Three in ten new positions no longer appear in the next 13F

Why it matters

Anyone following 13F portfolios is looking at an old snapshot: it reflects the end of the quarter and becomes public weeks later. A new line in a 13F is often read as a conviction of the manager. But the form shows what was held at quarter-end, is filed with a median delay of 38 to 40 days (2023–2025), and does not say when anything was bought or when it was sold. The practical question is: if I see a new position, or a manager's largest position, how likely is it to still be there when the next 13F comes out? The answer says nothing about returns; it is about how much the snapshot is worth as a description of the portfolio.

1. How long a new position lasts

Of the 114,587 clean new positions, 29.2% (95% CI: 28.0% to 30.4%) no longer appear in the next 13F. After one year, 40.6% (39.1% to 42.0%) remain, and after two years, 25.1%. The result is the same without positions disclosed by amendment (29.3%) and from 2015 onward (29.2%). It is close to the audit's independent recalculation, which gives between 29.5% and 30.3% depending on the cleaning; the difference comes from treating a 13F with a confidential omission as censoring at any horizon and from working at the issuer level.

Entry weight orders the curves. Among new positions below 0.5%, 34.1% no longer appear in the next 13F and 36.8% remain after one year. Among those of at least 10%, only 9.6% are missing from the next one, 65.7% remain after one year (CI: 62.3% to 69.2%), and the median is 7 13Fs.

The manager's prior turnover separates them even more. The share that no longer appears in the next 13F ranges from 16.2% in the lowest-turnover quintile to 43.1% in the highest. And managers differ a great deal: among the 1,039 with at least 20 new positions, that share has a median of 23.8%, with 10% of managers below 6.1% and another 10% above 50.0%. The manager's turnover matters as much as size: a small new position at a low-turnover manager lasts longer, at one year, than one of 2% to 5% at a high-turnover manager (48% vs. 29%).

Probability that a new position is still in the next 13F / one year later (4 13Fs), by entry weight and the manager's prior turnover

Number of new positions in parentheses. Kaplan-Meier, Q3 2013 – Q2 2026. "Low" turnover: quintiles 1 and 2 (up to 19.2% per quarter); "medium": quintile 3 (19.2% to 26.1%); "high": quintiles 4 and 5 (more than 26.1%).

What this means for anyone following 13F portfolios. Almost one time in three, a new line is something the next 13F will no longer show. Before giving it weight, it is worth checking two facts that come from the 13Fs themselves: how much it weighs in the declared portfolio and how much that manager usually turns over.

2. The life of the No. 1 position

We take the largest position (by value) of each concentrated manager in each quarter since 2015 (25,409 observations). One quarter later, the issuer appears in 96.9% of cases. One year later, in 86.6% (CI: 85.6% to 87.5%). After five years (20 13Fs), in 54.3%. But the ranking changes quickly. It is still No. 1 in 64.5% of cases the next quarter, 41.1% after one year (CI: 39.5% to 42.8%; equivalent to 47% of those still in the portfolio), and 16.2% after five years. After one year, 72.6% are still in the top 5.

The probability that the No. 1 is still in the portfolio after one year hardly depends on its weight: 86.8% if it weighed less than 10% and 86.9% if it weighed 20% or more. What does depend on weight is whether it remains the largest: 29.9% and 57.0%, respectively.

What it means. A concentrated manager's largest position rarely disappears from one year to the next. Losing the top spot is the norm, and on its own it does not indicate that the manager is leaving (it may be, for example, because other positions rose more).

3. Does a reduction warn of an exit?

We track 42,346 observations (position and quarter) of positions that weighed at least 10% of a concentrated manager's portfolio (1,317 managers). In the next 13F, 1,181 no longer appeared: these are "clean" exits, with no other class of the same issuer, no acquisition or delisting, and no corporate event. The median weight in the last 13F that included them was 12.7%. In 65.6% of those exits, the last 13F that included them showed no reduction in shares (of more than 1%) versus the previous one. That percentage is a base rate, not a feature of exits: among positions of the same size that stayed in the portfolio, 68.1% also showed no reduction.

