robustness/path-stability/diagram.svg.

The math

Position × dynamic width

Path Stability multiplies a position score by a width multiplier:

pct        = percentile of your original total return within the cloud (0–1)
position   = 100                                if pct ≤ 0.60
           = 100 · (1 − (pct − 0.60) / 0.40)    if pct > 0.60   (→ 0 at the very top)

width_mult = 1.0                                if worst-5% outcome ≥ 0   (all profitable)
           = clamp(1 − 0.20 · (−p5 / median), 0.40, 1.0)   otherwise

PathStability = position × width_mult

Position is asymmetric on purpose: a lucky draw (top tail) is penalized, but an unlucky one (bottom tail) is not — an unlucky backtest is conservative, and live results are more likely to beat it than disappoint. The width term uses −p5 / median — the worst-5% loss as a multiple of the typical gain — which is scale-free, so futures and equities, short and long backtests all normalize to the same units. The 0.40 floor means width alone can never zero a well-positioned strategy.

For optimization children, the whole component is then scaled by the Deflated Sharpe factor, because a result selected as the best of many trials is itself a kind of lucky draw.

In AlgoLift

Path Stability shows up in two places:

  • The Robustness Score waterfall. On the Summary sub-tab, each component contributes a labeled segment to the score breakdown. Path Stability is the leftmost segment because of its weight.
  • The Path Luck fan chart. Path Luck visualizes the same data Path Stability measures — the cloud of resampled equity curves with the original equity curve overlaid, including the original's percentile. A low Path Stability score corresponds to an original curve sitting in the lucky upper tail, or a fan whose downside runs wide relative to the typical outcome.
Placeholder · Screenshot

Path Luck — when the original equity curve sits at the lucky tail

Author hint: Path Luck sub-tab showing the equity fan chart with the original equity curve as a thick line. The original curve sits clearly in the top 10% of the cloud — a low Path Stability scenario. Capture with the percentile bands visible. Save to /public/images/guide/robustness/path-stability/path-luck.png.

Reading a Path Stability score

ScoreReadingAction
80–100Original sits in the typical zone of its cloud and the downside is contained relative to the gain.Treat the backtest's returns as representative.
60–79A mildly favorable position, or a moderate downside-width penalty.Cross-check Edge Concentration and the tail components.
40–59A clearly favorable draw, or a downside several times the typical gain.Treat the headline metrics with caution; find what drives it.
Below 40Original is a lucky-tail outlier of its own strategy.Treat the backtest's headline metrics as an upper bound. Re-validate before deploying.

Worked example

Two strategies tested on MES 5m from 2018-01-01 to 2024-06-30:

Placeholder · Worked Example

Two strategies, two Path Stability scores

Author hint: Strategy A: a balanced trend-follower, ~1200 trades, original total return $58,400, sim median $54,200, P5–P95 = −$8,000 to $96,000, original at 51st percentile → Path Stability ≈ 96 (central position, small downside relative to the gain). Strategy B: an over-optimized scalper, ~340 trades, original total return $94,800, sim median $41,500, P5–P95 = $12,000–$72,000, original at 96th percentile → Path Stability ≈ 12 (lucky top-tail draw; fan is all-profitable so width adds no penalty). Use real platform numbers when capturing.

What this shows. Strategy A's original return sat near the median of its alternate histories and its downside stays small next to its typical gain — Path Stability ≈ 96. Strategy B's original return was at the 96th percentile of its own resampled distribution: even though the headline return looks better, only about 1 in 25 orderings of the same trades produces that result, so Path Stability flags it as a lucky draw (≈ 12). Note that Strategy B's fan is entirely profitable — it won't blow up — but Path Stability isn't asking whether you'll lose money (that's Tail Discipline); it's asking whether the number you reported is real.

Common misunderstandings

Common Misconception
Myth
A high Path Stability score means the strategy will work live.
Reality
Path Stability says the equity curve you saw was a typical outcome of the trades the strategy generated — not that the strategy itself is durable across markets. A strategy that overfits the data can produce trades that consistently resample into similar outcomes; that's a separate failure mode caught by Tail Discipline, Edge Concentration, and the Deflated Sharpe correction.
  • "My Path Stability is low but my Sharpe is high. Which do I trust?" The low Path Stability. Sharpe is computed from one path; Path Stability is computed from ten thousand. The one with more information wins.
  • "My fan is huge but Path Stability is still 100." Expected when the whole fan is profitable. Width only penalizes downside relative to your typical gain, so a wide-but-all-green fan with a central original scores full marks — the dollar size of the fan is never penalized on its own.

When Path Stability is most informative

  • Strategies with few but large trades. Each large trade can swing the equity curve substantially; the resampled distribution will reveal whether the wins are clustered or spread.
  • Mean-reverters with short holds. These tend to produce moderately fat-tailed return distributions; Path Stability reliably flags ones that depend on one big tail event.

When it's less informative

  • Strategies with thousands of small trades. The relative fan tightens and the original almost always lands near the median, so Path Stability tends to sit high regardless. Lean on Edge Concentration and Sample Size Confidence to tell a strong edge from a merely numerous one.
Key Takeaway

Path Stability is the answer to "how lucky was your equity curve?" When it's high, the curve you saw represents the strategy. When it's low, you saw an outlier — and outliers don't repeat.