MIZAN · the survival table · computed 2026-08-04
Every track record has a price.
Measured in trials.
The Deflated Sharpe Ratio asks one question of any record: would this still be significant
if it was the best of N attempts? Below: the world's most famous return streams, and the largest N each
one survives at the 95% bar — computed by the MIZAN verification engine, monthly closes, every assumption
printed at the bottom of this page.
The number that decides every row — how many strategies were actually tried — is
self-reported across the entire industry. Nothing on earth checked it. That is the business we are in.
01
Apple1997-12 → 2026-07 · 343 months · Sharpe 0.94 annualized
358trials
The deepest record on this table — and even it has a price.
02
NVIDIA2019-01 → 2026-07 · 90 months · Sharpe 1.43 annualized
31trials
A 1.43 Sharpe over seven famous years — gone after 31 tries.
03
Netflix2002-05 → 2026-07 · 290 months · Sharpe 0.76 annualized
18trials
Two decades of the great growth story.
04
SPY — the S&P 500 itself1997-12 → 2026-07 · 343 months · Sharpe 0.66 annualized
13trials
The market itself. Thirteen.
05
Tesla2010-06 → 2026-07 · 193 months · Sharpe 0.81 annualized
8trials
Pedigree is not significance.
06
Microsoft1997-12 → 2026-07 · 343 months · Sharpe 0.58 annualized
8trials
Twenty-eight years of Microsoft: eight trials.
07
Bitcoin2017-08 → 2026-07 · 107 months · Sharpe 0.75 annualized
1trials
The most famous trade of the century survives exactly one trial.
08
General Electric1997-12 → 2026-07 · 343 months · Sharpe 0.35 annualized
1trials
A century of brand, one trial of significance.
09
Ethereum2020-08 → 2026-07 · 71 months · Sharpe 0.66 annualized
0trials
Not significant even as a single attempt.
Assumptions, in full. Monthly total-price returns from the named public series (dividends
excluded — a conservative simplification that lowers Sharpe uniformly). Integer fixed-point arithmetic identical
to the MIZAN in-circuit engine (era-v11 conventions): population moments, raw kurtosis, one-sided AR(1)
effective sample size, null at each N = best of N same-length noise strategies (σ_SR = 1/√n_eff, the MinBTL
convention, Bailey–Borwein–López de Prado–Zhu 2014), DSR = PSR(SR₀) at n_eff (Bailey & López de Prado 2014),
significance 0.95. Windows differ per asset and are printed on each row; shorter windows survive fewer trials —
that is the point, not a flaw.
What this table is not. Not investment advice, and not a claim these assets are bad —
buy-and-hold SPY needs no trial count because nobody searched for it. The table demonstrates one thing:
significance depends violently on a number nobody reports. A manager showing you a Sharpe below 1.0
with an unknown search behind it is showing you the right-hand column of this page, blank.