Prior Moves
Free · SEC 13F intelligence
Prior Moves is a free model that predicts the stocks famous investors — Buffett, Burry, Ackman, Loeb, Soros, Klarman and 19 more — are most likely to add to their next SEC 13F filing. Each prediction carries a calibrated conviction score and is backed by an honest, cost-adjusted backtest. Six weeks before the public sees the filing.
No account. No paywall on the core picks. Research tool, not investment advice.
Famous investors reveal their hand once a quarter, with a six-week delay, in a dense SEC filing almost nobody reads in time. Prior Moves turns that public data into a plain-English forecast: a ranked list of the stocks each investor is most likely to buy next — and one combined list across all of them.
Every tracked investor's predictions blend into a single PriorScore (0–100) per stock — weighing model conviction, how many high-track-record investors agree, price momentum, corroborating filings and tradability. Delisted and untradeable names are filtered out.
Choose Buffett, Burry, Ackman, Soros, Klarman, Druckenmiller and more. Each has their own model trained only on their own filing history, so a value investor and a quant aren't averaged into mush.
Every pick lists its evidence — which investors' models predict it, whether Congress or company insiders are buying, activist stakes, and sector. No black box, no blind trust.
Three steps, from a name you recognise to a forecast you can scrutinise.
Start with someone you know — Warren Buffett, Michael Burry, Bill Ackman — or jump straight to the combined Top Picks across all 25 modelled investors.
The model scores 1,200+ stocks and surfaces the names most likely to appear as a fresh buy in that investor's next 13F — ranked by a calibrated conviction score, weeks before the filing is public.
Open any pick to see the supporting signals: model conviction, agreement across investors, recent momentum, and corroboration from congressional buys, insider purchases or activist disclosures.
The single most important test: if you had bought the top-15 PriorScore names each quarter — equal-weight, held exactly one quarter, using only data available at the time, net of realistic trading costs — here is what happened versus the S&P 500.
Measured over 35 quarters (2017–2026) across 25 investors. Net of liquidity-scaled round-trip costs (≈3 bps for mega-caps up to ≈50 bps one-way for thin names). The 95% bootstrap confidence interval on the net edge is roughly +0.2 to +5.6 pts and excludes zero (t ≈ 2.0). Weighting recent quarters more heavily — the regime you would actually trade — lifts the net edge to about +5.4 pts/quarter (and +6.0 over the last eight quarters).
Methodology caveat. These are equal-weight, one-quarter holds on a still-modest sample, and taxes are not modelled. The numbers show that a signal exists — they are directional, encouraging evidence, not statistical proof or a live trading record. A live paper account trading the picks on real prices runs publicly inside the app as the honest counterpart to this backtest.
Built to be checked, not taken on faith. A short summary — the full breakdown and the reliability curve live inside the app.
The model sees 218 numeric features for every U.S. stock each quarter: insider trades (SEC Form 4), congressional buys under the STOCK Act, activist disclosures (SC 13D/G), what 80+ other super-investors hold, fundamentals, multi-source sentiment (Reddit, Hacker News, StockTwits, FT Alphaville), short interest, 10-K risk-factor length, and FRED macro series. A separate model per investor lets each style's strongest signals dominate.
Training is strictly walk-forward: each quarter is predicted using only data from earlier quarters. A leakage canary — shuffling the answer labels and retraining — collapses accuracy to a coin flip (~0.50 AUC versus ~0.83 real), proving the edge is genuine signal, not future information leaking in.
Raw model scores rank well but are over-confident, so Prior Moves applies isotonic calibration on 72,268 out-of-sample predictions. The displayed conviction percentage is the corrected historical hit rate, not the inflated raw number. Per-investor model accuracy ranges roughly 55%–90% AUC.
The headline edge is reported net — every name pays a realistic round-trip cost scaled by liquidity, once per quarter. 13F filings lag the underlying trade by up to 45 days, so a prediction means "what will likely show up in about six weeks", not "buy this today". We would rather under-claim than oversell.
Yes. The ranked Top Picks, per-investor predictions and the full backtest are free to use, with no account required. A free weekly digest of the top names is also available.
No. Prior Moves is a research and educational tool, not investment advice. Predictions are statistical estimates of which stocks may appear in upcoming SEC Form 13F filings — they are pattern-matches, not buy or sell signals. You are solely responsible for your own investment decisions, and should consider a licensed professional in your jurisdiction.
A separate machine-learning model is trained for each investor on 218 numeric signals per stock per quarter — SEC filings, insider and congressional trades, activist disclosures, fundamentals, sentiment and macro data — and ranks the stocks that most resemble that investor's next 13F buy. Training is walk-forward, so every quarter is predicted using only data from strictly-earlier quarters, with no look-ahead.
Across 35 quarters (2017–2026) and 25 investors, buying the top-15 PriorScore names each quarter (equal-weight, held one quarter, net of liquidity-scaled trading costs) returned about +2.8 percentage points per quarter over the S&P 500, beating the index in 25 of 35 quarters. The 95% bootstrap confidence interval excludes zero (t ≈ 2.0). It remains a backtest — directional evidence, not a live trading record.
A Form 13F is a quarterly report that large institutional investment managers must file with the SEC, disclosing their U.S. equity holdings. Filings are due about 45 days after each quarter ends, so they reveal what an investor held roughly six weeks earlier — and only long positions, never cash, shorts, options or private holdings.
The raw model is over-confident, so Prior Moves applies isotonic calibration on 72,268 out-of-sample predictions, which cuts the expected calibration error from 0.020 to 0.001 — meaning the displayed conviction percentage reflects the historical hit rate. A leakage canary (shuffling the labels collapses accuracy to a coin flip) confirms the signal is not look-ahead leakage.
See what Buffett, Burry, Ackman or Soros are most likely to buy next quarter — free, no account.
Open the app →