The illiquidity premium did not vanish from U.S. equities. It was cancelled out by a second exposure that rides along with it. Greg Leonard's August 2026 working paper argues that a standard long-illiquid, short-liquid portfolio is also, without intending to be, long a concentrated basket of stocks that are expensive to short. Those stocks have earned poor risk-adjusted returns, and they sit almost entirely on one side of the trade. In the post-2013 sample, 28% of the illiquid leg was short-sale constrained, against fewer than 1% of the liquid leg. A factor built that way is a net position, and the net has looked like nothing.
That much is the headline. The tables carry a sharper and less comfortable message. The robust result is the size of the contamination, not the size of the premium left once it is removed. For anyone who allocates on the strength of a factor's level rather than its decomposition, the distinction decides whether the paper is a reason to buy illiquidity or a reason to stop trusting the way it has been measured.
The Number That Holds Up, and the One That Does Not
Over the 105 months from July 2013 through March 2022, the equal-weighted all-stock illiquidity-minus-liquidity portfolio returned -0.32% a month raw. Adjusted for a Fama-French five-factor model with the size factor removed (size and the Amihud measure are close proxies for each other), the alpha was 0.17% a month with a t-statistic of 0.87. That is statistically indistinguishable from zero, which is where the "premium is dead" literature landed.
Strip out the top quintile of stocks by loan fee, which Leonard labels "special," and the alpha rises to 0.49% a month (t = 2.42). The increase itself, reported at 0.33% a month, carries a t-statistic of 3.53. (The rounded endpoints differ by 0.32; the paper reports the unrounded difference.)
The ex-microcap sample is where the disclosure matters. Excluding stocks below the 10th percentile of NYSE market cap, the alpha moves from -0.18% to 0.10% a month after removing special stocks. The 0.28% change is significant. The 0.10% level, at t = 0.46, is not. Once the smallest names are gone, what survives is evidence of a mechanism, not evidence of a harvestable premium.
That should reframe how the paper gets cited. "Illiquidity alpha rises to 0.49% a month" will travel. "Among stocks an institution can realistically hold, the cleaned premium is statistically zero, but the contamination effect is real" is the more accurate sentence.
Why Hard-to-Borrow Stocks End Up in the Illiquid Bucket
The co-location is not an accident of sorting. The rank correlation between Amihud illiquidity and loan fees is 0.47 across all stocks and 0.27 even after microcaps are removed. Special stocks are smaller, more volatile, more thinly held by institutions, more heavily shorted, and more price-sensitive to order flow. In the ex-microcap sample the median annual loan fee was 1.05% for special stocks against 0.56% for general collateral, and the 90th percentile of special-stock fees reached 7.97%.
Leonard's contribution is to argue that the arrow runs, at least partly, from borrow cost to illiquidity. Two channels do the work.
The first is inventory. A market maker that absorbs buying pressure ends up short. If borrowing the stock is expensive, carrying that short until inventory unwinds costs more, so the compensation demanded for supplying liquidity to buyers rises. The model predicts an asymmetry, and the reversal data show it. In the most illiquid ex-microcap quintile, short-term reversal earned 0.42% a month, of which 0.38% came from the short leg. Remove the high-fee stocks and total reversal falls to 0.09%, with the short leg at 0.06%. Among special stocks specifically, the short-leg reversal return is 0.51% a month gross and 0.20% after the realized borrow fee. Most of the apparent liquidity-provision profit is the borrow bill.
The second channel is adverse selection. When shorting is expensive, a trader who still chooses to short must hold a more negative signal to justify the fee, so sell orders in hard-to-borrow names carry more information. Liquidity providers who end up short against informed buyers also lose more. With stock fixed effects and controls for size, short interest, analyst dispersion, and institutional and passive ownership, a one-point rise in the annualized loan fee is associated with a 0.197 rise in immediate price impact and a 0.155 rise in permanent price impact. A ten-point fee increase would add roughly 1.55 to permanent impact against a sample mean of 1.85. That is not a marginal effect.
This is also where the paper parts company with older theory. Classic models predicted that short-sale restrictions could improve liquidity by removing informed pessimists from the order book. Leonard's point is that the logic fits a ban on some traders, not an ordinary borrow fee that every short, market makers included, must eventually pay. Expensive borrowing taxes the people who provide liquidity, and spreads widen accordingly.
