In January 2021, GameStop became the most famous settlement failure in market history, and almost everything said about it was wrong. Fails-to-deliver s...
In January 2021, GameStop became the most famous settlement failure in market history, and almost everything said about it was wrong. Fails-to-deliver spiked, message boards declared proof of a naked short conspiracy, and the actual mechanism, which was far less cinematic and far more instructive, got buried under the noise. The story of GameStop's fails is the story of how settlement plumbing actually works, and why the same data point can mean nothing on Monday and something real on Thursday.
What a fail-to-deliver actually is
A fail-to-deliver (FTD) is what happens when the seller of a security does not deliver the shares to the buyer by the settlement date. In US equity markets, a trade settles on the trade date plus one business day, a convention called T+1 since May 2024, and T+2 before that. If the seller cannot hand over the shares on time, the trade "fails." The buyer is still owed the stock; the delivery is simply late.
The critical thing to understand is that a fail is a delivery problem, not a trading problem. The trade happened. Someone bought, someone sold, money is owed, shares are owed. The fail is only about the shares arriving behind schedule.
Fails happen for reasons that have nothing to do with malice. A broker's stock-loan desk recalls shares at an awkward moment. An operational error mislabels an account. A market maker sells short to provide liquidity, fully intending to buy the shares back and deliver them, and the borrow arrives a day late. The plumbing is messy by default, and messiness alone produces a steady background hum of fails on nearly every liquid name, every day.
What this means for you: seeing a fail in a stock you own tells you nothing on its own. Treat a single day's number the way you treat a single tick, as noise until proven otherwise.
Why the mechanism produces fails even when everyone is honest
Consider why the system tolerates fails at all. The alternative, a market where every trade must be pre-delivered with certainty, would freeze liquidity. Market makers could not quote two-sided prices in size, short sellers could not act on information quickly, and the cost of that friction would land on every ordinary investor as a wider spread.
So the settlement system is built to absorb late deliveries. The National Securities Clearing Corporation (NSCC) nets down the enormous volume of daily trades so that only the imbalances need to settle. When a fail occurs, there are established procedures, including buy-ins, where the buyer's broker can purchase the shares in the open market and bill the failing party. Fails get resolved; they do not accumulate forever without consequence.
Here is the part intuition gets wrong. A fail-to-deliver is often read as evidence that "phantom shares" are circulating, that more shares exist than the company issued. That is not what a fail is. A fail is a late delivery of a real obligation, not the creation of a fake one. The buyer who is owed shares does not get to vote twice or receive two dividends; the plumbing tracks the obligation and closes it.
The confusion is understandable, because the visible effect, a large FTD number, looks like it should mean extra supply. What actually drives the number is the timing mismatch between when a sale is executed and when the borrow to cover it settles. Focus on the mismatch, not on any imagined phantom float.
What this means for you: before you read scarcity or manipulation into a fail, ask whether the number is consistent with ordinary borrow-market friction in a hard-to-borrow name. Most of the time it is.
When fails stop being noise
The GameStop episode is useful precisely because it contained both the noise and the signal. Early in the run, fails spiked as volume exploded and the borrow market seized up. That was mostly friction, an enormous surge of activity straining the delivery pipes. But the underlying condition, extreme scarcity of borrowable shares, was real, and that is the condition worth learning to read.
Persistent fails in a single name, day after day, week after week, are the tell. When the same security shows up on the fail list repeatedly and the numbers do not decay, the plumbing is not the story. The story is that shares are genuinely hard to obtain, borrow costs are high, and the natural resolution mechanisms are struggling against real scarcity. This is why regulators maintain Regulation SHO, which imposes mandatory close-out requirements once a security's fails cross a threshold for a sustained period. The rule exists because persistent fails, unlike transient ones, carry information.
The distinction is between a fail and a pattern of fails. One fail is a clerk's bad morning. A structural, repeating fail in a stock with a thin float, high short interest, and expensive borrow is the settlement system reporting that demand to borrow has outrun supply. That is a scarcity signal, and it can precede or accompany a short squeeze, because the same tight borrow that produces the fails also produces the pressure.
A useful frame for judging the magnitude is fails relative to trading volume. Suppose a stock trades ten million shares a day and shows fifty thousand fails. That is half of one percent, ordinary friction. Now suppose a thinly traded name turns over two hundred thousand shares a day and shows one hundred thousand persistent fails, session after session. That is fifty percent of daily volume failing repeatedly, and no amount of clerical error explains it. The ratio, not the raw count, is where the meaning lives.
What this means for you: read fails as a ratio to volume, and read them over time, not in a snapshot. Persistence and proportion are the two variables that separate signal from plumbing.
The limits of the data itself
Even when fails are persistent, the data has hard limits you must respect. The published FTD figures are released on a lag, typically twice a month, and by the time you see them the borrow-market condition they describe may have already resolved or worsened. You are reading a photograph of a moment that has passed.
The data also does not tell you why the fail occurred. It reports that a delivery did not happen; it does not distinguish an aggressive naked short from a market maker's legitimate delay from a back-office error. Anyone who claims a specific FTD number proves a specific intent is reading a motive into a record that contains no motive field. The number is real; the interpretation attached to it is frequently invented.
This is where the conspiracy reading fails on its own terms. The narrative treats every fail as deliberate suppression, but the mechanism produces fails constantly and innocently, and it also resolves them constantly through buy-ins and close-outs. A system designed to manipulate would not leave a public, regulated, mandatorily-closed-out trail. The visible drama of a large number distracts from the quieter, more reliable signal: does this fail persist, and is it large relative to how much the stock trades?
What this means for you: use FTD data as one input into a scarcity read, alongside borrow rates, short interest, and float. Never use it as a standalone proof of intent, because the data structurally cannot carry that weight.
The read worth keeping
A fail-to-deliver is a late delivery of a real obligation, and that single definition dissolves most of the mythology around it. The plumbing is messy by default, so isolated fails are noise; borrow markets can genuinely seize, so persistent fails relative to volume are a scarcity signal worth respecting. GameStop contained both, and the lesson is to separate them rather than collapse them into a single story.
The practical discipline is narrow and durable. Watch persistence, watch the ratio to volume, cross-check against borrow cost and short interest, and refuse to read intent into a number that records only timing. Do that, and the fail data becomes a modest but honest input into whether a stock is genuinely hard to hold, rather than a Rorschach test for whatever you already believed.
