Ergodicity & Path Dependence — Practical Guide for Builders and Traders

Why this matters

Expected value can look great on paper while real-life outcomes still fail repeatedly. The usual culprit is non-ergodicity: time-average growth (what one person experiences over time) differs from ensemble-average outcomes (what many parallel universes average out to).

In practical terms:

• One catastrophic drawdown can permanently impair compounding.
• Sequence matters: same average return, different path, very different final wealth.
• Survival constraints are first-class design inputs, not risk “afterthoughts.”

Core concepts in plain language

1) Ergodic vs non-ergodic

• Ergodic process: averaging across many samples is equivalent to averaging through time for one sample.
• Non-ergodic process: those two averages diverge.

Most compounding processes (portfolio growth, startup runway, reputation) are non-ergodic.

2) Path dependence

Outcome depends on order of events, not just their frequency.

• +50% then -50% => net -25%
• -50% then +50% => still net -25%

Arithmetic averages hide this asymmetry.

3) Fragility to ruin

If probability of ruin is non-zero and repeated often enough, ruin approaches certainty. “Small edge, frequent bets” fails if tails are not bounded.

Practical diagnostics (weekly checklist)

1. Ruin map

   • Define hard-fail states: margin call, capital < operating minimum, trust breach, legal breach.
   • Explicitly estimate probability and trigger conditions.

2. Path stress test

   • Reorder historical returns and simulate clusters of bad states.
   • Compare median terminal value vs worst-decile terminal value.

3. Time-average growth lens

   • Track geometric growth metrics, not only arithmetic expected return.
   • Monitor log-growth under slippage/fees/stress regimes.

4. Recovery math audit

   • Table required gain after loss: -10%→+11.1%, -30%→+42.9%, -50%→+100%.
   • Use this table to set max drawdown guardrails.

Execution rules that actually work

Rule A: Survival > speed

Cap risk per unit time. If drawdown acceleration exceeds budget, automatically reduce participation and leverage.

Rule B: Convexity hygiene

Prefer structures with bounded downside and scalable upside. Reject strategies with hidden short-gamma behavior unless explicitly hedged.

Rule C: Sequence-aware sizing

Increase size only after confirming regime stability; cut size faster than you add it. Upshift slowly, downshift quickly.

Rule D: Optionality reserve

Keep dry powder (capital, attention, compute, decision bandwidth). Optionality is wasted if no reserve exists when opportunity appears.

A simple policy template

State machine

• Green (normal): baseline sizing
• Amber (stress signals): -30~50% sizing, tighter stop conditions
• Red (dislocation): capital protection mode, only high-conviction or hedge trades

Trigger examples

• Volatility percentile > 90
• Slippage z-score > +2
• Fill quality deterioration + adverse selection markout worsening
• Liquidity depth collapse

Exit from defense

Require N consecutive intervals of normalized metrics before restoring risk.

Common mistakes

• Optimizing expected return while ignoring survival probability.
• Using single backtest ordering and calling it robust.
• Assuming “small losses” are symmetric with “small gains” in compounding systems.
• Treating tail risk as a narrative problem instead of a sizing problem.

TL;DR

In non-ergodic systems, the path is the outcome. Build strategies and workflows that maximize time-average growth and minimize ruin probability. Compounding belongs to survivors.