Part VIII · The whole market · Chapter 8I
Gamma squeezes
How to recognize one
Chapter 8H described four episodes after they ignited, and this chapter is its operational counterpart. It shows how to spot a gamma squeeze setup before the feedback runs, and which live indicators separate a real squeeze from a noisy uptrend. It also shows when not to chase one despite the headline pattern, and what the evidence says about each step.
- The three preconditions, the four episodes behind them, and how unevenly the record supports each (Chapter 8H).
- Gamma exposure, the sign chain from customer to dealer to hedge, and why the sign is an assumption (Chapters 8A and 8B).
- Short interest and short covering (Chapter 8H).
- A moving average: the average of the last values of a series, recomputed each day as the window moves forward.
1 · The three preconditions, as a checklist
In the Dealer Mechanics Lab's reading of the case studies of Chapter 8H, three preconditions consistently co-occur: Volmageddon in 2018, March 2020, GME in 2021 and August 2024 all had them. Any one alone is insufficient. Chapter 8H showed that the record supports them unevenly, and that the flow doing the documented damage was often not a dealer's. The checklist keeps them as a list of flows to look for, whoever carries them.
| Precondition | What it means | How to measure |
|---|---|---|
| Structural concentration | Either the dealer book is heavily short gamma in one direction, or a meaningful share of the float is shorted (single names) | Net dealer GEX sign and magnitude; FINRA short-interest reports for single names |
| Mechanical buying pressure | A flow source that requires purchases at progressively higher prices: short covering, dealer delta hedging, leveraged-ETF rebalancing | Volume composition, dealer-side flow inference, and the size of leveraged ETFs in the name |
| Liquidity constraint | Counterparty inventory or capital cannot expand to absorb the flow: wide bid-ask spreads, broker margin calls, exchange circuit logic | Bid-ask spread widening; broker outage signals; price paths that gap up rather than expand continuously |
When all three are present, squeeze conditions exist. Whether the squeeze triggers is a question of catalyst: news, positioning around options expiry, or retail coordination. FINRA, the Financial Industry Regulatory Authority, publishes the short-interest figures the first row uses.
The first row: who holds the options
The first row measures dealer gamma, and in a single stock its sign is exactly what public data do not show (Chapter 8B). The historical evidence points the other way from the squeeze story. In CBOE data for 1990 to 2001, investors other than market makers held more short than long open interest in both calls and puts. Market makers were therefore net long both (Lakonishok and co-authors 2004). The illiquidity premia in equity options fit market makers who hold large net long positions (Christoffersen and co-authors 2018). Kanne, Korn and Uhrig-Homburg (2023) find that the market makers' side differs from one underlying to another.
So in a single stock, the default is dealers long gamma, and the squeeze needs the exception: customers buying calls in size from dealers who sell them. Vendors do not even agree on the default. Barchart counts calls positive and puts negative for every ticker, while SpotGamma assumes dealers short both calls and puts on equities. Read the row's "short gamma" as a claim that needs evidence of who bought, not a convention.
The evidence that such buying can move a stock comes from a natural experiment. In January 2018 Robinhood made option trades free, where discount brokers had charged $11 to $28 a trade. Retail option trading then grew more in some stocks than in others, and Lipson, Tomio and Zhang (2023), in a working paper, find that it raised those stocks' volatility. Their channel is the one this chapter assumes: retail trades in higher-gamma contracts, which dealers take the other side of and hedge in a way that reinforces moves. The same paper finds the stocks' quoted spreads 1.8% smaller in the tercile with the most growth, so liquidity improved even as volatility rose.
A cautionary case sits in the same place. In GME, the SEC's staff found market makers buying calls, not writing them (Chapter 8H). A chain full of call open interest was not, by that evidence, a dealer book short calls.
The Edge Book's objection to the magnitude
The first row also measures GEX by its magnitude, and there the Edge Book disagrees. It keeps only the sign of GEX, as low-confidence regime context, and calls the magnitude too model-dependent to size on (Chapter 10H). It files dealer gamma effects under microstructure edges, which it considers institutional-only (Chapter 1G). It also rejects trading GEX levels for retail accounts (Chapter 10I). Read the row's GEX entry as a sign, checked against the conventions of Chapter 8B, and not as a number to size from.
