Buys are ambiguous. A fund can add to a position for a dozen reasons — rebalancing, inflows, a hedge. A complete exit is different: it is a decision with no half-measure. So we measured it separately.
The measurement, step by step
A signal is only as good as its measurement rule, so here is the rule in full. The panel is 83 funds tracked by the site, over 19 quarters from 2021Q3 to 2026Q1. A signal fires when three or more funds that held a stock in one quarter held none of it the next quarter — a complete exit, not a trim. The return is measured from the first closing price after that quarter's filing deadline, which is the first price at which an outside reader could have acted on the filing. That entry rule matters: 13F filings arrive up to 45 days after the quarter ends, so the signal is always stale by construction, and measuring from the deadline is the honest test of whether the information survives the delay.
Two details of the panel deserve attention. First, the headline computation ran over 81 managers with CIKs at the October 6 run, while the 2026Q2 formation section uses 83 — the pre-expansion roster. The two extra managers are the reason the extension is kept separate from the headline figures: mixing panels would silently change the denominator. Second, 190 exit signals occurred but only 151 could be measured with complete price data. The 39 that could not be measured are not a random subset — 38 had CUSIPs with no ticker mapping and one (BAM, 2021Q3) had a ticker with no usable price series — and unmeasured names are usually the worst outcomes, including delistings.
The comparison is always against the S&P 500 over the same window, so the headline figure is an excess return — the dumped stocks' return minus the market's return over the identical dates — not a raw return. That is what makes the −5.34 points interpretable: it is how much worse the dumped stocks did than the market, in the same quarter, over the same days.
The test
Same 83 funds, same five years. A signal fired when three or more funds that held a stock in one quarter held none of it the next. We measured returns from the first closing price after each quarter's filing deadline, exactly as in our buy test. 190 exit signals occurred; 151 could be measured with complete price data (the rest were mostly delisted or unmapped names — a survivorship caveat we flag below).
The result
- Dumped stocks' average return next quarter: −1.35%
- S&P 500 over the same windows: +3.99%
- Excess return: −5.34 percentage points (95% CI −8.17 to −2.50)
This is not a crash signal — about half of the dumped stocks still rose (50.3% of them). But on average they fell while the market rose. Of every signal we tested — crowded buys, tripled positions, exits — the exit was the strongest, and it wasn't close (−5.34 vs −0.43 and −1.48 points). It is also the only one of the three whose confidence interval stays clear of zero. The shape matters too: the median dumped stock was roughly flat (+0.72%), so the negative average was pulled down by a minority of large falls rather than by a typical stock collapsing.
Reading a confidence interval
The confidence interval is doing most of the work in this article, so it is worth saying plainly what it means. The headline excess of −5.34 points comes with a 95% interval of −8.17 to −2.50. That interval is a statement about precision: if the same measurement were repeated over many similar samples, the true excess would fall inside an interval of this width about 95% of the time. The key question is whether the interval includes zero. It does not — both ends are negative — so the data support a real negative effect rather than noise around no effect.
Contrast that with the subgroups. The exactly-four-funds group shows an excess of −4.18 points, which looks large, but its interval runs from −10.91 to +2.54 and therefore includes zero: the data cannot rule out that the true effect is nothing, or even positive. The same holds for the five-or-more group (−3.42, interval −12.94 to +6.09). A wide interval is not a small effect — it is an imprecise one, and the honest reading is "cannot tell", not "no effect". The width comes from the sample size: 21 and 15 events cannot pin down an effect of this size, however large the point estimate looks.
That asymmetry makes sense once you think about who is selling. A fund exiting entirely has usually finished its homework: the thesis broke, the numbers changed, or something better appeared. The filing arrives up to 45 days late, and even with that delay, the underperformance persisted into the following quarter.
Does louder consensus mean worse?
