Prior Movesmirror the world’s best investors

Institutional ownership research

Read the filings. Study what may follow.

Prior Moves organizes public institutional ownership disclosures and researches whether a manager’s filing history can help rank securities in a later filing. Forecasts are estimates about future disclosures, separate from current holdings and from expected returns.

Explore the research boardRead the methodologyEvaluate it yourselfContact
Evidence snapshot 2026-10-16Forward outcomes pending

The workflow

A research trail from disclosure to forecast.

01 / DISCLOSEDNormalize public filings

Reported positions and changes with source dates. A filing describes a past report period and may become public weeks later.

02 / RESEARCHEvaluate historical ranking

Does past information rank securities later present in a manager’s disclosed holdings? Results depend on the model and evaluation population.

03 / PROSPECTIVEKeep forecasts timestamped

Forward calls carry an as-of time and a target filing period. Outcomes stay unknown until the filing arrives and is graded.

Evidence, with its boundaries

Historical ranking, measured on held-out filings.

A historical US evaluation measured ranking discrimination over held-out manager-security observations: 124,731 rows, base rate 12.46%, 1,244,011 comparable pairs. It is a prediction test. It does not estimate return, profitability or alpha, and it is not the shared board’s own score.

Pair-weighted AUC

0.676295% investor-clustered interval 0.6563 to 0.7002

Panel

49 / 35managers / quarters

Stock-only labels

0.6611same predictions, option and bond rows removed

Japan, within investor

0.732519 investors; shuffled control 0.4968
  • Survivorship: securities that survived to 2026 are over-represented in the scored universe, worth +0.051 AUC per investor, paired, on scored names (28 of 33), a lower bound.
  • Option rows: current labels count purchased calls, puts and bond rows as stock; 31 of 49 managers change on stock-only labels.
  • The shared board uses a separate ranker. Its archived promotion gate measured AUC 0.5603 across 837 scored units, a different question from the figures above.

Forward record

Timestamped calls, graded when filings land.

70 distinct calls since July 2026, 65 matured, from 97,424 daily receipts. Japan: 4 hits on 22 graded against 0.99 expected by chance (p = 0.00784). The US cohort for the September quarter grades after the November filing deadline. The served model is frozen until 2 December 2026 so the record measures one engine.

Separate historical simulation

A positive cumulative result with wide uncertainty.

Across 35 complete quarters (2017 Q3 to 2026 Q1), the 15-name quarter-end book returned 489.5% net of modeled costs, versus 239.4% for SPY. Annualized returns were 22.48% and 14.99%. Its entry uses holdings that were not public until up to 45 days later. This hindsight upper bound could not have been captured by an outside investor.

Mean quarterly difference

+2.217 ptst = 1.313; 95% bootstrap -0.674 to +5.86

Estimators below p = 0.05

1 of 5none after removing the largest quarter

At actual filing availability

|t| 1.00best of 8 transforms; family-wise bar 2.57

Quarters won

23 / 35one-sided sign test p = 0.0448

At actual filing availability, the pre-specified 63-trading-day family had a maximum |t| of 1.00, below its 2.57 family-wise threshold; no transform cleared. Modeled costs were not executed trading costs. This is not live performance or proof of alpha.

Evaluation offer

A forward test a firm can grade itself.

  1. The firm picks the managers and agrees the candidate universe before quarter end.
  2. Before the 13F deadline we deliver ranked files, a SHA-256 hash and a timestamp receipt.
  3. The firm grades against the filings with its own code, against persistence, popularity and shuffled baselines.
  4. Trial: the first 3 months are free. Licence after that: USD 30,000 to 50,000 per firm per year.
  5. Success: the lower bound of the manager-clustered interval clears 0.5 and the best baseline.

What this does not claim

  • 13F report dates are not publication timestamps or executable trade dates.
  • A 13F shows net long holdings at quarter end, never the trades in between, shorts, futures or swaps.
  • Neither historical result establishes future or live performance.
  • Some board inputs have unresolved commercial-use or point-in-time questions; no data licence is implied.
  • Research is informational, not personalized investment advice or a managed account.

Methods and limitations · Public record · How to read the product