The second row: which flows are mechanical
Short covering and dealer hedging are the classic mechanical buyers. Leveraged ETFs are the third, because their daily rebalance always trades with the move (Chapter 8H). How much that rebalance moves a price is disputed. In US equity funds, investors' flows offset most of it (Ivanov and Lenkey 2018). Korea in 2026 shows the other case, with leveraged products large against single stocks. There, a preprint finds that the rebalance, and the traders who front-ran it, raised SK Hynix's annualized volatility from 100.0% to 136.7% over nine weeks (Zhao 2026). The products' mostly retail holders lost 17.6% of their initial investment. The size of the products against the stock's own trading is what the row's third measure asks about.
Short-dated options can add fuel of their own. Ellahie, Kaplan, Potter and Saha (2025), in a working paper, study the staggered listing of weekly options on single stocks. When weeklies were listed, the stocks' return volatility and trading volume rose. The authors propose that the higher gamma of short-dated contracts raised market makers' hedging needs. They do not identify whether the dealers were long or short that gamma.
2 · Four live indicators: P, V, G and D
The four univariate indicators of the SqueezeMetrics framework, as the Dealer Mechanics Lab presents them, map onto the preconditions as live, computable signals. The letters are the framework's own names: this G is a ratio of call gamma to total gamma, between 0 and 1, not the destabilizing gamma of Chapter 8D.
| Indicator | Computation | Squeeze interpretation | Maps to |
|---|---|---|---|
| P (price trend) | over 21 days, range −1 to +1 | P above +0.4 is a strong directional bias upward, so a mechanical-buying story is plausible | Mechanical buying pressure |
| V (volatility trend) | mad − MA(mad): recent absolute moves against their own moving average | V above 0 while P is strongly positive means volatility is rising into the trend, the rare pattern that separates a squeeze from a quiet uptrend | Liquidity constraint: expanding volatility means absorption is breaking down |
| G (gamma ratio) | call gamma ÷ (call gamma + put gamma), at a constant 20% volatility, weighted by open interest | G above 0.6 is a call-heavy book; with spot rising, dealers absorb more and more upside delta, and short-gamma feedback is live | Structural concentration |
| D (dark ratio) | FINRA dark-pool short volume ÷ total dark-pool volume, as a 5-day moving average | D rising while P also rises is hidden accumulation under a visible uptrend, so the squeeze is fueled and obscured | Hidden order flow and mechanical buying |
P, the price trend, divides the average daily return over 21 days by the average size of a return, , over the same days:
An average of returns can never be larger in size than the average of their sizes, so always lies between −1 and +1. It is +1 when every day in the window rose and −1 when every day fell. It is near 0 when the ups and downs cancel.
Step 1: a steady climb. Suppose that in the window 15 days rise 1% and 6 days fall 0.5%. The average return is , and the average size is . By (8I.1), , above the 0.4 line.
Step 2: a noisy drift. Now suppose 11 days rise 1% and 10 days fall 1%, so the average return is , the average size is 1%, and . The window ended higher, but nothing in it looks like mechanical buying.
V, the volatility trend, takes mad, the average absolute daily move over a recent window, and subtracts its own moving average. If recent moves average 1.5% a day against a moving average of 1.0%, V is +0.5 points: volatility is expanding. A quiet uptrend has P high and V near zero or below. A squeeze has both high, because the price is rising and the market is struggling to absorb the flow.
G, the gamma ratio, computes each strike's gamma at one constant volatility, 20%, so that the skew does not distort the comparison, and weights it by open interest. It then divides the calls' total by the total of calls and puts. A book with 3.0 units of call gamma and 1.5 of put gamma gives , a call-heavy book past the 0.6 line.
G has a blind spot worth naming. It counts gamma by option type, not by who holds it. A call-heavy book leaves dealers short gamma only if customers bought those calls from dealers. If customers sold them, as covered-call writers do, the same G describes dealers long gamma. The indicator can tell you that calls dominate the chain. Only evidence about the side of the trades can tell you whose calls they are.
D, the dark ratio, is the share of dark pool volume reported as short sales, averaged over five days. The usual reading rests on how off-exchange trades are reported. A market maker who fills a customer's purchase by selling short reports short volume, so a rising short share can mean rising buying underneath the visible price. That reading is SqueezeMetrics' own, from its 2018 white paper "Short is Long", which introduced the Dark Index (DIX), a version of the same ratio for the S&P 500. It is a vendor's argument, not a tested finding. A widely read practitioner test of DIX and GEX, a blog post of 2021 with positive results, was retracted by its author after a data error. The corrected advice was to normalize the measure by volatility.