If joint exits signal trouble, more sellers should mean more trouble — the data do not support it. Splitting the 151 measured signals by fund count shows the headline effect living almost entirely in the smallest group. Signals with exactly three exiting funds — 115 of the 151 — averaged −1.96% against +3.84% for the market, excess −5.80 points, interval −9.11 to −2.49, the only subgroup whose interval stays clear of zero. Signals with exactly four funds (21 events) show an excess of −4.18 points, and signals with five or more (15 events) −3.42 points — but both intervals are wide enough to include zero (−10.91 to +2.54, and −12.94 to +6.09), so neither can be called a real effect on its own. There is no dose response here: louder consensus did not perform worse. Small samples may explain it — 21 and 15 events cannot pin down an effect of this size. What the funds did together mattered; how many did it, beyond three, did not measurably add to it.
- Exactly 3 funds: n = 115 · avg −1.96% · S&P +3.84% · excess −5.80 (95% CI −9.11 to −2.49) · rose 48.7% · median −0.66%
- Exactly 4 funds: n = 21 · avg +0.42% · S&P +4.60% · excess −4.18 (95% CI −10.91 to +2.54) · rose 61.9% · median +5.69%
- 5 or more funds: n = 15 · avg +0.89% · S&P +4.32% · excess −3.42 (95% CI −12.94 to +6.09) · rose 46.7% · median −2.27%
- 4 or more funds combined: n = 36 · excess −3.87 (95% CI −9.36 to +1.63)
Where the headline lives
The subgroup split reveals where the headline actually comes from. The exactly-three group is 115 of the 151 measured signals — more than three-quarters of the data — and it is the only subgroup whose interval excludes zero. The headline excess of −5.34 is, to a first approximation, the exactly-three group's −5.80 diluted by the smaller, noisier groups. If the exactly-three group had shown no effect, there would be no headline at all.
That concentration has an implication for how the signal should be used. The rule that fires is "three or more funds exited", but the evidence for the rule is carried by the "exactly three" cases. A reader who sees a four- or five-fund exit should not assume a stronger signal — the data say the opposite, or more precisely, they say nothing at all about whether more funds mean a worse outcome. The threshold of three is where the evidence sits, and it sits there because that is where the sample is large enough to measure.
It also explains why the median and the average tell different stories. The median dumped stock was roughly flat (+0.72%), while the average was −1.35%. The gap means a minority of large falls pulled the average down, and that minority lives disproportionately in the exactly-three group, whose median was −0.66% — the only subgroup with a negative median. The headline is not a typical stock falling; it is a typical stock going nowhere while a minority falls hard.
Not every quarter
The headline average hides a ragged calendar. Across the 19 signal quarters, excess was negative in 13 and positive in 6 — leaning negative more often than not, but far from uniform. The two worst quarters were the most recent fully scored ones before 2026Q1: 2025Q3 (−30.36 points on just 5 measured signals) and 2025Q4 (−29.91 points on 9, with not one dumped stock beating the market). At the other end, the busiest quarter, 2023Q3 (16 signals), came out positive (+3.82 points), as did 2023Q1 (+3.52 on 8). And the latest quarter in the panel, 2026Q1, was slightly positive (+1.65 points on 11). Single quarters rest on 3 to 16 measured signals, so none should be read as a precise quarterly effect. The point is narrower: the headline average is not one quarter's accident — negative in more than two-thirds of quarters — but not a law that held every quarter, and not in the latest one.
- Quarters with negative excess: 13 of 19 · worst 2025Q3 −30.36 (n = 5) and 2025Q4 −29.91 (n = 9, win 0.0%)
- Quarters with positive excess: 6 of 19 · busiest 2023Q3 +3.82 (n = 16) · latest 2026Q1 +1.65 (n = 11)
One quarter, then nothing
The natural follow-up is whether dumped stocks keep falling after the first quarter. On the same measurement rule, one quarter later, the answer is: the effect fades. Take the 140 measured signals whose second-quarter window can already be observed (the 11 from 2026Q1 are excluded: their second window ends November 17, 2026, after the latest prices in our cache). In that second quarter, dumped stocks averaged +2.68% against +3.47% for the market — excess just −0.79 points, interval (−3.55 to +1.97) that comfortably includes zero — and 61.4% rose. The same 140 events in their first quarter had shown an excess of −5.89 points, with an interval (−8.82 to −2.95) clear of zero. The second quarter is neither a delayed crash nor a rebound — noise around zero, with very wide two-sided moves underneath (worst −57.87 points, best +53.10). Whatever information a joint exit carries, it is spent within one quarter.
