Part VIII · The whole market · Chapter 8H

When the framework breaks

Stress episodes

Everything built so far assumes a functioning market: finite spreads, available depth, and dealers who can rebalance in size without moving the book against themselves. Two episodes show what the system looks like when those assumptions break: Volmageddon (February 5, 2018) and the Great Liquidity Crisis (March 16, 2020). The Dealer Mechanics Lab reads both as textbook cases of dealer feedback running away from the standard model.

The point of these case studies is not to predict the next one. It is to calibrate your sense of when the linear GEX framework stops being a useful guide. It is also to see which structural conditions had to coexist for the feedback to become explosive.

Each episode comes in two layers. The first is the record: the prices, volumes and dates that exchanges, regulators and published studies document. The second is the reading: which mechanism produced the record. On the record the sources mostly agree, once a few numbers are corrected. On the reading they often do not, and the chapter says who holds which view, and on what evidence.

A flow that must buy after every rise and sell after every fall behaves like a short-gamma hedge, whoever carries it. Keep that in mind as you read. In Volmageddon, the best-documented flow of that kind was not a dealer's option hedge at all.

1 · February 5, 2018: Volmageddon

Before the event, an eight-month low-volatility regime had compressed the VIX to multi-year lows. Inverse-VIX exchange-traded products (ETPs), XIV by Credit Suisse and SVXY by ProShares, had grown large. The Dealer Mechanics Lab puts the two at a combined approximately $3.5 billion in assets under management (AUM), a figure no source checked for this book confirms. The Bank for International Settlements (BIS) gives about $4 billion at the end of 2017 for a broader group, select leveraged and inverse volatility ETPs (Sushko and Turner 2018).

These products were structurally short the front VIX future. Their rebalancing rule required them to buy VIX futures at the close in proportion to the day's VIX move. The BIS adds that the issuers rebalance daily by trading VIX-related derivatives, usually in the last hour of trading.

The rebalancing rule for any leverage

Let a product with assets $A$ target an exposure of $L \times A$, where $L = -1$ for an inverse product and $L = 2$ or 3 for a leveraged one, such as a leveraged ETF. After the underlying returns $x$, the exposure has become $LA(1 + x)$ and the assets $A(1 + Lx)$. The new target is $LA(1 + Lx)$, so the trade the product must make is

$$\begin{aligned} &LA(1 + Lx) - LA(1 + x) \\ &\quad = L(L - 1)\,x\,A \end{aligned}$$ (8H.1)

For $L = -1$ the factor $L(L - 1)$ is 2, which gives the worked example's $2xA$, and it is also 2 for $L = 2$ and 6 for $L = 3$. The factor $L(L - 1)$ is positive for every inverse or leveraged product, so each of them buys after a rise and sells after a fall. That is the same pro-trend flow as a short-gamma dealer's, which is why Chapter 8I counts leveraged-ETF rebalancing as a source of mechanical buying.

In the unit of gamma exposure, dollars per 1% move, a 1% move makes the product trade $L(L-1) \times A \times 0.01$, always with the move. By the hedge-flow formula of Chapter 8A, that is the hedge of a dealer book whose gamma exposure is $-L(L-1) \times A \times 0.01$ per 1%. An inverse product with $1 billion of assets therefore trades like a dealer book short $20 million of dollar gamma per 1%. Its sign is fixed by the product's design, not assumed from open interest.

Cheng and Madhavan (2009) set out this rule for leveraged and inverse funds and wrote that their daily re-leveraging "exacerbates volatility towards the close". Later studies disagree on how much of the rule's gross trade reaches prices. In US equity funds from 2006 to 2014, investors moved money into and out of the funds against the rebalance. Once those flows and standard risk factors are counted, Ivanov and Lenkey (2018) find the effect on late-day returns and volatility "economically insignificant". Shum and co-authors (2016) find the effect largest on the most volatile days. A survey by Lenkey (2024), an author of the negative result, judges the economic effects insignificant across the literature.

Volatility products were the exception. VIX futures are among the markets most dominated by exchange-traded products. There, Todorov (2023) finds that leverage rebalancing is the part of the products' demand with the largest effect on prices. In his words, it "amplifies price changes and exposes ETF counterparties to variance".