All four are normalized to roughly −1 to +1 with the hyperbolic tangent, applied over a rolling window, so that they can be plotted on one axis:
Near zero, tanh passes a value through almost unchanged; far from zero, it flattens toward ±1. By (8I.2), , and , so large readings are squeezed together while small ones keep their differences.
The interesting joint signals are rare, because on most days the four diverge. When they align, with P, V, G and D all high, the structural setup for a squeeze is in place. That is the moment to check the catalyst, not to bet ahead of it.
An alignment of all four indicators is a positive screen, not a diagnosis. When a condition is rare, even a specific screening test produces positives that are mostly false. That is why a positive screen leads to a confirmatory workup rather than to treatment. Here the workup has two parts. One is the catalyst check: news, expiry positioning or retail coordination. The other is a look at who holds the options, for customers who bought calls from dealers. Until those appear, the alignment has raised the probability of a squeeze, not established it.
Step 1: the base rate. Suppose that in 1,000 stock-months, 5 contain a real squeeze setup, a prevalence of 0.5%. These numbers are made up; no published study gives them.
Step 2: the test. Suppose the alignment of all four indicators catches 4 of the 5, a sensitivity of 80%. Suppose it also lights on 5% of the other 995 stock-months, about 50 of them, which is a specificity of 95%.
Step 3: the positives. About 54 stock-months light up, and 4 of them are real. The positive predictive value is : roughly 13 alignments for every true setup.
How rare the setup really is has only begun to be counted. A working paper by Baig, Zaynutdinova and Strong (2025) identifies 669 potential gamma-squeeze events in US stocks from 2019 to 2023, the first large sample of its kind. Its results beyond that count could not be checked for this book, so the denominator the screen needs is still missing.
The figure below lays out the worked example as a screening table, one square for each made-up stock-month. Sliders set the prevalence, from 0.1% to 50% and starting at 0.5%, the sensitivity, starting at 80%, and the specificity, starting at 95%. A 2 × 2 table under the grid counts the four cells, and the readout gives the positive predictive value.
| Screen | Real setup | No setup | Total |
|---|---|---|---|
| Alignment (positive) | 4 | 50 | 54 |
| No alignment (negative) | 1 | 945 | 946 |
| Total | 5 | 995 | 1,000 |
Mostly false alarms. An alignment raises the chance of a real setup from 0.5% to 7.4%. About 13 alignments light up for every real setup. Each one calls for the workup, not a bet: a catalyst check and a look at who holds the options.
At the start the table reads 4 true positives, 50 false positives, 1 missed setup and 945 true negatives. The predictive value is 7.4%: about 13 alignments light up for every real setup. Raise the sensitivity to 100% and it moves only to 9.1%. Return it to 80% and raise the specificity to 99%, and the value reaches 28.7%. At the start's settings, a prevalence of 5% gives 45.7%. With a rare condition the false positives come from the many stock-months without it, so specificity and prevalence decide what a positive means, and sensitivity barely does.
3 · Single-name squeezes against index squeezes
The mechanism is identical in the two kinds of squeeze, in the Dealer Mechanics Lab's account, but the structural conditions differ in kind.
| Aspect | Index gamma squeeze (Volmageddon, March 2020, August 2024) | Single-name gamma squeeze (GME) |
|---|---|---|
| Required short-interest concentration | None: the mechanism runs through dealer hedging alone | Critical: typically above 20% of float, often above 50% |
| Required catalyst | A volatility shock, idiosyncratic or macro | Coordinated retail call buying, plus a narrative |
| Speed | Hours to days | Days to weeks |
| Asymmetry | Symmetric: positive or negative GEX can blow out | Strongly asymmetric: the long side, with short-side squeezes rare |
| Liquidity constraint works through | Market-maker capital and market depth | Brokerage capital and DTCC margin |
| Resolution | Volatility fades once dealers' capacity to rebalance is restored | Retail attention fades, short covering completes, and broker restrictions break the loop |
Most published "gamma squeeze" commentary conflates the two. The trades are different, the risk profiles are different, and the resolution paths are different. Tag every squeeze read explicitly as single-name or index before you size it.
