Skill persistence (measured 2026-09-30)

A one-off, dated measurement — do the traders with the highest luck-adjusted edge in one quarter stay near the top the next?

A one-off measurement, not a live metric. Measured on 2026-09-30 (≈ 03:50 UTC) on Polymarket data. The numbers will move slightly on a re-run: the underlying rollup is rebuilt nightly and positions resolving into a window are added to it. The polyrank.app landing page quotes it; this page is its public method and caveats.

The question

If a trader is in the top 1 % by skill this quarter, how often are they still in the top 10 % next quarter? If skill were luck, the answer would be about what chance gives.

Definitions

Skill is the luck-adjusted edge shown on trader profiles: edge = Σ (won − w) / √ Σ w(1 − w), summed over the wallet's resolved positions, where w is how often any position bought at that same price wins. A positive edge means the wallet's picks won more often than the prices it paid implied.
Windows are non-overlapping calendar quarters of resolution time. The edge is computed separately in each quarter, for traders with at least 30 resolved positions in it.
Top 1 % and top 10 % are percentile ranks within that quarter's population.

Results — three readings of the same data

Reading	Windows	n (top 1 %)	Still in the top 10 % next quarter	What chance gives
Among traders who kept trading (≥ 30 positions in both quarters)	Apr–Jun → Jul 1–Sep 29 2026	741 of 74,005	81.4 %	10 %
Same, replication	Jan–Mar → Apr–Jun 2026	833 of 83,255	72.1 %	10 %
Counting everyone who stopped trading as a miss	Apr–Jun → Jul 1–Sep 29 2026	2,625 of 262,479	38.9 %	≈ 2.8 %
Same, replication	Jan–Mar → Apr–Jun 2026	2,497 of 249,616	37.5 %	≈ 3.3 %
Ranked per position (mean edge per position, no √n)	Apr–Jun → Jul 1–Sep 29 2026	741 of 74,005	44.3 %	10 %

Among those who kept trading, 92.2 % of the top 1 % still had a positive edge the next quarter (86.1 % in the replication), and only 7.3 % fell to the bottom half.

Caveats

Survivorship. The 81 % counts only traders who kept trading in both quarters. About half of the top 1 % (52.5 %) were still active the next quarter; counting the others as misses gives 38.9 %. Chance is lower under that counting too (≈ 2.8–3.3 %, because only 28–33 % of all traders stay active), so 38.9 % is still about 14× chance. Any sentence that does not say "who kept trading" should use 39 %, not 81 %.
Volume. The edge is a z-score, which grows with the square root of the number of positions, so part of the persistence is "many positions with a small, steady edge". Ranking by mean edge per position instead, 44.3 % of the top 1 % stay in the top 10 % (4.4× chance). The edge persists either way; the 81 % is partly a high-volume effect, as is the served ranking, which uses the same z.
Not a forecast of returns. Persistence of a forecasting edge is not a promise of future profit for anyone copying a trader. Nothing here is investment advice.

Reproduce

The exact queries (read-only, about 7 s each on the production database) are kept in the Polyrank repository under docs/measurements/skill_persistence_2026-09-30/, one file per row above plus a parameterised template.