The figure below applies the rule to products from −3× to 3× at once, each with the assets you set. Sliders set the leverage of the product the readouts describe, from −3 to 3 in steps of 0.5 and starting at −1, and its assets, starting at $100 million. They also set the day's move, from −30% to +30% and starting at +10%, and the share of the trade that investors' own flows offset, starting at none. That share is an assumption to play with, not Ivanov and Lenkey's estimate. Faded bars keep the gross trade when part of it is offset.

leverage L

−1×

assets A

$100M

the day's move x

+10%

offset by investors

0%

L(L − 1) = 2

trade at the close = +$20M

net of investors' flows = +$20M

like dealer GEX of −$2M per 1%

With the move. After a day up 10%, a −1× product's assets go from $100M to $90M and its exposure from −$100M to −$110M. Its target is now −$90M, so it must buy $20M at the close: 2 × 10% × $100M.

It opens on the worked example. $L(L - 1) = 2$, the inverse product's assets fall from $100 million to $90 million while its short grows from $100 million to $110 million, and so it buys $20 million at the close. That is the hedge of a dealer book with GEX of −$2 million per 1%. Set the assets to $50 million and the move to +20% for the check question's $20 million. Turn the move negative and every bar flips with it: each product sells into a fall. Only 1× never trades, and between 0× and 1× the dashed curve shows the rule trading against the move.

The Dealer Mechanics Lab replays the session from the open to just after the close, here with the open corrected:

Augustin, Cheng and Van den Bergen (2021) give the VIX as 18.44 at the open and 37.32 at the close. The Lab's replay gives 17.31 at the open and a rise of 115% in one session. That rise is the move from 17.31 to 37.32 (115.6%), which suggests that 17.31 was the previous trading day's close, not the open. Measured from the open, the VIX rose about 102% within the session, which is the rise of over 100% in one day that Augustin and co-authors report. The BIS puts the day's moves at about 4% for the S&P 500 and about 20 points for the VIX.

After the close, Credit Suisse's XIV acceleration clause triggered. An acceleration clause lets the issuer end a product early once its value falls past a stated threshold. The ETP was liquidated. The sources give different sizes for the loss:

• The Dealer Mechanics Lab: XIV's net asset value (NAV), the value of its holdings, down more than 80% intraday.
• The BIS: XIV lost 84% of its value.
• Augustin and co-authors: a one-day loss of more than 90% in short-volatility ETPs.
• The Edge Book: 96% in a single day (Chapter 10I).

The Lab adds that XIV's AUM went from $1.9 billion to effectively zero in one session, another figure this book could not confirm. Every source agrees on the outcome: the product was ended.

What the published record documents

Two studies reconstruct the day. A box in the BIS Quarterly Review of March 2018, by Sushko and Turner, works from VIX futures transaction data, the products' assets and their rebalancing rules. A case study in the Financial Analysts Journal by Augustin and co-authors follows the VIX, its futures and the short-volatility ETPs through the collapse. Two later studies measure the same channel across many days. Together they tell the mechanism through the products:

1. The inverse and leveraged products had to rebalance in VIX futures after the VIX rose, and their buying met a concentrated market. The spike at 16:08 traded "at highly inflated prices", in the BIS's words.
2. Augustin and co-authors conclude that such products can collapse when hedge and leverage rebalancing meets concentrated markets. The collapse came after nearly a year of low volatility. They liken the episode to portfolio insurance in 1987 (Chapter 8D). They also ask regulators and risk managers to watch hedge and leverage rebalancing when concentration and volatility are high.
3. Across the history of VIX products, leverage rebalancing moves VIX futures prices more than any other part of the products' demand (Todorov 2023).
4. The rebalancing reaches the index itself. Bangsgaard and Kokholm (2025) find that VIX-ETP rebalancing demand predicts the end-of-day return of S&P 500 futures, which then partly reverses. The channel they describe runs through the market makers who sell VIX futures to the products and can hedge that exposure in SPX options and futures. They find no indication that the impact comes from new information, or from the hedging of market makers in SPX or VIX options.

