Premium · Weekend Research Drop

The 13F filer study, re-run honest.

The event studies and backtests behind the desks — methodology in the open, including the failures and the corrections. This drop: does following the top-100 institutional filers actually beat the tape?

⚠ Data correction (May 31, 2026). An earlier version of this study double-counted amended 13F-HR/A filings, inflating the headline edge. The numbers below are the corrected, de-duplicated run. We publish the correction rather than quietly editing — that's the whole point.

The question

13F filings disclose what big institutions held 45 days after quarter-end. The folk wisdom says "follow the smart money." We tested it properly: take the top-100 filers by AUM, detect their brand-new positions, and measure the abnormal return of buying on the filing date.

100Filers studied
69Buffett new-stakes
+3.01%Mean abn (+5d)
p<0.05Significant filers

Pooled result

Across all top-100 filers, the pooled new-stake edge is real but modest — most of the "smart money" is already priced in by the time the 45-day-late filing lands. The signal is not uniform: a small subset of filers carries almost all of the edge.

Finding 1. The pooled abnormal return is positive but small once amended filings are removed. Following everyone is barely better than the index.

Out-of-sample check

We split the sample in time and re-ran on the held-out period. The filers that mattered in-sample largely kept mattering out-of-sample — the edge isn't just overfitting to a lucky window.

Finding 2. The walk-forward retained the bulk of the in-sample edge for the significant filers. Concentration, not breadth, is what survives.

Statistically significant filers (p < 0.05)

A handful of filers produce new-stake moves that clear the significance bar. Berkshire Hathaway is the cleanest: 69 new-stake events, 75% beat SPY at +5 sessions, +3.01% mean abnormal, and 83% win-rate on stakes ≥ $1B. That single filer is the backbone of the Buffett desk.

Finding 3. The edge is filer-specific. Trade the significant filers as named operators — not a basket of all 100.

What we shipped from it

This study is exactly why the Signal Desk tracks Buffett as a high-conviction operator and treats the broad "follow 13Fs" idea as noise. Methodology, data, and the correction are all here so you can replicate it — and disbelieve us if the numbers don't hold.

Built on free public data — SEC EDGAR 13F-HR filings + yfinance prices. Event-study abnormal returns are vs. SPY. Backtested results do not guarantee future returns; nothing here is financial advice.