- Second quarter after the signal: n = 140 · avg +2.68% · S&P +3.47% · excess −0.79 (95% CI −3.55 to +1.97) · rose 61.4% · median +3.44%
- Same 140 events, first quarter: avg −1.97% · excess −5.89 (95% CI −8.82 to −2.95)
- Second quarter by fund count: exactly 3: −1.36 (−4.65 to +1.93) · 4 or more: +1.06 (−3.79 to +5.91)
Why the effect fades in one quarter
The fade is the most informative result in the article, because it constrains what the signal can be. If joint exits reflected a permanent deterioration — a business in secular decline, a fraud slowly unwinding — the underperformance should persist into the second quarter and beyond. It does not. The second-quarter excess is −0.79 points with an interval that comfortably includes zero, and 61.4% of the dumped stocks rose. The information, whatever it was, was priced in within one quarter of the filing deadline.
That timing points to a specific mechanism. The funds exit, the filing arrives up to 45 days later, and the market reads the filing and reprices the stock over the following quarter. By the second quarter, the repricing is done. The wide two-sided moves in the second quarter (worst −57.87, best +53.10) are consistent with this: once the signal is spent, the stocks revert to moving on their own news, in both directions, with no systematic drift. A reader who acts on a joint exit in the second quarter is trading on information the market has already absorbed. The fade also sets a boundary on the signal's use: it is a one-quarter observation, not a standing short thesis. Any use that extends beyond one quarter is extrapolating past the evidence.
The same names, again
The 151 measured signals cover only 107 distinct stocks, so some names were exited jointly more than once. That repetition turns out to matter. Stocks exited jointly exactly once — 84 signals — averaged −3.17%, excess −6.68 points (−10.82 to −2.54), the strongest subgroup in this expansion and, like the exactly-three group, one whose interval excludes zero. Stocks exited jointly twice or more — 67 signals across 23 names — averaged +0.94%, excess −3.66 points (−7.37 to +0.06), an interval that just barely includes zero. The repeat names are mostly the market's biggest stocks — META 7 signals, UNH 5, BABA, V, GOOGL and AMZN 4 each — names so widely held that joint exits may partly reflect index and portfolio mechanics rather than a shared verdict. The comparison is descriptive, not a trading rule: repeat names were selected by being big and liquid enough to be exited repeatedly. But it sharpens the headline: the exit effect is strongest where it happens once, in names the funds left and did not revisit.
- Exited jointly once: n = 84 · avg −3.17% · excess −6.68 (95% CI −10.82 to −2.54) · rose 46.4%
- Exited jointly twice or more: n = 67 across 23 names · avg +0.94% · excess −3.66 (95% CI −7.37 to +0.06) · rose 55.2%
Still forming
The panel's story does not end at 2026Q1. On a separate, clearly labeled basis — the pre-expansion roster of 83 managers with CIKs, using the holdings cache as collected in October 2026 — 10 new joint-exit signals formed in 2026Q2: MSFT and INTU with four exiting funds each, and BKNG, FLUT, CTRA, PDD, HON, DIS, STZ and TMUS with three each. Their scoring window runs August 17, 2026 to November 17, 2026, the first trading days after the two filing deadlines, so no return can be claimed yet — and late 2026Q2 filings received after the cache build could still add signals. Two of the ten repeat the previous quarter: MSFT and INTU were also joint-exit signals in 2026Q1 (four and three funds respectively). The extension stays separate from the headline figures on purpose: the headline computation ran over 81 managers with CIKs at the October 6 run, while this formation uses 83, and re-running the same rule on the current cache reproduces the 2025Q3 and 2025Q4 sets exactly but gives 13 against the study's 12 for 2026Q1 — the cache has moved since the study. When the November window closes, these ten will either extend the pattern or join 2026Q1 as a quarter that did not.