Two rows of the table need a gloss from the evidence. The index column's examples are the Lab's. Chapter 8H found their documented mechanisms in product rebalancing, forced deleveraging and, in 2024, a quote artifact, with no measured role for dealers' option hedging. And the asymmetry row's "positive or negative GEX can blow out" cannot mean that long dealer gamma runs away. In every feedback model with price impact, only short gamma has a singular point; long gamma damps, and the damping is bounded (Chapter 8D). One working paper, by Sornette, Ulmann and Wehrli, models the directional effect and finds it "strongly asymmetric". The effect is far larger when the option market maker is short the option than when it is long. What can run either way is the direction of the move: short gamma amplifies a fall as readily as a rise. A book that a convention labels positive can also be short gamma in fact (Chapter 8B).
Theory is starting to give the single-name threshold a formula. A 2025 preprint by Dai derives a stability condition for the onset of a gamma squeeze. It reads as the point where the feedback loop's gain reaches one, the of Chapter 8D in single-stock form. It also predicts that low-beta stocks feel disproportionately strong feedback. It is new, single-authored and untested, so treat it as a sign of where the theory is going, not as a tool.
4 · When to chase, when to fade, when to step aside
Three rules follow from the case studies of Chapter 8H:
- Do not chase what is already running. Once GME hit $200, the squeeze was visible. In the Lab's reading, the dealer hedging that drove it from $20 to $200 had largely completed. The path from $200 to $483 was retail fear of missing out (FOMO) and short covering, not gamma, and its risk-reward was much worse.
- Liquidity constraints break loops abruptly. Broker restrictions, exchange halts and clearinghouse margin all act as discrete shock absorbers that can collapse the feedback in minutes. Size any squeeze trade on the explicit assumption that 50% of the intraday peak is the realistic exit price under broker-restriction scenarios.
- Cover before the catalyst clears. Squeezes resolve when the structural condition resolves: short interest covered, dealer hedge inventory normalized, or retail attention faded. Whichever resolves first defines the exit. Wait for the one you can actually observe, such as broker restrictions, dealer-side rebalancing footprints in the tape or short-interest reports. Do not wait for the one you would prefer, price exhaustion, which you cannot predict ahead of time.
The first rule's example is disputed. The SEC's staff, whose data showed market makers buying calls in GME, found the rally inconsistent with a gamma squeeze (Chapter 8H). The rule's logic does not depend on the example. Whatever mechanical buying drives a squeeze, it has had its turn by the time the move is on every screen. The buyers who come last are paying for it.
A model of 1987 explains why the last buyers get hurt. If the market mistakes mechanical buying for informed buying, the price rises above what fundamentals justify. It falls once the buying's mechanical nature is revealed (Jacklin, Kleidon and Pfleiderer 1992). Their case was portfolio insurers' purchases in a rally; call-driven dealer buying in a squeeze has the same shape.
The one case in this chapter also bears on the second rule. On January 28, 2021, GME closed at $193.60, about 40% of its $483 intraday peak (Chapter 8H), below even the rule's half. One episode is not a calibration, but it says the rule's 50% is not a conservative assumption.
Step 1: the scenario. Suppose you buy a squeezing stock at $120 and it runs to an intraday peak of $200. The second rule says to plan for an exit at half the peak, $100.
Step 2: the loss to size for. From $120, an exit at $100 loses $20 a share, about 17%, even though the stock traded far above your entry.
Step 3: the size. If you can accept a loss of $2,000 on the trade, you can hold shares. A stop 10% below entry, $12 a share, would have allowed 166 shares, and in the restriction scenario those shares would lose about .
Keep the checklist as a small table and update it as data arrive. Tag each precondition MET when it is measured and present, WATCH when it is building, NO when it is absent, and NEEDS DATA when you cannot measure it. P and V need only daily prices. G needs the option chain with its open interest, and D needs FINRA's dark-pool short volume. The first row also needs a view of who bought the calls, which open interest cannot give. Without one, tag that row NEEDS DATA, whatever G says. A NEEDS DATA tag is not a NO. It means the checklist is incomplete, and an incomplete checklist does not support a trade.