The Lab's reading

The Lab tells the mechanism in five steps, each feeding the next:

1. Inverse-VIX ETPs (XIV, SVXY) were structurally short the front VIX future. Their rebalance rule required buying VIX futures into the close in proportion to that day's VIX move.
2. Each VIX uptick demanded more buying, which lifted VIX futures, which lifted spot VIX, which triggered more buying: the canonical positive feedback loop.
3. Dealers on the other side of VIX call wings were short gamma in volatility of volatility, with negative vomma. As the VIX broke through 25, their hedges flipped from passive to forced. Vanna kicked in across SPX puts at the same time, forcing index-level selling on top of the VIX move.
4. Credit Suisse's XIV acceleration clause, triggered by NAV down more than 80% intraday, fired after the close, and the ETP was liquidated.
5. The combined flow into the closing auction created the 115,862-contract one-minute spike the BIS documented. It came from inverse-VIX rebalancing, dealer hedging of volatility of volatility, and panicked retail liquidations.

Steps 1, 2 and 4 match the published record. Step 3 is the Lab's interpretation. None of the four studies names dealers short the VIX call wing. The one study that checks for option market makers' hedging in the rebalancing's effect on the index finds no sign of it. The dealers that mattered in that study were the ones who sold VIX futures to the products and passed the exposure on to the index. The Lab elsewhere places that wing in FLEX options, out of public view, which none of the studies mentions either. Step 5 is partly the Lab's too: the BIS ties the spike to the products' rebalancing, and the Lab adds the dealers and the retail sellers. Read the vomma account as a hypothesis about the other side of the products' trades, one the published studies did not test.

Three preconditions had to coexist for the linear framework to fail, in the Lab's account. The record supports them unevenly:

• Dealers were deeply short gamma in the VIX wings, that is, in volatility of volatility. When the VIX broke through threshold levels, the wing hedging was no longer linear. No published study documents this one; it is the Lab's interpretation.
• Mechanical, rule-based rebalancing demand from inverse-VIX ETPs supplied unbounded one-sided flow. This is the best-documented of the three. The flow was one-sided, though not unbounded: by the worked example's arithmetic, a −1× product's purchase tops out at twice its starting assets, when a 100% rise erases them.
• Market depth had degraded structurally: the calm regime had pushed market makers to thin their quotes, and there was no real depth to absorb the closing-auction sell. The BIS's "highly inflated prices" and the concentrated market that Augustin and co-authors describe fit it.

When all three conditions are present, the effective-volatility equation of Chapter 8D, $\sigma_\text{eff}^2 = \sigma^2/(1 - \kappa G)^2$, approaches its singular limit. There $G$ is the destabilizing gamma, positive when published GEX is negative. Short gamma, whoever carries it, pushes $G$ up, and thin depth pushes $\kappa$ up, so both push their product up toward 1. The system is then no longer described by a probabilistic prior; it is governed by a literal forcing function. This is the right tail of the dealer-hedging distribution, and it does not respond to mean-reversion bets at any size.

Treat that as a picture of the episode rather than a measurement of it. No study has estimated $\kappa G$ for February 5, and none has found a feedback gain near 1 in the S&P 500 on any day (Chapter 8G).

2 · March 16, 2020: the liquidity crisis

Before the event, COVID-19 had spread globally over February and early March 2020. The VIX rose from 14 in late January to 53 by Friday, March 13, which the Dealer Mechanics Lab calls the largest two-week VIX increase in history at that point. The term structure had inverted, into backwardation, on March 9. Risk-parity funds began publicly de-grossing, cutting their gross positions, in the same week.

Jackwerth (2020) read the crisis from S&P 500 options, through the risk-neutral probability distributions their prices imply (Chapter 5C). On February 20 the index did not yet reflect the crisis. Only by March 16 was the full impact visible, with a pronounced second peak, a sizeable crash scenario, in the distributions implied by longer-dated options.

On Monday, March 16, 2020, the stress peaked:

• Before the open, ES futures, the S&P 500 futures of Chapter 2A, locked limit-down: the 5% rule triggered, and they could not trade lower.
• At the 9:30 AM open SPX gapped down massively. It closed at 2,386 (Chapter 2E), 11.98% below the previous close.
• The Dealer Mechanics Lab gives an intraday low of 2,280, "down approximately 12% from the prior close". No source confirms that low, and the two numbers do not fit together: a low of 2,280 would be about 16% below the prior close. The 12% was the fall from close to close.
• The VIX closed at 82.69, the highest close in its history. The Lab calls 82.69 the all-time intraday high, but the intraday high that month was 85.47, on March 18, and the highest intraday reading on record, 89.53, came on October 24, 2008.
• Level 1 circuit breakers, the first tier of market-wide trading halts, triggered on several days around it: March 9, 12, 16 and 18.