- 2026Q2 formations (unscored): 10 signals · 4 funds: MSFT, INTU · 3 funds: BKNG, FLUT, CTRA, PDD, HON, DIS, STZ, TMUS
- Window: Aug 17 – Nov 17, 2026 · unscored
How to read a joint exit
A reader who sees a joint exit should work through five questions. First, how many funds exited — exactly three, where the evidence sits, or four or more, where the data cannot confirm an effect. Second, whether the name was exited jointly once or repeatedly — the effect is strongest where it happens once and weakest in the biggest, most widely held names. Third, whether the scoring window is still open — a formation whose window has not closed carries no return yet. Fourth, whether the exit could be mechanical — a fund closure, an index change, a mandate shift. And fifth, whether the quarter is the first after the filing deadline, because the effect is spent within one quarter and a second-quarter reading is noise around zero.
The honest caveats
39 of 190 exit signals couldn't be measured — 38 because their CUSIPs had no ticker mapping, and one (BAM, 2021Q3) because its ticker had no usable price series — and unmeasured names are usually the worst outcomes (delistings among them), so the true figure may be worse than −5.34. Exits can also be mechanical (fund closures, index changes), which adds noise in the other direction. The subgroup and quarterly figures rest on small samples — 15 events for five-or-more funds, 3 to 16 per quarter — and their wide intervals say so plainly; none should be read with headline confidence. The 2026Q2 formations are unscored and use a slightly different panel, so they stay separate from the headline average. We report it as we measured it.
The survivorship caveat, in full
The 39 unmeasured signals deserve more than a footnote, because they bias the headline in a specific direction. A CUSIP with no ticker mapping is usually a name that left the public markets — a delisting, a buyout, a bankruptcy — or a name too small or too obscure to map. A ticker with no usable price series is, in the BAM case, a name whose data simply stops. In both cases the missing names are disproportionately the worst outcomes: a delisted stock's return is, for the holder, typically a large loss, and excluding it removes the worst results from the average. That means the headline excess of −5.34 points is likely an understatement of the true effect — the measured average excludes the names that fell hardest, so the real average (if every signal could be measured) would be more negative, not less.
The bias runs in the other direction for mechanical exits. A fund that closes, merges, or changes mandate exits every position at once, and those exits carry no information about the stocks. They add noise that pulls the measured effect toward zero. The two biases partly offset each other, but they do not cancel neatly, because they act on different subsets and with different magnitudes. The honest statement is that −5.34 is the measured figure on the 151 signals with complete data, that the true figure on all 190 is probably more negative because of the unmeasured names, and that the exact size of either bias cannot be recovered from the data at hand.
If you read 13Fs, read the exits first.
Data: SEC Form 13F filings, 83 funds, 2021Q3–2026Q1 headline panel; 2026Q2 formations on the pre-expansion panel; prices through October 2026. Methodology and full scoreboard on our Track Record page. Not investment advice.
Revision history
- October 10, 2026 — figures unified. The numbers above were recomputed on the site's single measurement standard: the October 2026 SEC CUSIP-to-ticker mapping, with entry at the first closing price after the filing deadline. As first published, this article reported a preliminary run on the mapping then in force — an average return of +0.66%, an excess of −3.94 points, and 130 of 190 events measured — while the Track Record page and the companion win-rate study each carried a different value from their own runs. All four places now report the same computation: average −1.35%, S&P 500 +3.99%, excess −5.34 points (95% CI −8.17 to −2.50), 151 of 190 measured. The conclusion is unchanged and, on the unified figures, stronger: exits remain the largest measured effect on this site, and the only headline signal whose interval excludes zero.
- October 31, 2026 — expanded with new analyses. Added threshold, quarterly, second-quarter persistence, repeat-name, and 2026Q2 formation sections from the same signal data; headline figures re-verified unchanged on recomputation.