Does Copying 13F Filings Actually Work? We Tested It
+187.2% Druckenmiller — vs QQQ +59.6% over the same window. Copying the wrong book (Buffett's AI slice, +37.3%) lagged that same index.
The numbers — winners and the loser together
| Whose AI book you copy | Return since Aug 2024 | QQQ, same window | Verdict |
|---|---|---|---|
| Stanley Druckenmiller | +187.2% | +59.6% | Beat, by ~3x |
| Cathie Wood (ARK) | +170% | +59.6% | Beat |
| David Tepper | +108.2% | +59.6% | Beat |
| Warren Buffett (AI-related slice) | +37.3% | +59.6% | Lagged |
Four caveats belong on the same screen as these numbers, not in a footer: ① this is each investor's AI-related slice, not their whole portfolio; ② a 13F cannot see short positions or option hedges, so a hedged book looks naked here; ③ which holdings count as AI is this site's editorial judgement; ④ any period where price coverage fell below 50% of the basket is skipped, not guessed. Past performance is not a promise of anything.
Why filing-day prices are the whole point
A 13F becomes public roughly 45 days after the quarter ends. Backtests built on quarter-end prices — which is nearly all of them — therefore price the copy at a moment when the filing did not exist. That is not a small technicality: it is the difference between measuring a strategy and measuring time travel. Our ledger enters every position at the filing-day close and re-enters at each subsequent filing, using adjusted daily closes, with holdings taken from the line-by-line EDGAR filings themselves.
The academic base rate points the same way: Martin and Puthenpurackal's Berkshire study ("Imitation is the Sincerest Form of Flattery", SSRN 2008) found that mimicking Berkshire's disclosed holdings from 1976–2006 would have beaten the market substantially even when the copy traded a month after disclosure. The lag is survivable. What our test adds is the part the literature can't give you: which of today's AI books was worth copying, at prices you could actually have paid — and the answer splits hard by investor.
Run it yourself — your investors, your date, your amount
The Copy-Homework calculator answers "if I had copied them starting then, what would it be worth today" — pick any mix of investors and a starting filing date; it computes in your browser, each leg benchmarked against QQQ over that investor's own window, result permalinked. One tap in the tool follows your pick with an update when the next 13F lands.
Open the calculator → Who holds what now →What this test cannot tell you
- Whether it keeps working. Eight rebalances is a real record and a short one; the next filing season recomputes everything, including our own headline.
- Anything about whole portfolios. The AI slice is the part this site can grade against its tracked predictions; Berkshire's full book is a different (and unmeasured-here) animal.
- Which investor to copy next. The ledger reports what happened at tradeable prices. It does not rank the future — and a site that graded predictions for a living would be the first to tell you those are different jobs.
FAQ
Does copying 13F filings actually make money?
It can — the question is who you copy. Our backtest buys each investor's disclosed AI holdings at the closing price of the day the 13F was filed and holds to the next filing. Over 8 rebalances since August 2024 (as of 2026-08-16): Druckenmiller +187.2%, Cathie Wood +170%, Tepper +108.2%, against QQQ +59.6% over the same window — while Warren Buffett's AI-related slice returned +37.3% and lagged the index. Copying is not one strategy; it is a different strategy per investor.
Isn't the 45-day filing delay fatal to copying?
The classic academic result says no: Martin and Puthenpurackal's study of Berkshire Hathaway (SSRN, 2008) found a portfolio mimicking Berkshire's disclosed holdings from 1976 to 2006 would have beaten the market substantially even when trades were made a month after the disclosures became public. Our test is the tradeable version of that idea: it never uses quarter-end prices (which nobody reading a filing 45 days later could get) — it buys at the filing-day close, the first price a real person opening EDGAR could actually pay.
Why did copying Buffett lag while others beat the index?
Because the AI slice of Berkshire's book is deliberately conservative: cash-flowing businesses adjacent to AI (Apple, and Alphabet added through 2026) rather than accelerator names. That slice returned +37.3% over the test window against QQQ's +59.6% — a fact about this AI-only slice and this window, not a verdict on Berkshire's whole portfolio, which the 13F-visible slice does not represent.
How is this different from other 13F tracker sites?
Three ways. Pricing: every other backtest we found uses quarter-end prices, which assume you traded 45 days before you could have read the filing; ours uses filing-day closes. Honesty: the loser (Buffett's AI slice, lagging) is published beside the winners, and four caveats sit on the same screen as the numbers. Auditability: holdings come from the line-by-line EDGAR filings, returns use adjusted daily closes, and the whole thing recomputes each 13F season rather than being written once.
Browse all AGI questions · Holdings come only from line-by-line 13F filings; the ledger recomputes quarterly and this page's numbers update in the same run. Nothing here is investment advice.