Does a 13F Copycat Strategy Work? We Backtested 16 Managers

The 13F Copycat Backtest by the Numbers

What we tested and how

The appeal of a 13F copycat strategy is obvious. Large managers must disclose their long US equity holdings every quarter, so anyone can read the list and buy the same stocks. We wanted to know what that would actually have earned, so we built the simplest honest version of the idea and ran it over every manager’s filing history. The setup, in plain terms:

1. Filings. We pull each manager’s quarterly 13F-HR from SEC EDGAR, using the XML-era filings with report periods from June 30, 2013 onward.
2. Tickers. Each CUSIP is mapped to a ticker through OpenFIGI.
3. Portfolio. The copycat buys the manager’s top 20 holdings, weighted by reported value, at the close of the first trading day after the filing date, and holds them until the next filing.
4. Costs and benchmark. We assume 10 basis points of transaction cost and compare with SPY on a total-return basis.
5. Missing prices. Holdings we could not price, mostly delisted or merged companies, are dropped. Every result carries a coverage figure: the share of portfolio value that could be priced.
6. The blend. For the sixteen managers together, we use an equal-weight portfolio of the sixteen copycat portfolios.

The timing matters. A 13F is due 45 days after quarter-end, so our copycat always trades on information that is already stale by the time it is public. The backtest buys only after the filing date, which means that delay is built into every number below rather than assumed away. For background on what a 13F does and does not contain, see our explainer on 13F filings and hedge fund holdings.

Results: sixteen managers, four winners

The headline is not flattering. The equal-weight blend of all sixteen copycat portfolios returned 14.0% a year against 14.3% for SPY over about 13 years, a shortfall of 0.3 percentage points a year. Its information ratio was 0.03, which is another way of saying it was statistically indistinguishable from the index. Only four of the sixteen individual copycats beat SPY: Greenlight, Appaloosa, Duquesne and Scion. The other twelve lagged.

Two features of the table stand out. First, the spread is wide: from Scion at +12.1 points a year to Baupost at −11.0. Second, the dispersion is mostly noise around zero. The blend lands within a rounding error of the index, and the Sharpe ratio of 0.75 and maximum drawdown of −36% say the blend behaved like a broad equity portfolio, not like something that sidestepped risk.

Skill, or the luck of picking 16 names?

Four winners out of sixteen is easy to over-read, so we checked how strong the evidence is. A rough t-statistic is the information ratio times the square root of the years of data. Duquesne’s 0.62 over 13.1 years gives t ≈ 2.2, the only positive result that clears a conventional threshold of 2. Scion’s 0.42 over 10.1 years gives t ≈ 1.3. On the other side, Baupost’s −0.61 over 13.1 years gives t ≈ −2.2 — just as “significant”, but negative.

That is not the end of the story, because we tested sixteen managers. When you run sixteen tests, one or two clearing a t of 2 is close to what chance alone produces. One strong winner and one equally strong loser out of sixteen is roughly what chance alone produces, so Duquesne is at best a candidate for a genuinely skilled copycat, and even that result deserves skepticism. Scion illustrates a different problem: the +12.1 points a year came with a −73% maximum drawdown and 64% average turnover per rebalance, the highest of the group. A copycat that earns the best return by riding a portfolio that fell nearly three-quarters at one point is not one many investors would have held through.

The most-copied 13F: Berkshire Hathaway

Berkshire’s filing is the one most people have in mind when they say “copy the smart money,” and the copycat trailed SPY by 3.1 percentage points a year (11.2% against 14.3%). Its coverage was 98%, so this is not an artifact of missing prices: almost all of the portfolio could be priced. Our Berkshire 13F playbook covers what the filing reveals and what it hides.

This does not say Buffett is unskilled. The research literature points to a more complicated picture. Frazzini, Kabiller and Pedersen’s “Buffett’s Alpha” (NBER working paper 19681; Financial Analysts Journal, 2018) reports that “Berkshire Hathaway has realized a Sharpe ratio of 0.76, higher than any other stock or mutual fund with a history of more than 30 years,” and that Berkshire’s alpha is insignificant after controlling for Betting-Against-Beta and Quality-Minus-Junk. The authors also describe average leverage of about 1.6-to-1. Their sample is 1976–2017, a different window from ours, so the numbers are not comparable, but the lesson is. The record of the company is not the record of a portfolio assembled from its disclosed stock list. A copycat takes the equity list, not the financing structure or the businesses Berkshire owns outright.

