We Backtested the Most Popular Ways to Trade 13F Filings. None of Them Beat the Market.
June 24, 2026 · FindataFox
We backtested the four most-marketed ways to trade hedge-fund 13F filings — the honest way, with the 45-day filing lag, transaction costs, and out-of-sample discipline. Every single one underperformed the S&P 500.
TL;DR
13F-tracking sites all sell some version of "copy the smart money and beat the market." We had the data and the discipline to test that — so we backtested the four strategies people actually pitch, across 13 years (2012–2025), modeling the things that turn paper edges into fiction: the 45-day filing lag (you can't act on a 13F until it's public), transaction costs, survivorship (delisted losers booked, not deleted), and an out-of-sample split (design on 2013–2019, validate untouched on 2020–2025). Returns are factor-adjusted (market, size, value, momentum), because beating the market by holding more small-caps isn't skill — it's a tilt you can buy in an ETF for three basis points.
The scoreboard: 0 for 4. None produced a statistically significant factor-adjusted alpha, and every basket underperformed the S&P 500.
| Strategy | The pitch | Factor-adjusted alpha | Significant? | Verdict |
|---|---|---|---|---|
| Patient capital | "buy what long-term holders own" | +0.4%/yr (long-short) | No (t=0.1) | ❌ No edge |
| Net-flow momentum | "buy what institutions are buying" | −1.9%/yr (long-short) | No (t=−1.0) | ❌ No edge (mildly inverted) |
| Crowding fragility | "avoid the crowded names" | ~0%/yr | No (t=−0.2) | ❌ No edge |
| Best Ideas clone | "copy small concentrated managers' top new bets" | +3.5%/yr | No (t=0.7) | ❌ No edge |
This isn't a list of strategies we tried and gave up on. It's a list we measured — and the measurement says the public, 45-day-lagged 13F signal has been arbitraged flat. Below is exactly what we did and what each one showed.
The rules (why you can trust the nulls)
A backtest only means something if it can't flatter itself. Every test here obeys the same non-negotiable rules:
- 45-day lag. You can't buy what a fund filed until the 13F is public — ~45 days after quarter-end. We enter at the price on that date, never at quarter-end. (This one rule kills most "follow the whale" edges on its own.)
- Transaction costs. 25 bps round-trip per quarterly rebalance.
- Survivorship handled. A stock that delists or goes bankrupt doesn't vanish from the basket — we book it as a loss. Dropping it (the most common silent backtest bug) would delete the disasters and fake an edge.
- Factor-adjusted. We strip market/size/value/momentum and report only the alpha left over. Underperforming the S&P while showing positive "alpha" just means the alpha came from a small-cap tilt the model gives back.
- Out-of-sample. Design on 2013–2019; the 2020–2025 leg is never used to tune anything.
- Publish the failures. A disproven strategy is a finished result, not something to bury. That's the whole point of this exercise.
An aside on rule #3 and #4: while building this we hit both classic traps. An early version used raw (split-unadjusted) prices and dropped delisted names — and produced a beautiful, totally fake "+11% significant alpha" on a basket of every stock. The tell was a market beta of −0.08 (impossible for a real stock basket). Fixing the price data, the survivorship, and the factor timing made the phantom alpha evaporate. We mention this because it's exactly how a less careful backtest convinces someone they've found an edge they haven't.
1. Patient capital — "buy what long-term holders own"
The idea: stocks owned by patient, low-turnover institutions should outperform — long holding periods proxy for conviction and informed, sticky capital.
What we found: nothing. The stocks held longest by institutions returned no more than the stocks held shortest. The long-short spread (most-patient minus least-patient) was +0.4%/yr full-period (t=0.1) and negative in the in-sample window. Both ends underperformed the S&P.
2. Net-flow momentum — "buy what institutions are buying"
The idea: the single most-marketed 13F signal — the "biggest buys this quarter" lists. Institutional demand should push and predict price.
What we found: no edge, and if anything mildly inverted. The most net-bought stocks slightly trailed the most net-sold ones — the long-short spread was negative in every window (full −1.9%/yr, t=−1.0). After the 45-day lag, the demand-driven move has already happened. We do not claim "fade institutional buying" works (the inversion isn't significant) — only that following the buying doesn't.
3. Crowding fragility — "avoid the crowded names"
The idea: over-owned "hedge-fund hotels" are fragile and underperform — both the statically most-held names and the ones being rapidly crowded into (a comomentum reversal).
What we found: no fragility, either way. Most-crowded vs least-crowded: spread ~0 (t=−0.2). Fastest-crowding vs slowest: the sign even flipped between in-sample (continuation) and out-of-sample (reversal), both noise. Crowding — static and dynamic — is not a tradeable signal here.
4. Best Ideas clone — "copy small concentrated managers' top new bets"
The idea: the most academically-supported version (Cohen–Polk–Silli). Don't follow funds broadly — clone only the high-conviction new initiations of concentrated, smaller managers, where 13F should still carry signal.
What we found: no edge — and a clean control proved why. We compared the concentrated-small-manager basket against the same high-conviction new buys from any fund. They performed the same. The "concentrated small manager" filter — the entire thesis — added nothing. Full-period alpha +3.5% (t=0.7), and it held up across conviction thresholds from 5% to 15% of book. Underperformed the S&P throughout. The literature's best shot doesn't survive the 45-day lag.
What this means
- The 13F edge, if it ever existed, is gone by the time you can act on it. Forty-five days is forever; the dataset is the most-scraped filing in finance.
- "Beat the market" and "have skill" aren't the same. Most of these baskets made money — they just made less than the S&P, and what they did make was ordinary factor exposure you can buy cheaply.
- This is the honest answer to the question every 13F site dodges: can you actually trade this? For these four strategies, no.
It also lines up with our two companion studies: institutional skill doesn't persist out-of-sample, and the famous "stock pickers" show no significant factor-adjusted alpha in their visible books. Three independent angles, one conclusion.
The honest caveats
- 13F is a partial view — long U.S. equity only; no shorts, options, leverage, international, or private holdings. A real edge can live outside what we can see.
- Quarterly snapshots miss intra-quarter trades; we assume the filed book was held.
- Our betas run a little low (~0.5–0.6) for these equal-weight baskets — partly the quarterly factors not perfectly matching a 45-day-lagged window, partly the stale-price beta attenuation typical of small-cap baskets. This biases the measured alpha upward, so the nulls are conservative — a cleaner adjustment would push them lower, not higher.
- We didn't test everything. Activist/13D targets, true-hedge-fund-only consensus, and insider-confluence signals need data we don't yet ingest — those are open questions, not claimed nulls.
- This is research and education, not investment advice. Past performance does not predict future results — which is, once again, the entire finding.
Companion studies: skill doesn't persist · the famous pickers are closet indexers. Full method, every strategy, and the exact numbers live in our backtest registry. NOT investment advice.