Short Interest as Information
Short interest is a census of positions, not a census of opinions — the aggregate mixes bets, hedges, and arbitrage legs. The information lives in the borrow cost and the exit congestion.
Short interest is the rare data point that two opposing camps read with equal confidence and opposite conclusions: the fundamental investor reads a high number as informed bearishness, the squeeze hunter reads the same number as trapped fuel. Both are reading a figure that, on its own, supports neither conclusion — because short interest is not a poll. It is a census of positions, and positions are established for reasons that range from a researched conviction that the equity is overvalued to reasons that involve no opinion about the company whatsoever. Until the number is decomposed, it is not bullish, not bearish, and not information.
The framing we use: short interest is a census of positions, not a census of opinions. At least three distinct populations live inside the aggregate. The first is the directional short — the researched bet that the stock is worth less than the price, the only population that means what investors assume the whole number means. The second is the hedge and arbitrage complex: merger arbitrage shorting an acquirer against a target position, convertible arbitrage shorting equity against a bond, index and pairs hedges, market-neutral books that are short a name only because they are long its comparable. These positions can be enormous and carry no view on the company — the short leg is plumbing, not opinion. The third is market-making inventory and hedging flow, mechanical by construction. The proportions vary by name and by corporate situation: a company with a convertible bond outstanding, or one involved in a merger, can print towering short interest that is almost entirely opinion-free. Reading the raw level without asking which population dominates is the original sin of short-interest analysis, and nearly every downstream error descends from it.
The triad: level, borrow, cover
The decomposition becomes tractable because the short position leaves two additional traces beyond its size, and each trace prices a different property. Our working triad: the level tells you who is there, the borrow cost tells you what conviction costs, and days-to-cover tells you how narrow the exit is.
The borrow cost is the underappreciated leg. To sell a share short you must borrow it, and the borrow has a price set in a genuine market where lenders of stock meet demand from shorts. A stock that is heavily shorted but cheap to borrow is telling you supply of lendable shares is ample and the marginal short faces no scarcity — consistent with hedge-driven positioning or with a negative view so consensual that expressing it is frictionless. A stock whose borrow is expensive and getting more so is telling you something sharper: shorts are paying a running fee, sometimes a punishing one, to hold the view. The borrow rate functions as the price of the negative opinion — the closest thing markets offer to a real-time auction for bearish conviction. Days-to-cover — short interest scaled by trading volume — measures something else entirely: not conviction but congestion, how many sessions of ordinary liquidity the entire short cohort would need to exit through. It is a fire-code number, an occupancy figure for a room with one door, and it belongs to the risk analysis, not the opinion analysis.
Why the directional short tends to be informed
The illustration: when the congestion channel ate the information channel
The meme-stock episode of early 2021 is the modern teaching case, and it teaches the opposite lesson from the one usually drawn. In the most famous name, short interest had reached extraordinary levels relative to the shares available to trade — and the information channel and the congestion channel, which usually coexist quietly, came apart in public. Whatever the merits of the shorts' fundamental thesis, the occupancy figure had become the dominant fact: the exit was so narrow relative to the crowd that a coordinated burst of buying could force covering, and the covering itself became the buying that forced more covering. Prices moved by multiples on no fundamental news, and short positions were unwound at catastrophic cost regardless of whether their underlying analysis was sound.
The generalizable lesson is not that squeezes lurk everywhere — episodes of that magnitude require a rare conjunction of extreme congestion, thin float, and a coordinating mechanism. The lesson is that the same number carries both channels at once. Short interest is simultaneously a measure of informed negative opinion and a measure of mechanical exit risk, and which channel dominates depends on the congestion parameters, not on the thesis. At moderate days-to-cover in a liquid name, the information channel dominates and the shorts' presence is mostly signal. At extreme days-to-cover in an illiquid one, the congestion channel dominates and the position's forced-unwind mechanics can overwhelm any information it contains — the shorts become the tinder for the move against their own view. An investor who reads only one channel will be systematically blindsided by the other.
The strongest case against this framework
The sharpest objection inverts the informed-shorts premise: if constraints on shorting are severe — borrow scarce, fees punishing, recall risk live — then plenty of negative information never gets expressed as a position at all. The names with the worst outlooks may show modest short interest precisely because they are too dangerous or too expensive to short, while high, expensive short interest may mark names where the bear case is already so public that its expression is crowded. On this reading, short interest measures the negative view that was possible, not the negative view that exists — and the constraint, not the opinion, sets the number.
This objection is substantially correct, and the framework absorbs it rather than fighting it, because the borrow market is where the constraint itself is priced. A modest short-interest level combined with an expensive, scarce borrow is exactly the signature the objection describes — large unexpressed negative demand pressing against a small lendable supply — and the triad reads it correctly where the raw level reads it backwards. This is the deeper reason the borrow cost outranks the level in our hierarchy: the level measures expressed positioning, while the borrow price clears expressed and unexpressed demand against supply. The residual truth in the objection is a humility clause worth keeping: short-interest data is also lagged and periodic rather than continuous, so the census is always somewhat stale — a name-specific instance of the general disclosure-lag problem — and sharp recent changes in the borrow market will be visible before they are visible in the printed number. The framework reads the print for structure and the borrow for state.
How to apply this framework
- Decompose before you interpret. Check for convertibles, pending deals, and index or pairs dynamics that generate opinion-free short positions. Short interest in a name with a convertible outstanding is a different object from the same number in a name with none.
- Rank the borrow above the level. Cheap borrow with high short interest reads as hedging or consensus; expensive and tightening borrow reads as paid conviction — and expensive borrow over a modest level can mark negative demand the constraint is suppressing. The borrow market clears what the census cannot count.
- File days-to-cover under risk, not opinion. It is an occupancy figure for a one-door room. Elevated congestion in an illiquid name means price action can be dominated by forced covering mechanics in either direction — a volatility and sizing fact, whatever your directional view.
- Treat conviction shorts in your holdings as a subsidized second opinion. Someone is paying a running fee to disagree with you. Reconstruct the strongest version of their thesis and test the position against it — and treat a rising borrow cost in a name you own as a standing prompt to re-run that work.