Eleven Anomaly Factors Earn Their Returns from the Same Crowded Stocks
A translation of a Chinese forum post analyzing the paper "Crowded spaces and anomalies" by Ludwig B. Chincarini, Renato Lazo-Paz, and Fabio Moneta, *Journal of Banking and Finance*, vol. 182, article 107579, January 2026 (DOI: 10.1016/j.jbankfin.2025.107579).
The thought experiment
Suppose eleven well-established, post-publication-robust stock market anomalies. Now apply one filter: drop the most institutionally crowded segment of each anomaly portfolio and trade only the least crowded half. The result: equal-weighted monthly risk-adjusted alpha falls from 0.390% to 0.009%, and the t-statistic falls from 6.42 to 0.18. Tested one by one, none of the eleven anomalies produces positive alpha after removing crowded stocks.
Measuring crowding: Days-ADV
The main crowding metric is Days-ADV: the dollar market value held by all institutional investors at quarter-end, divided by the stock's average daily dollar trading volume over the prior quarter. It answers: at normal volume, how many trading days would institutions need to exit their entire position? A one-quarter lag accounts for 13F disclosures being up to 45 days late.
The metric captures both position size and liquidity. Metrics based only on the number of institutions or ownership percentage omit liquidity and do not produce significant alpha in the authors' tests. A companion measure, ActRatio (active investor ownership relative to turnover), correlates with Days-ADV at roughly 0.99 and performs nearly identically.
A forty-year crowding curve
- The number of institutional investors grew from about 400 in 1980 to about 6,000 in 2021 (~15x).
- The number of companies they held peaked at 5,756 in the late 1990s, then fell to 2,761 by 2021.
- As of December 2016, the ten largest institutional investors managed about 26.5% of total equity market capitalization.
- Event studies: crowded vs. non-crowded anomaly portfolios during the financial crisis (Nov 2006 – Feb 2010, difference t = 1.97) and COVID (Mar 2019 – Feb 2021, t = 2.03); both significant.
- Panel regressions: log Days-ADV predicts crash risk — NCSKEW coefficient 0.011 (t = 3.29), DUVOL coefficient 0.018 (t = 5.13), with stronger effects on anomaly short legs.
- In the transcribed Table 6, one anomaly is labeled ROA, inconsistent with capital investment (CI) in Table 1 and the text; Table 1's definition is authoritative.
- Baseline sample: 1980Q1–2021Q4. A post-1993 subsample exists; Figure 2's caption mentions a 1995 structural break while the text and tables use 1992Q4 — reproduction should follow the text and tables.
- Sample: US NYSE, AMEX, and Nasdaq common stocks with price above $5, excluding utilities and financials.
- Ludwig B. Chincarini, Renato Lazo-Paz, Fabio Moneta, "Crowded spaces and anomalies", *Journal of Banking and Finance*, vol. 182, article 107579, January 2026, DOI 10.1016/j.jbankfin.2025.107579.
- Figures and t-statistics are drawn from the journal page and IDEAS/RePEc abstract page.
- 17.28% is a linear annualization (1.44% monthly FF3 alpha × 12), not an actual return.
More money chasing fewer investable names: that divergence is the source of crowding.
Crowding itself is an alpha
Each quarter, stocks are sorted into quintiles by Days-ADV; buy Q5 (most crowded), short Q1 (least crowded). Value-weighted monthly FF3 alpha:
| Portfolio | Monthly alpha | Annualized (linear) | t | |---|---|---|---| | Q5 most crowded | +0.54% | +6.48% | 8.87 | | Q1 least crowded | -0.90% | -10.80% | -7.86 | | Q5 − Q1 | +1.44% | +17.28% | 9.67 |
The often-cited 17% figure is 1.44% monthly alpha multiplied by 12 — a linear annualization, not an actual return. Under FF5P, FF5A, and FF5AM specifications, Q5−Q1 monthly alphas are 0.958%, 0.892%, and 0.796% (all t > 6), and significance survives adding liquidity and momentum factors.
The core finding: removing crowded stocks zeroes out anomalies
Each anomaly is double-sorted: first on the anomaly variable, then within each leg by Days-ADV — taking the most crowded 30% of the long leg and the least crowded 30% of the short leg. The eleven anomalies come from Stambaugh et al. (2012): composite equity issuance, net equity issuance, total accruals, net operating assets, gross profitability, asset growth, capital investment, investment-to-assets, momentum, Ohlson's O-score, and failure probability.
| Construction | Equal-weighted FF3 alpha | t | |---|---|---| | Anomaly sort only | 0.390% | 6.42 | | Anomaly + crowding double sort | 1.693% | 11.09 |
Value-weighted alpha rises from 0.268% to 1.024%. But once crowded stocks are excluded:
| Remaining non-crowded anomaly portfolio | Monthly FF3 alpha | t | |---|---|---| | Equal-weighted | 0.009% | 0.18 | | Value-weighted | 0.008% | 0.09 |
In the internet appendix, per-anomaly tests show none of the eleven produces positive alpha after excluding crowded stocks. Eleven factors validated over a decade earn their returns from the same batch of stocks — the most crowded ones.
Post-publication performance
Splitting the sample at each anomaly's original publication year:
| Period | Alpha | t | |---|---|---| | Full sample | 1.693% | 11.09 | | Pre-publication | 1.957% | 9.32 | | Post-publication | 1.609% | 7.67 |
Decay is real (1.957% → 1.609%) but the post-publication alpha remains economically large and significant. Value-weighted results decline from 1.184% to 0.973%, also significant. In Fama-MacBeth regressions, the short leg shows stronger post-publication evidence (Pos-Pub × Short × LADV t = 2.92) than the long leg (t = 1.62).
Alpha's cost lives in the tail
The authors' explanation follows classic limits-to-arbitrage logic: many institutions hold the same anomaly stocks, use similar signals, and sometimes use leverage. A market shock triggers simultaneous selling, draining liquidity and crashing prices — so investors demand higher compensation.
Evidence:
The 17% is not free: it is compensation for crash risk. The return shows up in the mean; the cost shows up in the tail.
Caveats
What to watch next
1. Alternative crowding metrics — only the liquidity-inclusive Days-ADV and ActRatio work; independent replication with a new metric would nail the result down. 2. Application to China A-shares and Hong Kong stocks, where retail participation, price-limit rules, and short-selling constraints change both the crowding measure and the unwinding mechanism. 3. Implications for the factor zoo — if eleven anomalies' returns concentrate in the same stocks, multi-factor portfolios built from them may carry far more risk than models assume.
Final reminder: this paper provides evidence on limits to arbitrage, not a recommendation. All figures are historical statistics on 1980–2021 US data; the ~17% annualized alpha corresponds to fatter tails.
---