The Tension Inside the Finding
Put the two halves of the paper together and an awkward implication emerges. If borrow fees genuinely cause illiquidity, then special stocks are not mislabeled. They are truly illiquid, by the very price-impact measure the factor relies on. Yet they earn negative alpha: -0.80% a month in the most illiquid ex-microcap quintile, against 0.13% for general-collateral stocks in the same bucket.
So the market does not appear to pay for illiquidity uniformly. It pays for illiquidity that arises from thin trading and small size, and it appears to charge, rather than pay, for illiquidity that arises from short-sale constraints. The more precise reading is that illiquidity is priced conditionally on its source. A single Amihud sort cannot see the source, which is why the measured factor has drifted toward zero.
That reframing matters for how the post-2013 weakness should be interpreted. Loan-fee data are only comprehensive from 2013, so the paper uses reversal asymmetry as an indirect diagnostic for earlier decades. It suggests constrained stocks contaminated illiquid portfolios historically, but less heavily than in the recent sample. Leonard points to Regulation SHO and its amendments, which closed routes for maintaining short positions without borrowing, as a plausible reason the borrow-cost link to liquidity tightened. If that is right, part of the factor's apparent decay is a regulatory artifact rather than a sign that investors stopped demanding compensation for trading costs. That link is inferred from a model-based proxy, not tested directly, and should be held loosely.
The Strongest Case Against the Read
The cleanest objection is that removing special stocks would lift almost any long-biased portfolio of small, volatile names. Hard-to-borrow stocks underperforming is an established anomaly in its own right. Dropping a known negative-alpha segment from the long leg of IML will mechanically raise IML's alpha, whether or not anything called an illiquidity premium is being "revealed." On this view, the paper repackages the short-constraint anomaly rather than rescuing a liquidity premium.
Leonard's mechanism evidence is the best answer available: if borrow costs cause the illiquidity, the two exposures are entangled at the root and cannot be dismissed as a sorting coincidence. But that answer explains why the contamination exists. It does not, on its own, establish that what remains after cleaning is compensation for liquidity risk rather than a residual small-stock effect. The ex-microcap result, where the cleaned level is 0.10% and insignificant, leans toward the skeptic.
Three further weaknesses deserve plain statement:
- The loan-fee window is 105 months. A t-statistic of 2.42 over under nine years is suggestive, not settled, and the paper has not been peer reviewed.
- The portfolios are equal-weighted and rebalanced monthly. That design isolates expected-return differences but tilts toward the smallest, costliest-to-trade names, precisely where capacity is thinnest.
- Borrow data are delayed, incomplete, or simply unavailable to many investors, and borrow conditions can change or vanish quickly through recalls. A screen that works on a researcher's clean fee series may not be reproducible in a live book.
None of this undermines the decomposition. It does undermine any jump from the 0.49% headline to a strategy.
What Investors Should Take From It
The practical consequence is narrower and more useful than "illiquidity works again."
For factor researchers and anyone evaluating a small-cap or illiquidity tilt, a raw Amihud sort should now be treated as a bundle of at least two exposures with opposite signs. Reporting the factor without conditioning on borrow cost is reporting a net number whose components could each be moving.
For long-short managers, the reversal results are a direct warning. A gross backtest of short-term reversal in hard-to-borrow, illiquid names overstates implementable returns by most of the short-leg profit. The gap between 0.51% and 0.20% a month among special stocks is the difference between a strategy and an accounting error.
For long-only investors, borrow data still carry information even if they never short. A high loan fee flags a segment where liquidity is likely to be worse than volume alone suggests and where expected returns have been poor. Liquidity screens that rely only on volume or spreads will miss it. The appendix result using bid-ask spreads, where removing high-fee stocks again strengthens the all-stock premium, suggests the effect is not an artifact of the Amihud measure specifically.
The view changes on one observable. If a longer or out-of-sample test shows the cleaned ex-microcap premium becoming statistically meaningful, rather than staying near the 0.10% this paper reports, the case for a revived, investable illiquidity premium strengthens considerably. If it stays near zero while the contamination effect keeps replicating, the right conclusion is that illiquidity was never a single priced factor, and the debate over whether it is "gone" was asking the wrong question.