The checklist is built from four episodes, read after the fact, and the joint signal is rare. The record behind it is thinner than it looks. The showcase single-name squeeze is disputed, and the index episodes ran mostly through flows other than dealers' option hedging (Chapter 8H). Loops also break abruptly, often in minutes, when a broker restricts buying or an exchange halts trading. The price can then gap through any stop. That is why rule 2 sizes for an exit at half the peak rather than at a stop price.
- All three preconditions met, no catalyst yet: check the catalyst; do not bet ahead of it.
- A call-heavy chain: do not tag the first row met until you have evidence that customers bought the calls from dealers.
- Squeeze already visible and running: do not chase it; the mechanical part of the move is largely done.
- In a squeeze trade: size for an exit at half the intraday peak, or lower, and exit when the first observable structural condition resolves.
- Before any of this: tag the setup as single-name or index, and read any GEX input as a sign, not a size (Chapter 10H).
The headline pattern, a stock up several times in days, looks like the setup. By then the mechanical buying that drove it has largely completed. What remains is attention and short covering, which can stop in minutes. The GME path from $200 to $483 was that phase, in the Lab's reading, with much worse risk-reward than the move before it.
A high G says that calls dominate the chain, not that dealers sold them. In single stocks the historical evidence has market makers net long options, and in GME the SEC's staff found them buying calls. Treat a high G as a question about who bought, and look for the answer before you tag the first row.
Commentary often uses one word for both kinds of squeeze. An index squeeze, as the Lab describes it, needs no short interest and resolves as dealers' capacity to rebalance returns. A single-name squeeze needs heavy short interest and retail call buying, runs on the long side and ends with attention, covering or a broker's restriction. Tag the setup before you size it.
- A squeeze needs structural concentration, mechanical buying and a liquidity constraint together, plus a catalyst to trigger it. Any one alone is insufficient.
- In single stocks, dealers have historically been net long options, so "dealers short gamma" needs evidence that customers bought the calls; a call-heavy chain is not that evidence.
- P, V, G and D turn the preconditions into live signals. Their alignment is rare and is a positive screen: it calls for a catalyst check, not a bet.
- Single-name and index squeezes share a mechanism but differ in short interest, catalyst, speed, asymmetry and resolution; only short gamma can run away.
- Do not chase what is running, size for an exit at half the intraday peak or lower, and exit on the first structural condition you can observe.
5 · Check yourself
Question 1
Over 21 days a stock rises 0.8% on 14 days and falls 0.4% on 7. What is P, and does it pass the 0.4 line?
Answer
The average return is and the average size is . So , which passes the line.
Question 2
A stock has short interest of 60% of its float and a gamma ratio of 0.7. Its spreads are normal and its price rises steadily without gaps. Do squeeze conditions exist?
Answer
Not yet. Structural concentration is met by the short interest alone, although the gamma ratio says only that calls dominate the chain, not that dealers sold them. Mechanical buying may be building, but there is no sign of a liquidity constraint. Tag that row NO, or WATCH if spreads start to widen, and wait.
Question 3
You bought shares at $80 in a squeezing stock, and the intraday peak is $150. What exit price does the second rule tell you to plan for, and what does it cost you?
Answer
Half the peak, $75, a loss of $5 a share against your entry, although the stock traded almost twice as high. Size the position so that this loss fits your budget, and remember that GME closed at about 40% of its peak on the day of the restrictions.
Question 4
All four indicators are high on the same day. What do you do next, and why?
Answer
Check the catalyst: news, positioning around options expiry or retail coordination, and look for evidence of who bought the calls. The alignment is a positive screen that raises the probability of a squeeze, not a reason to bet ahead of one. With a rare condition, most positive screens are false.
Question 5
Why tag a squeeze as single-name or index before sizing it?
Answer
The two differ in what they require, how fast they run, which way they can break and how they end. An exit plan that suits one can be wrong for the other. A single-name squeeze can run for days to weeks, while an index squeeze can run its course in hours.
Question 6
A commentator says a positive-GEX index can "blow out" as easily as a negative one. What is right and what is wrong in that?
Answer
A move can run either way: short gamma amplifies a fall as readily as a rise. But long dealer gamma damps moves, and in the feedback models only short gamma has a runaway point. A "positive" reading that blows out is more likely a book whose sign the convention got wrong (Chapter 8B).