What we do
13FPulse tracks institutional money flows behind 8,222 US stocks, distilling public SEC 13F holdings into one question: which stocks are being accumulated by a rapidly growing number of institutions?
We don't predict prices or give recommendations. We present a pattern tested across 50 evaluable quarters — together with every failure case and limitation we know of.
The data
The SEC requires institutions managing over $100M to disclose US equity holdings within 45 days after quarter-end. The current signal release is 2026Q2, with 8,697 institutions, 8,714 effective filings and 3,376,353 holding rows; signal history covers 53 quarters. Full restatements and additive amendments are resolved by SEC semantics, and each filing's value unit is checked against quarter-end market prices.
The core signal: breadth change
Each quarter we compute, for every stock, the net change in institutional holders (current holders minus prior-quarter holders, or d_holders), ranked cross-sectionally. It is not the count of first-time buyers: d_holders of +100 means 100 more institutions initiated than fully exited. The intuition is that one index fund rebalancing represents one decision, while a net increase of hundreds of holders signals consensus spreading through professional circles — historically an early-to-mid-stage feature of major rallies.
Why not track share amounts? We tested it: changes in aggregate shares held have no predictive power (slightly negative, likely picking up dilution). What matters is the net change in holder count, not how much was bought — consistent with academic findings (Chen, Hong & Stein, 2002).
Strict backtest rule: public first, measured second
The biggest trap in 13F analysis is the timeline: March 31 holdings become public in mid-May. Any "return" computed from March 31 prices is paper profit no real investor could capture.
Our rule: signal visibility date = the actual SEC filing deadline + a 2-day buffer. Weekend and federal-holiday adjustments are applied first; entry uses the next trading session. Every return figure starts only after the data was public.
Backtest results (2013Q3–2025Q4, 50 evaluable quarters)
Holding the top decile of breadth-change each quarter for ~6 months:
| Single stock beats universe | 49.6% — a coin flip |
| Single stock positive return | 58.8% |
| Portfolio beats universe, share of quarters | 32/50 = 64% |
| Portfolio 6-month return | quarterly mean 5.9% vs 4.8% for the universe |
Read this carefully: the signal does not turn any single stock into a high-probability bet. Its use is as a basket that broadens the chance of catching large rallies; a single-name bet remains close to a coin flip.
The clearest failure regime was the 2020 post-crash V-recovery, when stocks institutions had dumped hardest bounced hardest. Trend-following signals can fail sharply in that environment. We do not hide this, because it will happen again.
What it caught
Breadth-change percentile on the day the signal became visible (1.000 = strongest in market):
- NVDA, 2023Q1 holdings (visible May 2023): 0.999, +62% over the next 6 months, followed by a two-year run ⟲
- Micron (MU), 2025Q2 (visible Aug 2025): 0.985, +241% over the next 6 months ⟲
- SanDisk (SNDK), Palantir (PLTR): lit continuously from 2025Q3 / 2023Q2 respectively
What it missed (read this)
Tesla, mid-2019: the stock sat at a major bottom while institutions were fleeing — breadth-change percentile 0.016 (near the bottom of the market). The stock rose +290% in the following 6 months. ⟲
This failure defines the signal's boundary: it confirms trends; it does not call bottoms. Fundamentals-driven, progressively accumulated rallies (NVDA/MU-type) get caught early-to-mid. Retail- and squeeze-driven V-reversals (TSLA-2019-type) are structurally invisible to it — it may even read inverted. Do not use it to hunt for washed-out bottoms.
Exit reference: signal extinction
For stocks previously in the top 20% of breadth change, the following quarter's state matters:
| Next-quarter state | Excess return, quarter after |
| Still lit | +1.3% |
| Extinguished, price holding | +0.4% (usually digestion — false alarm) |
| Extinguished AND price falling | -0.4% (-0.9% over 6 months) |
Extinction alone is noise; extinction confirmed by price weakness (our red light) is what statistically precedes underperformance. Micron's two red lights (2018Q3, 2022Q1) both landed near memory-cycle peaks. ⟲
Our limitations (as important as the results)
- Survivorship bias: delisted/acquired stocks drop out of return statistics; historical figures are optimistic. Relative rankings between signals are unaffected; discount the absolute numbers.
- 45-day lag: you see holdings at least six weeks old. This signal earns the middle of trends, never the start.
- Long-only: 13F excludes shorts; hedges may mask true views.
- Quarterly resolution: intra-quarter round trips are invisible.
- Statistics, not causality: a 64% quarterly win rate means losing 36% of the time, with no guarantee the pattern persists.
Our commitment
A quarter is published as provisional after it passes the coverage gate, while late filings and SEC amendments continue to arrive; it becomes final 30 days after the deadline. Every release must pass count, unit, period-consistency and build checks. When the next quarter arrives, prior lists are settled against real returns — including losing quarters.
Disclaimer
This site is a statistical organization of public data and does not constitute investment advice. Historical patterns do not predict future results. Invest at your own risk.