A day's delay halves the one-month copy alpha from buying newly disclosed positions of famous 13F filers. Smolinski and Saridevi study Buffett, Icahn, Burry, Ackman and Peltz. Their buys outperform as a group for a few weeks, though Icahn's own do not. The authors themselves doubt that their copy-portfolio alphas make an implementable trading rule. They put weight on the absence of reversal: the alpha never turns negative, which they read as information entering prices rather than herding. That interpretation may be right. The trading case still weakens as alpha fades toward zero by six months.

What the paper measures

The premise is straightforward. A discretionary 13F discloses positions backed by research the manager paid for and copiers receive free. Earlier work cited by the authors finds long-horizon returns from copying ordinary funds' disclosures. Smolinski and Saridevi ask whether attention speeds up that return. Investors can follow only so many filers; a famous manager's disclosure may therefore move a price faster than an obscure manager's disclosure carrying the same information. They call the extra return the Buffett effect.

Their search begins with Wikipedia's "American hedge fund managers" category. They screen its 215 entries to 24 concentrated, low-turnover discretionary funds, then rank those 24 using a recognition index. It averages z-scores of log Wikipedia pageviews (2015 to 2025) and log ProQuest news mentions (2000 to 2025). Each fund gets an obscure control matched on size, concentration and longevity. The sample contains 32,785 new-position and complete-exit events from 2000 to 2025. Events are dated to the SEC's EDGAR filing day; returns are measured as Carhart four-factor CARs.

The reported coefficient is a difference-in-differences, which the authors call the disclosure premium. It compares a fund's buy-minus-exit return gap with the same gap for its controls. For the top tier, recognition z of 1 or more across five funds, the premium is +1.422% over the filing day and the next day, rising to +3.091% by day 20. Copying that tier's buys yields 9.8 bps a day over 21 days. TAQ shows retail trade counts up 15.12 log points, while average trade size moves -1.37 (p=0.547). The authors see famous names drawing retail attention to information prices have yet to absorb.

How much rests on the five?

Across all 24 recognizable funds, the fully specified premium against controls is +0.200% at [0,1] (p=0.159). It reaches +0.499% at [0,5] (p=0.031). The 19 moderate-recognition funds show +0.088% (p=0.481). The authors acknowledge the limitation: five managers drive the high-recognition premium.

Results differ within that group. Icahn's own 30-day premium is -2.003% (p=0.005). Without Buffett, the two-day estimate remains +1.122%, significant at 1%, while the 30-day estimate drops to +1.722% and loses significance. The restricted wild bootstrap returns p=0.071 at [0,5] and p=0.148 at [0,30]. I accept the immediate reaction: its restricted-bootstrap p is 0.011, and the estimate remains +1.122% without Buffett, +1.52% under the rolling index and +1.244% with the skill control. The longer-window result looks fragile.

There are few chances to trade it. Figure 3 records 815 high-recognition buy events across 25 years, roughly 33 a year. The copy portfolio averages about ten holdings.

Waiting a day

The headline portfolio enters on the first post-filing day, defined in the paper as the close of the filing day itself. Timing carries much of the result. In the event study, high-recognition stocks beat controls by 67.5 bps on day +1 alone, following +43.2 bps on day 0.

The authors recognize that a copier may miss that close, so they delay portfolio entry by a day. The 21-day alpha falls from 9.8 to 4.9 bps a day (p=0.064); the 63-day alpha falls from 4.4 to 3.2 (p=0.029). Their nearest implementable version, buy-and-hold with day-two entry, earns 4.8, 2.7 and 0.7 bps a day at 21, 63 and 126 days (p=0.075, 0.068, 0.493). Every figure is gross of costs.

Costs matter most where the signal is strongest. The two-day premium is +2.211% in small stocks against +1.104% in large ones (p=0.009), and +2.089% in illiquid stocks against +1.097% in liquid ones. The paper annualizes the 21-day long-short alpha at roughly +26% and says a real copier would likely capture less. I agree.

Does the alpha last?

The abstract's claims that alpha persists for months and does not reverse both hold. No intercept turns negative. Yet the long-only tier's daily alpha falls from 9.8 to 4.4 to 1.4 bps (p=0.164 at 126 days). The long-short alpha falls from 10.4 to 4.8 to 0.8 (p=0.438).

A lasting price change weighs against pure herding. It offers little assurance that a six-month book pays.

Past skill makes persistence harder to pin on recognition. On the subsample with a skill measure, adding each fund's past disclosure CARs cuts the two-day estimate by about a sixth, from 1.476% to 1.244% (p=0.001). The 30-day estimate drops further, from 2.372% to 0.819% (p=0.381). Recognition and that skill measure correlate at r=0.47. The authors also acknowledge that their static recognition index can reflect fame acquired late in the sample. A rolling, as-of Wikipedia index from 2016 yields +1.52% against +1.77% for the static index on the same subsample. It covers only filings from mid-2016, under 40% of the sample period. The fund list, meanwhile, comes from a category retrieved in May 2026 and requires funds to be active near the sample close.

The less famous funds offer a result I find more interesting. Moderate-recognition buy portfolios have no significant two-day premium (+0.088%, p=0.481), yet earn 2.5 bps a day at 63 days and 2.1 at 126 (both p<0.01). Controls earn 0.5 and 0.8 over those same horizons. Slow copying of less famous managers may be the sturdier signal.

Why we did not backtest the copy strategy

We could not test this ourselves. We lack manager-level 13F holdings with EDGAR filing timestamps, which prevents us from identifying new positions and enforcing post-disclosure entry. We also lack historical Wikipedia pageview and ProQuest counts. Our equity price history does not extend into the 2000 to 2009 portion of the sample.

I would reconsider if the buy-and-hold, day-two version cleared realistic small-cap costs on filings after 2025, with recognition measured as of each filing date.