Abstract. Stocks jointly held by hedge-fund-like 13F filers co-move beyond what their characteristics explain, and they fall harder in the worst market months. Both effects are statistically significant. Neither clears the promotion hurdle this site applies before a result is treated as tradable, so crowding is classified as a risk measure rather than a signal.

What was tested

The August 2007 “Quant Quake” showed that positions held in common across leveraged managers unwind together, so that one fund’s deleveraging becomes every co-holder’s drawdown (Khandani and Lo, 2011). The question is whether that channel is visible in advance in disclosed holdings, and whether what is visible is strong enough to trade. Form 13F is the quarterly Securities and Exchange Commission holdings filing. A stock’s crowding is the number of hedge-fund-like filers holding it, read only once the filing deadline has passed. The claims, the fund filter and the number of tests were fixed before any filing was retrieved, so the correction for multiple testing existed before the results did.

Sample and construction. Thresholds fixed in advance.
MeasureValue
Form 13F panel53 quarters, 2013 through mid-2026 (2.9 GB)
Filers screened11,750
Hedge-fund-like filers1,216
Fund filterfewer than 250 positions; name turnover above 25% a quarter
Point-in-time lag45 days (filing deadline)
Co-held pairs, matched on sector and size decile3,312
Excess residual correlation, six standard factors removedapproximately +0.9 correlation points
Crowded minus uncrowded quintile, worst decile of market months since 2014−142 basis points per month
Crowded minus uncrowded quintile, all months−33 basis points per month
Tests declared in the pre-registrationn = 2
Promotion hurdleraw t of about 5, after halving gross return and doubling costs

What the data show

If many funds hold the same two stocks, those stocks should co-move beyond what sector, size and the standard factors explain (Anton and Polk, 2014). They do, against matched controls the crowd does not co-hold, and the effect is real but economically small. The stress result is the larger one, and the crowd pays a modest toll in ordinary months as well. That is what the deleveraging channel of 2007 predicts.

Table 1. Crowding effects and promotion outcome.
ClaimEffectNW tOutcome
Co-held pairs co-move (3,312 pairs)+0.9 corr pts4.09held, not promoted
Crowded underperform in stress (14 months)−142 bps/mo−3.90held, not promoted

Why neither is promoted

Promotion requires a claim to survive a re-audit run with a margin of safety, at a threshold set in the spirit of the multiple-testing corrections of Harvey, Liu, and Zhu (2016). Neither t-statistic reaches it. The stress result also rests on the handful of months the pre-registration anticipated, and a small sample remains a small sample however well it agrees with the prior. The finding stands as pre-registered. Crowding is a risk measure, not a tradable signal: it tells a risk desk where liquidity is narrow, and a trading desk nothing about what to buy.

Limitations. Recorded before and during the study rather than after.
LimitationConsequence
Survivorship-biased panel (current constituents only)crowding is counted among survivors; reported statistics are upper bounds
Pre-registration specified industry matching; the available data support sector matchingdeviation recorded in code before results existed
Pre-registered crowding measure specified scaling by float market cap decile; the implementation used unscaled countsa second deviation, not previously listed here; it tilts the crowded quintile toward mega-caps and is only partially absorbed by the FF5+Mom residualization
13F filings disclose neither short positions nor swaps, and the fund filter (fewer than 250 positions, turnover above 25% a quarter) excludes dealer banks by constructionan exposure of the Archegos type is invisible; equity held on swap by multi-manager platforms, which appears if anywhere in the prime broker’s own filing, sits outside this measure entirely. The crowding measured here is a floor, not a ceiling
The study is retained here because most desks would have shipped these two t-statistics as a live trading product. Significant but not promoted is a distinct category, and maintaining it is what keeps the signals that do pass worth believing.

References

  1. Anton, M., and Polk, C. (2014). Connected Stocks. Journal of Finance, 69(3), 1099–1127.
  2. Harvey, C. R., Liu, Y., and Zhu, H. (2016). …and the Cross-Section of Expected Returns. Review of Financial Studies, 29(1), 5–68.
  3. Khandani, A. E., and Lo, A. W. (2011). What Happened to the Quants in August 2007? Evidence from Factors and Transactions Data. Journal of Financial Markets, 14(1), 1–46.