The mechanism was layered across several feedback loops at once, in the Lab's account:

1. Vol-targeting funds were de-grossing: risk-parity and volatility-targeting strategies, which size their positions to a target portfolio volatility, were forced to sell equities and bonds together as realized volatility exploded. A European Central Bank simulation gives a sense of scale. A stylized leveraged volatility-targeting portfolio, run through March 2020, had to sell assets worth nearly 225% of its capital to meet its target.
2. Portfolio-margin requirements expanded: as Chapter 2E discusses, they expanded violently in the first two weeks of March. Many accounts were forced into liquidation as their stress-tested requirements doubled or tripled (Chapter 1E).
3. Dealer vanna ran in reverse: with the VIX rising from 30 to 80 in a week, the deltas of out-of-the-money puts became dramatically more negative. Dealers short these puts had to sell increasing amounts of SPX equivalents to stay delta-neutral, and the hedging flow was overwhelmingly one-sided selling.
4. Treasury bonds and gold sold off too, a sign that participants were raising cash through any channel that still had liquidity. The cross-asset correlation collapsed, and nothing was diversifying.
5. The Federal Reserve intervened with several emergency announcements between March 15 and March 23. They eventually included unlimited quantitative easing (QE), which is open-ended central-bank buying of bonds, and expanded repo facilities for short-term lending against collateral. They also included dollar swap lines with foreign central banks. This stopped the cascade, but only after market structure had been visibly damaged.

What the record shows about the dealers

The third loop is the one this part of the book is about, and it is the one nobody measured. The sources behind this chapter contain no peer-reviewed estimate of SPX dealer gamma in March 2020. What the record does show bears on the loop's direction, not its size.

Who bought protection, and when. Jackwerth (2020) finds that "firms bought crash protection ahead of the index crash, whereas retail customers bought it as the index was already recovering". Firms' put buying before the fall fits the Lab's premise that dealers were short the protection institutions bought. Retail's late buying shows demand arriving after the move.

The standing position. The data on who holds what agree that dealers have long been net short index puts. End users were net long SPX options, above all out-of-the-money puts, in 1996 to 2001 (Garleanu, Pedersen and Poteshman 2009). Public investors were net buyers of deep out-of-the-money SPX puts from intermediaries in 1991 to 2012 (Chen, Joslin and Ni 2019). Constantinides and Lian (2021) model the same split. A working paper by Terstegge (2025) rebuilds dealers' positions from Cboe data for 2011 to 2023. It finds a persistent net short of about 19 million put contracts, mostly deep out of the money, with calls near flat. Without hedge adjustments, a 10% fall in the S&P 500 would cost roughly 200% of the dealers' option inventory value.

Deep out-of-the-money puts carry little gamma until a crash brings them toward the money, and then their gamma grows (Chapter 5C). A book that damps moves on normal days can therefore amplify them in a crash, as a worked example in Chapter 8B shows with a made-up book. That is exactly the pattern the Lab's third loop needs.

Capacity shrinks when it is needed. Chen, Joslin and Ni read a fall in the public's net buying of deep out-of-the-money puts as a sign that intermediaries' constraints are tightening. In their data, such a fall goes with more expensive options, higher risk premia across many assets, worse funding liquidity and broker-dealers cutting leverage. In Constantinides and Lian's model, market makers become more credit-constrained in writing puts as disaster risk rises. Dealers' willingness to carry the short-put book shrinks exactly when stress arrives.

The narrative. The story that "dealers flipped to short gamma" in March 2020 was told as it happened. Bloomberg News, on March 23, 2020, reported strategists saying that dealers, "once a buffer to volatility, have morphed into short-term momentum traders, amplifying swings". That was practitioner commentary built on sign conventions and dealer estimates, not a test.

The same three preconditions were present, in a different form, in the Lab's reading:

• Dealers were short gamma in the wings, in the put protection that institutions had bought as COVID-19 escalated. The direction fits the record above; the size was never measured.
• Mechanical de-grossing from vol-targeting funds and forced portfolio-margin liquidations supplied unbounded one-sided flow. These are rule-based flows whose sign is known, and the ECB's simulation gives a sense of their scale.
• Market depth had degraded acutely, as every intermediary tried to raise cash at the same time.

