Skip to main content
⚖

WHO'S RIGHT?

The theories judged on unseen years, then today's reading

THE VERDICTS MEASURED BY THE LAB

Each theory is written as an exact rule, with the parameters of its original paper fixed in advance, then judged by the Strategy Arena Lab on a sealed period: the most recent 30% of the series, which the rule never read. Seven rules, seven proofs, one protocol hashed before the first. What these theories bring, measured: mostly protection. None beats its reference significantly; six of seven reduce the maximum drawdown of the sealed period (1.1 to 10 times shallower).

TheorySealedVerdictMax drawdown (theory / ref.)Per year (theory / ref.)Proof
Taleb30 yNo proof of an edge yet-5.0% / -50.3%3.2% / 10.6%SAP-dc1ad089ec3c
Mandelbrot29.8 yNo proof of an edge yet-25.8% / -50.3%8.2% / 10.4%SAP-e0773f1a8194
Dalio13 yNo proof of an edge-15.0% / -21.5%6.2% / 9.3%SAP-7b6971dcef99
Markowitz13 yNo proof of an edge yet-16.6% / -18.9%4.4% / 7.3%SAP-ae3c415a13a2
Choueifaty13 yNo proof of an edge yet-16.2% / -18.9%5.5% / 7.3%SAP-2ff4ea84ffb2
Sharpe13 yNo proof of an edge-16.1% / -18.9%9.3% / 7.3%SAP-4f2eaa967e91
WifeyAlpha10.5 yNo proof of an edge-27.2% / -22.9%8.5% / 9.0%SAP-16036a84c421
Talebmax drawdown ÷ 10.0
Barbell: 90% Treasury bills, 10% US stocks, reset monthly (Taleb, Antifragile, 2012). The convex options leg is not tested.
No proof of an edge yet: the gap with the reference is not significant
Over 30 unseen years, the 90/10 barbell earned far less than stocks (+155.1% vs +1,945.5%), but its worst drawdown was 5.0% while stocks lost 50.3%: the maximum loss divided by 10. That is Taleb's promise: never be ruined.
Proof SAP-dc1ad089ec3c · 30 sealed years (1996-09 → 2026-08) · the theory page
Mandelbrotmax drawdown ÷ 1.9
US stock exposure = min(1; 12% / volatility of the last 126 days), the rest in bills, monthly, no leverage (Barroso and Santa-Clara 2015, after Mandelbrot's 1963 volatility clustering).
No proof of an edge yet: the gap with the reference is not significant
Cutting exposure when volatility rises, the idea drawn from Mandelbrot's volatility clusters, divided the maximum drawdown by 1.9 over 29.8 years (25.8% vs 50.3%) at the same Sharpe (0.57 vs 0.57). Return is lower (+942.1% vs +1,815.0%) because the rule is often half invested: the idea holds on protection, not on return.
Proof SAP-e0773f1a8194 · 29.8 sealed years (1996-11 → 2026-08) · the theory page
Daliomax drawdown ÷ 1.4
Unlevered risk parity: each asset weighted by 1/volatility (last 36 months), monthly (Asness, Frazzini and Pedersen 2012). Gold and commodities (monthly averages, World Bank) are held at the average price of the month after the decision: correction of 20/11/2026, which replaces the 19/11 proof.
No proof of an edge: picking the window on the past would not have held (PBO 58%)
Without leverage, risk parity does not beat the 60/40 over 13 years (+117.4% vs +217.2%), but it divides the maximum drawdown by 1.4 (15.0% vs 21.5%) with 6.5% volatility instead of 9.6%: Dalio's all-weather promise, steadiness, not return.
Proof SAP-7b6971dcef99 · 13 sealed years (2013-09 → 2026-08) · the theory page
Markowitzmax drawdown ÷ 1.1
Minimum variance, long only, covariance of the last 60 months, rebalanced monthly (Markowitz 1952; Jagannathan and Ma 2003). Gold and commodities (monthly averages, World Bank) are held at the average price of the month after the decision: correction of 20/11/2026, which replaces the 19/11 proof.
No proof of an edge yet: the gap with the reference is not significant
Markowitz does not beat the equal-weight basket over 13 years, but cuts volatility from 8.0% to 6.0% a year and the maximum drawdown from 18.9% to 16.6%: exactly what the minimum-variance portfolio promises, less risk, not more return.
Proof SAP-ae3c415a13a2 · 13 sealed years (2013-09 → 2026-08) · the theory page
Choueifatymax drawdown ÷ 1.2
Maximum diversification, long only, covariance of the last 60 months, monthly (Choueifaty and Coignard 2008). Gold and commodities (monthly averages, World Bank) are held at the average price of the month after the decision: correction of 20/11/2026, which replaces the 19/11 proof.
No proof of an edge yet: the gap with the reference is not significant
Maximum diversification does not beat the equal-weight basket on return over 13 years, but it cuts the maximum drawdown from 18.9% to 16.2% and volatility from 8.0% to 6.3%: the idea holds on protection, not on return.
Proof SAP-2ff4ea84ffb2 · 13 sealed years (2013-09 → 2026-08) · the theory page
Sharpemax drawdown ÷ 1.2
Tangency portfolio (best estimated Sharpe), long only, means and covariance of the last 60 months, monthly (Sharpe 1964). Gold and commodities (monthly averages, World Bank) are held at the average price of the month after the decision: correction of 20/11/2026, which replaces the 19/11 proof.
No proof of an edge: picking the window on the past would not have held (PBO 75%)
The tangency portfolio did better than the equal-weight basket over these 13 years (+217.6% vs +148.5%, Sharpe 0.99 vs 0.69), but the Lab cannot yet tell this gap from luck: it is not significant, and picking the estimation window on the past would not have held. Sharpe's contribution that stands without debate is his measure: it is what judges every other theory on this page.
Proof SAP-4f2eaa967e91 · 13 sealed years (2013-09 → 2026-08) · the theory page
WifeyAlpha (Antonacci)max drawdown × 1.2
Dual momentum (Antonacci 2014): each month, the better of US and non-US stocks over 12 months if it beats bills, otherwise the 10-year Treasury.
No proof of an edge: picking the window on the past would not have held (PBO 90%)
Over the last 10.5 unseen years, dual momentum did not keep its protection promise: maximum drawdown 27.2% vs 22.9% for the three markets at equal weights, similar return (+135.5% vs +147.4%). On the training period (1991-2016, already known), the same rule had cut the drawdown from 35.3% to 18.4%: it protects in long bear markets, less when rebounds are fast. A simple, forecast-free rule anyone can judge again every year.
Proof SAP-16036a84c421 · 10.5 sealed years (2016-03 → 2026-08) · the theory page

