US held-out AUC
0.676295% CI 0.6563 to 0.7002, resampling managersInstitutional ownership research
Public 13F filings, read forward.
Prior Moves keeps a dated record of what 58 large managers disclosed, ranks the securities most likely to appear as new positions in each manager’s next filing, and publishes the held-out tests and timestamped calls that measure that ranking. It is a shared model-ranked stock board, not a calibrated probability or a return forecast.
Reported fact, model ranking13F · Jun 2026
Disclosed
Warren Buffett, largest positions
AAPL22.0%
AXP17.1%
KO10.9%
GOOGL9.4%
Ranked for next filing
Shared board, PriorScore 0 to 100
OKTA89.3
CRWD88.3
PANW86.7
SNOW86.4
What is observed, forecast and proven
Seven layers, each with its own evidence.
Data, descriptions, forecasts and returns are different claims. This table says which ones Prior Moves can support today.
| Layer | What Prior Moves has | Status |
|---|---|---|
| Observed | US 13F long positions for 58 managers; large-shareholding reports in Japan, the UK, Korea, the EU and Hong Kong. | Public filings |
| Described | New positions, adds, trims and exits per filing; typical position size; turnover. | Descriptive |
| Forecast | Ranked candidates for each manager’s next new positions, combined into one shared board. Adds, trims and exits are research models only. | Served, frozen to 2 Dec 2026 |
| Tested on history | US held-out ranking 0.6762 (49 managers, 35 quarters); Japan 0.7325 (19 investors). Two open biases, below. | Measured, with limits |
| Tested live | 70 distinct calls timestamped since July 2026, 65 matured; the US Q3 cohort grades after the 13F deadline. | Building |
| Returns | At the real filing date, the best of 8 pre-specified transforms reached |t| = 1.00 against a 2.57 bar. | Not demonstrated |
| Institutional delivery | A forward evaluation on SEC-derived data. No feed, licence or service level today. | Evaluation only |
The forecast board
Names ranked for the next filing.
One shared board across the tracked managers, re-scored daily on market data. The filing signal updates when new 13Fs land. Board scored 16 Oct 2026; forward quarter 30 Sep 2026.
PriorScore is a rank within the current board, not a calibrated probability that a particular investor will buy a stock. Open the full board with market filters.
Why not copy the last filing
A 13F is already old when it is published.
DAY 0Quarter ends
The manager’s positions on this date are what the filing will report.
UP TO DAY 45Filing becomes public
Copy-trading tools start here, with a portfolio up to 45 days old.
BEFORE THE NEXT FILINGPrior Moves ranks what it may show
Each call is stamped with its as-of date, so it can be graded when the filing lands.
A filing is a snapshot. Two snapshots show the net change in reported holdings at two dates. A rise from 100,000 to 150,000 shares means the position ended 50,000 larger; the manager may have traded far more in between, and a position opened and closed inside one quarter never appears. Short positions, written options, futures and swaps are not in 13F at all.
Historical evidence
How well the ranking has worked on held-out filings.
The model is trained only on filings public before each test quarter, then scored against what each manager actually filed. AUC is the share of pairs where a real new position outranked a stock the manager did not add; 50% is a coin flip.
Sample
49 / 35managers / quarters; 124,731 held-out rowsTop-decile lift
3.17×the 12.46% base rateJapan held-out AUC
0.732519 investors; shuffled control 0.4968- Survivorship. The scored universe favours securities that survived to 2026; on scored names that is worth +0.051 AUC per investor, paired, (28 of 33), a lower bound. Read 0.6762 as an upper estimate.
- Option rows. The current labels count purchased calls, puts and convertible-bond rows as stock. Scored on stock-only labels, the same frozen predictions give 0.6611 on 119,124 rows; 31 of 49 managers change. A clean refit waits for the model freeze to end.
- Two engines. These figures measure the per-manager model. The shared board uses a separate ranker with its own, weaker held-out test.
Sources: methodology and the gate files it cites. Historical ranking tests are not trading returns.
The forward record
Calls stamped before the filing, graded after.
70 distinct manager-security calls from 97,424 daily receipts, 2026-07-07 to 2026-10-14. Daily re-stamps are not counted as new calls. The US cohort for the September quarter grades after the 13F deadline.
| Market | Calls | Graded | Hits | Reading |
|---|---|---|---|---|
| HK | 2 | 2 | 0 | Too few to read |
| JP | 26 | 22 | 4 | Above its base rate |
| KR | 23 | 23 | 0 | No different from base rate |
| UK | 19 | 18 | 0 | No different from base rate |
The served model has been frozen since 2 September 2026 so the record measures one engine. Full record and receipts.
Return evidence
A positive simulation that is not yet significant.
Quarter-end book
489.5%vs 239.4% SPY, 35 quarters, modeled costsMean quarterly margin
+2.217 pts95% CI -0.674 to 5.86; t = 1.313At filing availability
|t| 1.00best of 8 transforms; bar 2.57Largest quarter
+47.5 pts2026-03-31; mean without it +0.884Holdings were not public for up to 45 days, so the quarter-end book is a hindsight upper bound no outside investor could have captured. At filing availability, none of the pre-specified transforms cleared the 2.57 family-wise threshold. Costs were modeled, not executed. This is not live performance or proof of alpha.
For research teams and data buyers
Test it on filings that do not exist yet.
- You choose the managers and agree the candidate universe before quarter end.
- We deliver ranked files with a SHA-256 hash and a timestamp receipt before the 13F deadline.
- You grade them against the filings with your own code: within-manager AUC and top-decile hit rate against simple baselines.
- Pass if the lower bound of the manager-clustered interval clears both 0.5 and the best baseline.
What is and is not on offer
A forward evaluation on SEC-derived inputs, so the file carries no third-party data question. There is no institutional feed, API licence or service level today, and two inputs on the public board still need commercial-use permission.
Questions
Fair questions.
Is this investment advice?
No. Prior Moves publishes research and a shared model-ranked stock board. PriorScore orders names within that board and is not a calibrated probability that any manager will buy a stock. We do not execute trades or hold money.
What does a 13F filing actually show?
A US 13F lists a manager's long positions in US-listed securities at quarter end, published up to 45 days later. Comparing two filings shows the net change in reported holdings. It does not show the trades in between, short positions, written options, futures or swaps.
What exactly is forecast?
For each tracked manager, which securities are likely to appear as new positions in the next filing. Adds, trims and exits have been studied as research models and are not served as forecasts.
Is the investment edge proven?
No. A quarter-end 15-name book returned 489.5% against 239.4% for SPY over 35 quarters, net of modeled costs, but it enters before the holdings were public. At the real filing date none of 8 pre-specified transforms cleared the 2.57 family-wise threshold.
Why is it free, and what does a subscription add?
Every board is free to preview, five rows a table, and the track record is free in full. Subscribers see every row. Your portfolio is never the business: we never touch it.
Can a fund or data buyer evaluate it?
Yes, as a forward test: the firm names the managers, we lodge hashed and timestamped rankings before the filing deadline, and the firm grades them against the filings with its own code. No feed, licence or service level is offered today.
The weekly digest
The predicted next buys, fresh disclosures, and the week’s catalysts. One email a week. Honest numbers, never advice.
Prior Moves is a research publication, not personalized investment advice. Rankings are model expectations for future 13F filings; PriorScore is not a probability of purchase. We do not execute trades or custody assets. Full disclaimer.