3 · What both cases share

When the three preconditions coexist, the effective-volatility equation approaches its singular limit. The linear GEX framework is then no longer a useful guide. It still gives a direction, but in a regime where the magnitude of the move it predicts is grossly understated. Position sizes calibrated to normal-regime volatility are catastrophically large in this regime.

The actionable lesson is asymmetric: when the three conditions coexist, your size goes down and your stops widen, regardless of whether the catalyst ever arrives. Most weeks, nothing happens. In the weeks when something does happen, the framework built across Parts II to VIII stops describing the market. The cost of being wrong then scales non-linearly with the gamma you do not see.

4 · Reading the warning signs

Both episodes had observable precursors, days or weeks before the explosion. In the Dealer Mechanics Lab's account, the VVIX, the VIX of the VIX, had been elevated for weeks in late January 2018 while the VIX itself slept. That was a tell that positioning in volatility of volatility was deteriorating beneath the surface.

Before March 16, 2020, the VIX term structure inverted. The Dealer Mechanics Lab dates the inversion in two ways: to late February, seven trading days before March 16, and to March 9. Either way the inversion came before the crash, as did the public de-grossing of risk-parity funds from March 9.

Neither reading required forecasting the catalyst. Both required noticing that the conditions for feedback were quietly being installed. The market dashboard of Chapter 11A tracks both the VVIX and the term structure among its markers.

5 · August 5, 2024: the yen carry unwind

A third index episode is the unwind of the yen carry trade on August 5, 2024. In a yen carry trade, a position borrows in yen, at low interest, to hold higher-yielding assets elsewhere (Chapter 10E). The Bank of Japan had unexpectedly raised rates the week before, and Friday's US jobs report came in weak. Leveraged yen-carry positions around the world were then forced to unwind.

In the Dealer Mechanics Lab's account, the Nikkei fell 12% on Monday, August 5, its largest daily decline in percentage terms since Black Monday in 1987. SPX gapped down approximately 3%. The VIX printed above 65 intraday, which the Lab calls the third-highest reading in its history at that point. For a moment it stood at levels seen only in March 2020 and in the 2008 financial crisis.

The Lab reads the unwind as the same mechanism. Leveraged carry positions, which are essentially short volatility on the yen, required forced selling of yen-funded assets all at once, with no time for an orderly unwinding. In its account, SPX dealer gamma was short going into the event, and cross-asset correlations spiked toward 1. The episode resolved relatively quickly, with the VIX back to 20 within a week. For the Lab, it confirmed that the three-precondition framework continues to describe the right-tail behavior of modern markets.

What the VIX print measured

Two official studies took the day's most-quoted number apart. The BIS Bulletin of October 29, 2024, by Todorov and Vilkov, decomposes the VIX change option by option. Li and Askanazi (2025), in a working paper for the SEC's Division of Economic and Risk Analysis, decompose the VIX calculation itself. They agree on four points:

1. The print came before the open. The VIX rose from 23.9 to over 65 in pre-market hours, while VIX futures stayed below 35 (Li and Askanazi). The BIS calls it the biggest one-day spike in the VIX's history, up 180% to almost 66.
2. The market was thin. Pre-market option volume was about 80 times lower than in regular sessions.
3. Quotes, not trades, moved the index. The VIX is computed from the mid-quotes of out-of-the-money SPX options (Chapter 9A). Bid-ask spreads widened asymmetrically, which lifted those mid-quotes, above all for less liquid out-of-the-money puts, which accounted for more than 85% of the spike. More out-of-the-money puts entered the VIX's strip of strikes than usual, and fewer calls qualified.
4. Hedging flows played no meaningful part in the print. In the BIS's words, "VIX ETFs, market makers' option hedging, and dispersion trades do not seem to have contributed meaningfully to the spike". Once regular trading opened, volume surged and spreads narrowed sharply.

What this does to the dealer-gamma reading

The Lab's statement that SPX dealer gamma was short going into August 5 has no source behind it that this book could find. The BIS looked for market makers' hedging in the VIX spike and did not find it. Calling the 65 print the third-highest reading is the Lab's framing, not the BIS's. The level itself measured quotes in a thin market, not a traded volatility.

