Performance Persistence — Definition, Evidence, and How to Measure It Honestly
Performance persistence is the tendency for past investment performance to predict future performance. The academic evidence is mixed; the appearance of persistence is often an artifact of survivorship bias.
- Performance persistence is the question of whether past investment performance predicts future performance.
- The academic evidence is mixed and depends heavily on whether the dataset includes failed funds and accounts.
- On survivorship-corrected data, most apparent persistence in active management collapses to roughly the level expected from luck alone.
- An honest registry that retains every entry — including dead accounts — is the only structural way to measure persistence without bias.
Definition
Performance persistence is the empirical question of whether a manager who outperformed in one period is more likely than chance to outperform in the next. If past returns carry information about future returns, performance persists. If they do not, last year's top decile is statistically indistinguishable from any other group of managers next year.
Why the question matters
Persistence is the central premise of every track-record-based allocation decision. Allocators do not care about a manager's past performance for its own sake; they care because they implicitly assume past performance carries some information about future performance. If persistence is zero, the entire active-management allocation industry collapses into pure luck-chasing. If persistence is meaningful, then past performance is genuinely informative — and the rigour of measuring that past performance becomes critical.
What the academic evidence shows
The literature stretches back decades. Hendricks, Patel, and Zeckhauser (1993) found short-term 'hot hands' in mutual funds, where last year's winners outperformed in the next year. Carhart (1997) extended that work and concluded most apparent persistence is explained by common factors (size, value, momentum) rather than skill, with the exception of a small group of persistent under-performers. Brown, Goetzmann, Ibbotson, and Ross (1992) showed that on data including failed funds, much of the visible persistence disappears — what looked like skill was largely the surviving cohort having had a good run by luck.
“Persistence of performance is largely an artifact of survivorship bias. When the analysis is corrected for survival, much of the apparent persistence disappears.”
The picture from the modern hedge fund and crypto literature is similar. Persistence on the upside is weak and shrinks once survivorship is controlled for. Persistence on the downside — bad managers staying bad — tends to be more robust. The reason is asymmetric: a manager with no skill can still produce a great year by luck and disappear before the next bad year, but a manager with negative skill keeps producing bad years until they stop.
How NakedPnL is structured to support honest persistence measurement
Honest persistence measurement requires three things. First, every account that ever existed must remain visible — dead accounts cannot quietly disappear. Second, the metric must not be retroactively editable, so past returns can be checked against today's measurement of future returns without doubt about what was published when. Third, the methodology must be the same across all entries, so an apparent persistent winner is not just an artifact of methodology drift.
NakedPnL meets these conditions by design. The hash chain is append-only — a delisted account remains verifiable in the chain even when not visible in the live registry. The OpenTimestamps Bitcoin anchor lets a researcher confirm that yesterday's published number is the same one published today. And the TWR methodology is fixed across accounts, so a like-for-like persistence study is feasible. The methodology guide on survivorship bias goes into more depth on the structural commitments needed.
Related terms
- Survivorship bias — the systematic over-statement of population performance when only survivors are visible.
- Time-weighted return — the metric used to measure period performance in a way that is comparable across periods.
- Verified track record — the substrate honest persistence measurement requires.
- Hash chain — the data structure that prevents retroactive editing of historical entries.