Why copying a 13F can fail

Our data do not isolate each cause, so what follows is qualitative reasoning about why the copycat is a weaker product than the original, not a measured decomposition.

The information is old. A 13F describes holdings at quarter-end and arrives up to 45 days later. Our copycat trades after the filing, so any position the manager changed in the meantime is already different from what we bought. Managers who trade actively, as Scion’s turnover suggests, lose the most from this.

The filing shows one leg. A 13F lists long US-listed equity positions only. It has no shorts, no bonds, no foreign shares and no cash. For a fund that hedges, the longs alone are not the strategy; they can be a different risk profile entirely. The copycat portfolio is not the manager’s portfolio.

Concentration cuts both ways. Taking the top 20 holdings by value keeps the manager’s largest bets, which is in the spirit of Cohen, Polk and Silli’s “Best ideas” research on mutual fund managers’ highest-conviction positions outperforming. But it also leaves you holding a few large positions through the full drawdown, as the −73% at Scion and −66% at Tiger Global show, with none of the manager’s ability to cut risk in real time.

What this means in practice

The evidence supports using 13Fs as an input to research rather than as a portfolio. A filing can flag a stock worth studying, show where a respected manager is concentrated, or confirm a thesis you formed independently. It does not, on this evidence, justify mirroring a portfolio mechanically. For a comparison of slow quarterly disclosures with faster signals, see insider buying clusters versus 13F clusters, and for how tracking tools present this data, Alpha Suite versus WhaleWisdom.

Limitations

The most important caveat is survivorship. Holdings we could not price, mostly delisted or merged companies, were dropped, and that biases every result upward: the true copycat return would include those positions. Average coverage runs from 77% to 98%, and it is lowest for exactly the managers whose results look weakest or most extreme, Baupost (77%), Greenlight (77%) and Scion (81%). The reported returns are therefore best read as upper estimates. For more on this bias, see survivorship bias in investing.

Second, this is one backtest with one parameter set: top 20 holdings, value-weighted, a one-day lag after filing, 10 basis points of cost. Results are sensitive to these choices and we did not search for better ones, which keeps the test honest but also means it is not the last word on every possible copy rule. Third, the manager list is chosen with hindsight. These are famous managers selected in 2026, so the sample already favors names that are well known today, and it still produced a blend that trailed the index. Finally, a 13F shows only long US equities, so the copycat is not what the manager actually runs, and we make no claim about the managers’ own performance. None of this is investment advice.

FAQ

Does a 13F copycat strategy work?

Not reliably in our test. An equal-weight blend of sixteen copycat portfolios returned 14.0% a year against 14.3% for SPY over about 13 years, and only four of the sixteen individual copycats beat the index. Survivorship bias in the data means the real figures are likely lower.

Does copying Warren Buffett’s 13F work?

Over our window, no. The Berkshire Hathaway copycat returned 11.2% a year against 14.3% for SPY, a gap of 3.1 percentage points, with 98% of portfolio value priced. A 13F cannot capture Berkshire’s whole business or financing.

How delayed is 13F data?

A 13F is due 45 days after quarter-end, and it lists long US-listed positions only. Our backtest buys at the close of the first trading day after the filing date, so the delay is included in every result.

Which manager’s copycat did best?

Scion had the highest excess return at +12.1 points a year, but with a −73% maximum drawdown and 64% average turnover. Duquesne returned +7.7 points with an information ratio of 0.62, the only positive result near conventional significance; Baupost’s copycat was equally significant on the downside, and with sixteen managers tested that pattern is close to what chance alone produces.

References

1. Alpha Suite. (2026). “13F copycat backtest (October 2026).” backtest_13f.py, run October 3, 2026; methodology and results summarized in this article.
2. Cohen, R. B., Polk, C. & Silli, B. (2010). “Best Ideas.” SSRN Working Paper. doi:10.2139/ssrn.1364827
3. Frazzini, A., Kabiller, D. & Pedersen, L. H. (2018). “Buffett’s Alpha.” Financial Analysts Journal, 74(4), 35–55. doi:10.2469/faj.v74.n4.3