Two cautions keep this from overturning the Lab's whole reading. The BIS explains the index print, not the equity sell-off, so the role of dealer gamma in the sell-off itself is untested rather than refuted. And the forced unwinding of carry positions, in the Lab's account, is a mechanical flow of the kind the framework describes, whatever the dealers were doing. The lesson for you is narrower and firmer: a stress episode's most-quoted number can be a measurement artifact.

6 · January 2021: GME, the disputed squeeze

The Volmageddon, March 2020 and August 2024 cases were macro, index-level events. The Dealer Mechanics Lab calls the 2021 episode in GME (GameStop) the canonical single-name gamma squeeze. That is a rally driven by dealers buying the stock to hedge the calls that customers bought. It counts the episode as the most operationally relevant case study for understanding retail-driven dealer feedback in individual stocks. It is also the most disputed of the four, because the regulator that examined the trading data did not find the squeeze.

Before the event, the setup had four parts, as the Lab gives them:

• GME traded at about $20 in early January 2021.
• Short interest was above the whole float. The Lab gives approximately 140% of the float, a figure press reports carried for early January, and the SEC staff's report puts the January peak at about 123%. Either way the stock was heavily shorted through stock loans.
• An active retail community, WallStreetBets, coordinated a call-buying campaign.
• The options chain showed heavy retail call buying at strikes of $20, $30, $40 and $50, with very low premiums of under $0.10.

Short interest is the number of shares sold short and not yet bought back, often quoted as a percentage of the public float. It can exceed 100%, because a borrowed share, once sold, can be lent again by its buyer and sold short again.

The Lab's account

The mechanism ran in six stages:

1. Retail call buying created dealer short-call inventory: Citadel Securities, Susquehanna and other wholesalers absorbed the retail call flow as its natural counterparty (Chapter 2B). The resulting dealer position was short calls with growing notional exposure.
2. Dealers hedged by buying GME shares: each short call requires a long stock hedge equal to delta × open interest × multiplier. As GME rose, the deltas of out-of-the-money calls rose toward 1.0, so dealers had to buy progressively more shares.
3. The feedback loop ran: rising GME moved the calls toward the money, which raised dealer deltas, which meant more share buying and more upward pressure, over and over. It was classic short-gamma feedback in single-name form.
4. Short sellers covered: hedge funds buying back shares at progressively higher prices, which is short covering, added to the buying pressure.
5. The run accelerated to its peak: GME hit $483 intraday on January 28, 2021, about 24 times its early-January level. Implied volatility on GME options exceeded 1,000% on a one-day basis.
6. A brokerage liquidity crisis followed: on January 28, Robinhood restricted purchases of GME. The reason, according to later congressional testimony, was its clearing deposits at the DTCC (Depository Trust & Clearing Corporation), which clears US stock trades. They had ballooned from about $700 million to more than $3 billion because of the volatility, and Robinhood could not post enough capital to keep accepting buy orders at scale.

The House Financial Services Committee heard that testimony, and its staff memorandum gives the numbers behind it. On the morning of January 28, the automated notice from the DTCC's stock clearinghouse, the NSCC, asked Robinhood for about $3.7 billion. That included a charge based on value at risk of nearly $1.3 billion, up from $696 million. It also included a further surcharge of over $2.2 billion, which the memorandum calls an excess capital premium charge. Later that morning the demand was cut to $1.4 billion.

In the Lab's account, the Robinhood restrictions broke the feedback loop. GME closed on January 28 at $193.60, down 44.3% from the previous close and about 60% below the $483 intraday high. The Lab's text says the close was down 44% from the intraday high; the 44% was the fall from the previous close. Over the following week, in the Lab's account, retail buying faded, dealer hedges unwound, and GME normalized to about $50.

What the SEC's staff found

The Securities and Exchange Commission's staff examined the trading with regulatory data: the consolidated audit trail, options data by type of participant, short interest and failures to deliver. Their report of October 14, 2021 reached three conclusions that bear on this chapter.

On the options, the staff find that the "increase in options trading volume was mostly driven by an increase in the buying of put, rather than call, options". They go on:

On the rally, they find that neither a short squeeze nor a gamma squeeze caused it, and that positive sentiment, not buying to cover, sustained it. Short covering did occur, especially from January 22 to 27, when the price rose as short interest fell, but it was a small part of the buying. On the restrictions, they trace them to the clearing-deposit requirements above.

