← Our researchOctober 7, 2026 · Portfolios

193 portfolio optimizers, one honest scoreboard

Ten years of walk-forward testing: once you account for luck, none of 193 optimizers beats simply holding the S&P 500. The robust ones are the ones that fear drawdowns.

Optimizers
0 of 193beat the S&P 500 after deflating for luck

The question.

Mean-variance, risk parity, HRP, CVaR, Black-Litterman… which way of splitting the money actually works, when every method plays by the same rules?

0.21best deflated Sharpe vs the S&P 500 (0.95 needed)
−25%worst drop for min-drawdown, vs −34% for the index
0.23correlation between past and future rankings
#141where the 2017–2021 champion finished after

How we tested it

1The field193 optimizers: 8 covariance estimators × 7 risk optimizers, tail-risk methods, expected-return models, Black-Litterman and more.
2The rulesThe 100 most liquid US stocks (200 as a control), long only, 10% cap per name, 10 bp per unit of turnover.
3The testWalk-forward, rebalanced monthly from February 2017 to October 2026 using only information available that day.
4The judgeDeflated Sharpe against the S&P 500, bootstrap tests, and the rank in 2017–2021 against 2022–2026.

What we found.

2017 → 2026CAGRSharpeWorst drop
Min-CDaR (minimize drawdown)14.1%0.94−25.0%
Min-Ulcer13.9%0.93−24.8%
Max-Sharpe · James-Stein · Ledoit-Wolf19.9%0.95−32.7%
S&P 500, buy and hold15.4%0.88−33.7%
  • No optimizer beats the index demonstrably. What you can get is the same return per unit of risk with much smaller drops.
  • Minimizing drawdown is the robust winner across both universes.
  • Picking the in-sample champion does not work: the best of 2017–2021 finished 141st afterwards.
  • Using price-target upside as expected return concentrates risk in noise: its predictive power is close to zero.
193 portfolio optimizers, one honest scoreboard · Finaxion