No testable rule: Satoshi and Szabo (a protocol and an architecture, not a market rule). Testable but not judged for lack of long BTC data under a clear licence: Herlin (gold + BTC) and Choueifaty's BTC/gold version. Data: Ken French Data Library (stocks, bills), Board of Governors FRED/H.15 (10-year yield, public domain), World Bank Pink Sheet (gold, commodities, CC BY 4.0). The 10-year Treasury is rebuilt from its yield; gold and commodities are monthly averages, held at the average price of the month after the decision (correction of 20/11/2026). Research on past data, not advice.

Have your own variant judged in the Lab →

PANTHEON, WAVE 2: TEN FAMOUS RULES, 36 PROOFS

Ten rules written and preregistered on 12/10/2026 with their sources' parameters, played once per market and judged in certified mode by Lab 270 on the period after each source's publication. Bogle has no rule to judge: he is the reference. In this wave, 36 proofs pitted ten famous theories against plain buy-and-hold, each on unseen years: 0 "survives", 3 do better than buy-and-hold over their period (only one clearly, Jegadeesh and Titman's sector momentum), and 20 of 36 have a smaller maximum drawdown. What Bogle brings, measured: beating "doing nothing" is rare, and what rules offer most often is less drawdown, not more gain.

TheoryProofsUp toVerdictsProofs (excerpt)
Meb Faber219.1 y2 no proof yetSAP-4497a1a6480b SAP-4f3589dd7254
Moskowitz, Ooi & Pedersen714.3 y4 no proof yet, 3 no proofSAP-ed106b457ad4 SAP-3bb2d1e84c17 …
Bouman & Jacobsen413.0 y3 no proof yet, 1 no proofSAP-3c3916056432 SAP-f0939b09d16c …
Robert Ariel418.3 y2 no proof yet, 2 no proofSAP-ca9be788f6f8 SAP-4a7e5be91e5b …
Gary Antonacci111.5 y1 no proof yetSAP-059ebfe4a789
Jegadeesh & Titman133.0 y1 no proof yetSAP-635556a921c1
Blitz & van Vliet120.0 y1 no proof yetSAP-cac545e82971
Larry Connors & Cesar Alvarez317.7 y3 no proofSAP-0d4bb9a2890d SAP-4e81e4a4fe2d …
Richard Dennis & William Eckhardt1218.7 y12 no proofSAP-9331c1d0a12b SAP-36e1c240bcf0 …
Brock, Lakonishok & LeBaron133.7 y1 no proof yetSAP-cfd1492ddc37

Also: Bogle (the reference), PlanB (a forecast, not a rule), and what the Lab does not judge (Seykota, Gann, Elliott, Simons). Research on past data, not advice.

Archive: simulation stopped on 20 November 2026. Figures show the last computed state and no longer change.

TODAY'S READING (SIMULATION ON BTC AND GOLD, NOT A VERDICT)

⏳
Loading...

VERDICTS BY THEORIST

DATA USED

Loading...
This page is an educational analysis based on live data. This is not financial advice. Markets are unpredictable — that's exactly what Taleb and Mandelbrot demonstrated.

The Mathematicians' Verdict

This page puts side by side the ideas of 10 thinkers (Taleb, Mandelbrot, Dalio, Markowitz, Sharpe, Satoshi, Szabo, Choueifaty, Herlin, WifeyAlpha). At the top, the verdicts measured by the Lab on a sealed period; further down, a daily reading computed on market data, which is a simulation and not a verdict.

Click each thinker to discover their dedicated page. To have your own rule judged on a period it never saw: the free online judge or the free Lab. Method: how the judge decides.