The staff's second sentence goes to the root of the Lab's account. By the sign chain of Chapter 8A, a dealer who buys calls from customers is long gamma, and its hedge sells into a rally rather than buying it. If market makers were net buyers of calls, the first stage of the Lab's six did not happen as described.

The academic reply

An ad hoc committee of academics answered the staff report (Mitts, Battalio, Glosten and Kochuba 2022). Using public trade and options data, they argue that GME may have been subject to both a short squeeze and a gamma squeeze. They find that a nontrivial fraction of volume consisted of short sellers' covering purchases. They also argue that the staff's gamma analysis should be extended to market makers' hedging of puts as well as calls.

That last point has the sign chain behind it too. A dealer who sells a put to a customer is short gamma, just as one who sells a call is. Its hedge also buys as the price rises. The put's delta shrinks toward zero as the price climbs, so the dealer buys back the stock it had sold short against the put. Heavy put buying by customers therefore does not rule a gamma effect out. It moves the question to who sold the puts and how much of their hedge the rally triggered, a size this book could not check.

Two facts bear on how much weight each side carries. The committee worked from public data, less granular than the SEC's regulatory records. And one of its authors, Kochuba, founded SpotGamma, a vendor of GEX data, an interest worth noting. The staff report, for its part, gives the views of the SEC's staff, not of the Commission.

How to read the dispute

The honest statement has three parts. The mechanism is plausible: customers who buy options leave dealers short gamma, and a short-gamma hedge buys as the price rises (Chapter 8A). The SEC's regulatory data, finer than the public data its critics used, did not show the call side of it. Academic critics, with coarser public data, argue that a gamma squeeze may have contributed, through puts as well as calls. Treat GME as a disputed case, not the settled example the Lab makes of it.

The case, in the Lab's account, demonstrates four things. Read each with the dispute in mind:

• Single-name gamma squeezes require specific structural conditions: very high short interest, coordinated retail call buying, and enough dealer absorption to create meaningful delta-hedge inventory. GME had the first two; the third is what the staff's data put in doubt.
• Liquidity constraints can break feedback loops: brokerage capital limits and DTCC margin act as natural shock absorbers, though the same constraints can amplify later stress. The staff's report confirms that the restrictions came from clearing-deposit requirements.
• The mechanism is the same as in the index cases: rising spot raises dealer delta and forces buying. It runs at single-name scale, with higher per-share gamma because of the lower notional base.
• Implied-volatility readings during the squeeze were misleading: with spot moving 20% to 50% a day, Black-Scholes-Merton (BSM) implied volatility breaks down. The model's assumptions of lognormal returns and continuous hedging are violated, so practitioners use realized-volatility measures during such episodes.

Three preconditions were present in GME, in the Lab's reading, mirroring the index cases:

• Dealers were short gamma in the call wings, from retail buying. The SEC's data contradict this as stated; the committee's put channel is the untested alternative.
• Mechanical buying pressure came from retail fear of missing out (FOMO) and from short covering. Both occurred, and the staff found covering a small part of the buying.
• Liquidity was constrained at the brokerage level, through DTCC margin. This one is documented.

Structurally, the Lab reads this as the same three-precondition framework as Volmageddon and March 2020, at single-name scale. The single-name version differs in two ways. Its "stress" is asymmetric, on the long side rather than both sides. And its liquidity constraint works through brokerage capital rather than market depth.

7 · What the record adds

Set side by side, the four episodes teach something the Lab's framing does not quite say:

• In the index episodes, rule-based flows did the documented damage. In 2018 the inverse and leveraged volatility products rebalanced into a concentrated market. In 2020 volatility targets and margin calls forced selling by rule. Dealers' option hedging was not found in the studies of 2018, was not measured in 2020, and played no meaningful part in the headline VIX print of 2024.
• The best-known single-name case is disputed. The regulator's data did not show the call-side squeeze; its critics argue for a put-side one.
• These episodes are the tail. On a typical day, the measured effects of dealer gamma are modest (Chapter 8G), and the tail is the least measured part of the subject.

So keep the three preconditions as a checklist of flows, not as a claim about dealer books. A flow that must trade with the move counts, whoever carries it: an inverse or leveraged product, a volatility target, a margin call or a dealer's hedge. Several of these have a sign you can know, which is more than any GEX model can say of the dealers' (Chapter 8B).

8 · Check yourself