Method Apr 2026
Benchmarks that flatter
Why naive baselines quietly overstate a model's real performance — and the honest comparisons we hold ourselves to.
Show a strategy next to the right benchmark and it can look like genius or like noise — same returns, different yardstick. Most performance claims fail not because the numbers are fabricated, but because the comparison was chosen, consciously or not, to flatter.
A benchmark is a hypothesis about what was easy. If the easy thing would have made most of your return, the strategy added little. The whole question is whether you picked an honest definition of easy.
The baselines that quietly cheat
Three flattering baselines show up again and again, and all three are avoidable:
- Cash or zero — comparing a long-biased strategy to nothing, so a rising market reads as skill.
- A static index with hindsight weights — a benchmark that already knows which names survived.
- An untradeable composite — a yardstick no one could have actually held, net of costs.
The right benchmark is the best thing you could have done instead — not the worst thing, dressed up as a fair fight.
What an honest comparison costs
Held to a tradeable, cost-aware, point-in-time baseline, most edges shrink. That is not a failure of the strategy; it is the first time the number meant anything. The ones that survive an honest benchmark are the only ones worth trusting — and they are far rarer than the marketing version suggests.
Method note
Benchmarks here are constructed point-in-time and net of estimated costs. A result that only beats a hindsight baseline is recorded as unproven.