The right comparison runs the other way. If the manager had reduced the position, the probability that it does not appear in the next 13F is 3.0% (CI: 2.6% to 3.4%). If it had not been reduced, 2.7% (2.4% to 3.0%). The difference, 0.3 pp, cannot be distinguished from zero (CI: −0.1 to 0.7 pp). A prior increase does not protect either: 3.0%. Positions with no change are the most stable (1.6%). One caveat: requiring the position to weigh at least 10% in the last 13F leaves out many that were cut sharply, because they no longer reach that weight. Two situations do change the picture, although they are exploratory cuts (we defined them in this report and the audit did not review them). A position that was new in the previous 13F is missing from the next one in 9.5% of cases. And after a cut of more than half, in 16.7% (CI: 11.6% to 22.4%). But such cuts are rare: 216 cases, 0.5% of positions, accounting for 36 of the 1,181 exits.

What it means. In a 13F, a moderate reduction of a large position is not a warning of an exit, and the absence of a reduction offers no reassurance. Most exits from large positions arrive without a prior reduction visible in the filings.

4. Cascading exits

A secondary section, using the figures from the audit's recalculation. Unit: a stock in a quarter, held by at least 5 concentrated managers (45,042 observations, 3,416 stocks, 44 quarters from 2015 to 2026). If no one exited in quarter t, 11.0% of the remaining holders exit in t+1. If more than a quarter exited, 18.9% exit (90% CI: 18.1% to 19.4%).

About 3 of the nearly 8 points of difference are composition: where there was an exodus, the remaining holders are managers with higher turnover. If the expected rate is computed from each holder's own exit rate over its four previous quarters, the excess is +3.3 pp (90% CI: 2.6 to 3.7) in the above-25% bracket (where no one exited, the observed rate is 1.5 pp below the expected one), and +5.2 pp in stocks with at least 20 holders. In a quarter-by-quarter regression, the relationship appears in all 44 quarters. Verified example: of the 55 concentrated managers that held Fiserv (CUSIP 337738108) at the end of Q3 2025, 25 no longer held it in Q4 2025. Of the 35 that held it in Q4 2025, 15 did not hold it in Q1 2026. The audit checked 5 of those Q4 2025 exits on EDGAR, and all 5 match.

What it means. A mass exit in one 13F tends to continue in the next, but the effect itself is moderate (about 3 points). Much of what looks like an ongoing exodus is simply that the remaining holders are the ones with the highest turnover.

Cases verified on EDGAR

Data-quality counterexample. ADAMCAPITAL Gestao de Recursos Ltda reported its 19 positions with made-up CUSIPs in Q2 2021 and with the real ones in Q3 2021. Without filtering this out, 18 false "one-quarter positions" would appear.

Conclusion

What we observe. Among concentrated managers, what appears new in a 13F has a short life: 3 of every 10 new positions do not make it to the next filing and 6 of every 10 do not make it to one year. Large new positions at low-turnover managers are the exception. The No. 1 position usually stays in the portfolio, although it loses the top spot. Exits from large positions are almost never preceded by a prior reduction, and group exits continue only slightly more than can be explained by who the remaining holders are.

A possible interpretation (not proven here). Many small new positions look like trial or tactical positions, and the 13F, because it is quarterly, captures only part of the decisions. To read a 13F usefully, it pays to look at the weight and the manager's style before the novelty of the line.

Methodology

• Concentrated managers: filers that meet GestorLens's criteria in that quarter (between 5 and 100 stocks, top 10 ≥45%, ≥$250M in stocks, average turnover ≤50%, options ≤20%, ETFs ≤20%, 8 quarters of history, eligible filer type). The criteria are applied quarter by quarter (methodology).
• New position: an issuer (the first six characters of the CUSIP, which group share classes together) that appears in the 13F for t and not in the one for t−1, at a manager that is concentrated in t and filed a 13F with stocks in t−1. Entry weight: the issuer's value over the declared 13F stock portfolio, not over the manager's total assets.
• "No longer appears": in t+k the manager filed a 13F and the issuer (or its successor after a CUSIP change) is not in it. We never say "it sold": the 13F only shows that the line is missing.
• Cleaning: no ETFs or funds (closed-end funds included), no CUSIPs with an invalid check digit, none of the 19 made-up sequential CUSIPs from ADAMCAPITAL Gestao de Recursos Ltda (00507V109 to 00507V127, confirmed on EDGAR), and no new positions below the threshold that allows a position to be omitted (fewer than 10,000 shares and less than $200,000; SEC FAQ 39). No portfolio copies that arrive through another filer's notice filing.
• Censoring (cases that stop being observed without counting as an exit): the manager did not file, the security left the universe through an acquisition or delisting, the 13F for t+k declares a confidential omission, or the data runs out (Q2 2026).
• Calculation: discrete quarterly Kaplan-Meier curves. 95% confidence intervals via a manager-level bootstrap (2,000 replicates, with managers in a fixed order and a fixed seed: two runs give identical intervals): whole managers are resampled because their positions are not independent. Prior turnover = the average quarterly turnover of the declared portfolio over the four quarters before the new position (measured by weight changes; only information available at the time).

Limitations

• "No longer appears" does not mean "sold". It may be a transfer to an entity that does not file a 13F, a hedge outside the form, a position that fell below the FAQ 39 threshold, or a filer error. Nor does the 13F say when within the quarter the change happened: a "one-quarter position" may have lasted anywhere from one day to six months.
• The data reflect amendments filed later (today's version). Without the positions disclosed by amendment, the headline figure does not change (29.3%).
• Turnover is measured by weight changes, which mix price moves and trading. A "traded" turnover (shares × price) gave the same ordering.
• Filer type is a current attribute applied to the entire history. There is a slight retroactive classification bias.
• In section 3, the "≥10% in the last 13F" cut leaves out many positions that were cut sharply. The cuts by size of the move (a cut of more than half, a position new in the previous 13F) are exploratory and have small samples (216 and 1,614 cases).
• The No. 1 position reading is direct at each horizon (it does not require the position to have appeared in every intermediate 13F), and the sample shrinks at long horizons.
• No comparison was made with non-concentrated managers. The academic literature on the duration of institutional holdings and the persistence of institutional demand was not reviewed: the originality claimed is that of the practical approach, not of the phenomenon.
• What cannot be claimed: that the manager sold in that quarter, or when; anything about returns or the reason for an exit; that one exit "causes" others; anything that amounts to a recommendation.

Sources

• SEC, Frequently Asked Questions About Form 13F, questions 39 (small positions that may be omitted) and 56 (confidentiality in arbitrage), accessed Oct 18, 2026: sec.gov.
• Berkshire Hathaway Inc (CIK 1067983): Q2 2019, 0000950123-19-008356 · Q3 2019, 0000950123-19-011362 · Q4 2019, 0000950123-20-002466 · Q1 2020, 0000950123-20-005345 · Q2 2020, 0000950123-20-009058 · Q4 2021, 0000950123-22-002973 · Q1 2022, 0000950123-22-006442.
• Philadelphia Financial Management of San Francisco, LLC (CIK 1351407): Q2 2019, 0001214659-19-005281 · Q3 2019, 0001214659-19-007202.
• Brilliance Asset Management Ltd (CIK 1665751): Q4 2024, 0001665751-25-000003 · Q1 2025, 0001665751-25-000005.
• ADAMCAPITAL Gestao de Recursos Ltda (CIK 1732080): Q2 2021, 0001567619-21-014402.
• On GestorLens: methodology and how-to guide.